TBPN - OpenAI Prepares ChatGPT for Ad Driven Era, The AI Paradigm Shift | Keith Rabois, Alfred Lin, Keith Sakata, M.D., Talia Goldberg, Dave Girouard, Kylan Gibbs, Sam Jones, Zach Pogrob

Starting point is 00:00:00 You're watching today. TBPN. Wednesday, August 13th, 2025. We are live from the TBPN Ultradome, the temple of technology, the fortress of finance, the capital of capital. Ramp.com. Time is money. Save both. Easy to use corporate cards, bill payment and accounting and a whole lot more all in one place.

Starting point is 00:00:21 Timelines and turmoil again, except this time it's not the substack. It's technically called Passport. It's Ben Thompson's version of Substack. because semi-analysis and Strathekery are both I think on the same technology platform the same blogging platform but they have slightly different takes

Starting point is 00:00:39 obviously they agree on a lot but we are going to play the bull and the bear today we have we have some we have some some new hats in the studio wait did you was I going to play the bear so we're talking about Google

Starting point is 00:00:55 because recently Ben Thompson came out with a post talking about how he has he's he's reviewed Google's technology strategy their AI placement and maybe things are good and maybe things are going to are going to go well for the company Bucco Capital bloc says Ben Thompson's Google bull posting is accelerating he says a quote from their from Ben Thompson's Sharp tech with Andrew Sharp says I am becoming a Google fanboy let Google abuse their search monopoly as much as they want to humanity is benefiting leave google alone i will say it is hard to participate in this yeah i don't think this is going to work at all i really i can't

Starting point is 00:01:41 see anything this doesn't work at all this is yeah i don't think dylan i don't think dillan tested these out i don't think they look cool and that's it i can only i can see like one one pixel here let me let me see if i can hit the gong can you hit the gong can you hit the gong Well, we had our fun with that. Did you just miss? Okay. Okay. Well, so, so thank you to restream.

Starting point is 00:02:10 One live stream, 30 destinations, multi-stream and reach your audience wherever they are. This stream is made possible by re-stream. So, basically, let me set the stage and then we'll debate it a whole bunch. So last week, Ben Thompson wrote a Stratory article titled Paradigmuch's and the winner's curse and he weaved through some of the opportunities in front of Google and if you were to play the Google Bowl it looks something like this they have a fantastic AI chip with the TPU this allows them to serve frontier models at low the lowest possible cost they're they're pretty dominant on the Pareto frontier as we've seen yes they have incredible cloud scale with gCP Google cloud

Starting point is 00:02:52 platform and that's accelerating we saw in the recent run of earnings Azure did quite well GCP did quite well, AWS was kind of lagging there. So they're positioned well in terms of like building big data centers, big capax, big AI token factories. They have an amazing lab, deep mind, which we will get into some of the deep mind folks who made the updated version of the Metis list. They are producing top tier models, Gemini, obviously very impressive,

Starting point is 00:03:19 but also V-O-3, completely state-of-the-art. And Genie 3 now, definitely state-of-the-art. And Ben highlights that Google is hard to analyze, because Larry Page and Sergei Bryn famously weren't particularly interested in business or in running a company. They just wanted to do cool things with computers in a college-like environment like they had at Stanford, that the company nearly 30 years later is still doing cool things with computers in a college-like environment. Maybe maddening to analysts like Ben who want clarity and efficiency. It also may be the key to not just surviving, but winning across multiple paradigms.

Starting point is 00:03:55 So Ben has become a Google fanboy by his own account. But on the other side of the argument is Dylan Patel and the crew over at Semi Analysis. In this new post on Semi Analysis, GPT5 set the stage for ad monetization in the super app. They lay out a path to complete to complete chat GPT dominance in the advertising space. It's a great read. We'll go through some of it. But you should go subscribe to both Ben Thompson's Tretecory and Dylan Patel's analysis because they are truly fantastic.

Starting point is 00:04:29 So basically when you go to chat ChaptiPT with a highly monetizable query and they pick the funniest possible example, which is DUI lawyer near me. I love these guys. But apparently that is extremely monetizable because if you need a lawyer, if you get a DUI and you're going to pay that lawyer a lot of money. And so it's not unheard of somebody's on the side of the road. Frantically in ChachyPT, it's, yeah, DUI lawyer near me. And if you're if you're a lawyer that,

Starting point is 00:04:55 that represents clients in DUI cases, they're going to pay you maybe $100,000. You're happy to pay $1,000 for a referral fee to Google or to ChatGPT in the future. And so basically, when you go to Chat ChappetPT with a highly monetizable query like that, like if you ask, you know, what's the capital of Wisconsin?

Starting point is 00:05:14 Like it knows that we can't really make money off of that. So let's not light the GPUs on fire. This query is simply too good. I got to find some way to make money on this. I got to make money off of it. If it's the DUI lawyer, example. So the new model router in chat GPT and GPT5 will be able to understand that they could potentially earn hundreds or even thousands of dollars in referral traffic if they help you find

Starting point is 00:05:37 the best person for the job. This means firing up the biggest model, the most expensive GPUs, to make sure you get the best possible answer. This applies to lots of other domains to. Companies are now building out clones. They're called reinforcement learning environments with verifiable rewards. Basically a clone of DoorDash, clone of Amazon.com, a clone of other UI heavy shopping experiences. Exactly. And then the companies can go RL on top of those environments, those virtual environments. To get really good at buying stuff. Learn how to use the real DoorDash.

Starting point is 00:06:10 Learn how to use the real Amazon.com. Learn how to check the box that says, I'm not a robot. And select the bicycle. Literally, yes. Literally, yes. I mean, that was what the semi-analysis crew's takeaway from GPT-5 was that it was not a bigger pre-training model. That was the Death Star. The Death Star was the idea that GPT-5 would be a bigger model or some sort of foundational change.

Starting point is 00:06:38 No, they blew that up. They blew up the idea that GPT-5 would be a much bigger model. And instead, they focused on they RLed the hell out of it, according to the semi-analysis crew. And so it's highly good at very specific things. is the spikiest intelligence we've had. So basically they will URL on DoorDash, Amazon.com, other websites so you can check out on your,

Starting point is 00:07:01 so that the agent can check out on your behalf. It might be expensive for an AI model to jump through all those hoops to actually order you a new pair of headphones, but it'll be worth it if there's a affiliate commission on the end of the line. This obviously poses a major threat to Google search ad revenue.

Starting point is 00:07:17 Don't you assume that the labs are also just training on the real? applications and the real websites themselves? No, because every time you check out on Amazon, you're spending like 50 bucks at least, right? If you're going through the flow and the flow for buying a TV. It's too expensive to do the volume. Millions of times. Exactly.

Starting point is 00:07:36 So why not just simulate the whole thing? Yeah. I'm sure that they are, they probably have. They do test runs. Of course. Of course. Yeah. When Sam Altman needs to go through the Konigseg Configurator, he's using that as training data.

Starting point is 00:07:50 I remember when Sam, when they launched deep research, the example that Sam gave was he was trying to buy this obscure Acura in Japan. An NSX. An NSX. Obscure to some people, not to me. And he was like, yeah, I just used deep research and I found it. And people, a lot of people didn't pick up on that a ton at the time. But that was a highly monetizable deep research report. For sure, for sure.

Starting point is 00:08:19 And so Ben Thompson and others have noted within weeks of chat GPTs in the initial launch that there was a threat to Google. The question is how fast this shift happens. How much will Google adapt to the new paradigm and what the economics of the consumer tech industry look like in a world where we no longer operate on top of zero marginal costs? The cost to serve the one more Google search was zero, but the cost to serve one more DUI lawyer lookup will be. 50 bucks. And so semi-analysis has a bunch of good charts and graphs that we can kind of look through and then maybe we'll go back to the bull case after. But let's look through the bear case. So fabricated knowledge, Doug, who from semi-analysis, came on the show last week, fantastic, hour-long interview. Every time we grab one of these semi-analysis guys, we're like, yeah, yeah, yeah,

Starting point is 00:09:09 the standard interviews an hour, don't worry about it. Like, send them an hour. And they're like, it's always so good. Most of these DBPN interviews are 10 minutes. Like, why, why do you need an hour? my time is because you're gold we love you so fabricated knowledge says so if you can't tell i wrote the f out of this also i know we're getting a lot of pushback but the affiliate model feels inevitable timeline is this instacart adopts a gentic purchase in january instacart's CEO leaves that's fiji simo to open a i in may sam a tone shift uh router for for control of query so let's pull up the videos of Sam Altman on AI ads and I think it will crystallize a little bit of like what we mean when we mean like monetizing a free LLM, a free AI chat app. It doesn't necessarily mean

Starting point is 00:10:00 stuffing display ads in there just like the answer to Facebook's monetization problem was not banner ads on the in the right bar. It was in feed ads that look if you're watching reels and you see a Reels ad, it looks exactly like a reel. And in fact, the best performing ads on Instagram Reels feel just like user-generated content. They don't look like Super Bowl ads. They're additive. Yeah, they're additive and people often enjoy them. And so that will be, at least this is the semi-analysis argument that I sort of agree with, that

Starting point is 00:10:33 will be the, like what we say about ads in AI, it will be more like commissions for agentic checkout, at least to start. So let's pull up the first video. And the example, right now, there's a lot of people that have websites that monetize with referrals to Amazon. Yep. And they're frustrated because a lot of the, I mean, traffic, just like organic SEO, is way down. Yep. And the general read here is that OpenAI will ultimately start to earn that same type of revenue that the publishers historically did.

Starting point is 00:11:07 Yep. So there was a fireside chat at Harvard Business School with Sam Altman. Let's give it up for Harvard Business School. It's the Harvard of Business School. That's what they've been saying. They've been saying. So he got a question from the audience about ad monetization. We'll hear how Sam Altman responded to it.

Starting point is 00:11:26 Although fair, it could be a barrier for early stage entrepreneurs or startups or even small businesses. Given this context, do you envision open AI exploring alternative monetization strategy that could include free API, perhaps supported by advertising or other methods to faster innovation in the future. I will disclose just as like a personal bias that I hate ads. I think ads were important to give the early Internet of business model, but I think they do somewhat fundamentally misalign a user's incentives with the company providing the service. I'm not totally against them.

Starting point is 00:12:11 I'm not saying open-air. I never consider ads. But I don't like them in general. And I think that ads plus AI is sort of uniquely unsettling to me. You know, when I think of like GPT writing me a response, if I had to go figure out, you know, exactly how much was who paying here to influence what I'm being shown. I don't think I would like that. The knowledge engine should go on.

Starting point is 00:12:36 I think I would like that even less. So there's something I really like about the simplicity of our model, which is we make great AI and you pay us for it, and it's like we're just trying to do the best we can for you. And then given that that has some inherent lack of access and inequality, we commit as a company to use a lot of what basically the rich people pay to give free access to the poor people or the poorer people. You see us do that today with the chat GPT free tier. You'll see us do a lot. more to make the free tier much better over time.

Starting point is 00:13:14 And I'm interested in figuring out how we bring the equivalent concept to the API. But I kind of think of ads as like a last resort for us for a business model. I would do it if it meant that was the only. That's where he says, he says, I kind of think of ads as a last resort as a business model. But recently he dropped a new podcast. This was from, I think, a month ago. and it's from the Open AI podcast. So you have to imagine that the run of show

Starting point is 00:13:45 and the talking points in here are very carefully selected to move the narrative forward and kind of educate the community on where the company is going. And so we'll pull up Sam Altman's interview on AGI, GPD-5, and what's next from the Open AI podcast.

Starting point is 00:14:03 So that brings up the other question from people who are using this or skeptical is that opening eye now has access to this data, and there's the concern one was about training, which open eyes has been very clear about when or when not it's training. You know the options to turn that off. The other thing is like advertising, things like that. What's open eyes approach towards that?

Starting point is 00:14:23 How are you going to handle that responsibility? We haven't done any advertising product yet. I kind of, I mean, I'm not totally against it. Turns not totally against it. Yeah. Areas where I like ads. I think it adds on Instagram. Kind of cool.

Starting point is 00:14:39 Yes, that's on Instagram. Very cool. Let's go. I am like, I think it'd be very hard to, we take a lot of care to get right. I have faith. I think you can do it. People have a very high degree of trust in chat,

Starting point is 00:14:51 which is interesting because like AI hallucinates. It should be the tech that you don't trust that much. My friends will be ready to. So I trust them. People really do. But I think part of that is if you compare us to social media or, you know, web search or something where you can kind of tell that you are being monetized in the company is trying to like deliver you good policies no doubt but also yeah you can kind of like this is the

Starting point is 00:15:20 monetized block this is the non monetized block whatever like you know how much how much do you believe that like you're getting the thing that that company actually thinks is the best content for you versus something that's also trying to like interact with the ads I think there's like there's a psychological thing there. So for example, I think if we started modifying the output, like the stream that comes back from the LLM in exchange for who is paying us more, that would feel really bad. Yeah, this is a great solution. Give you the actual answer you want, but hey, these are the best headphones for you, but if you want me to buy them, I'm going to have to go cook as an agent. I'm doing the work and I'm going to take a cut of that. That's amazing. I'm so down for that.

Starting point is 00:16:07 like the agent's like, I'm getting paid either way. Yeah, exactly. And I could say, okay, which headphones do I want? Do I want the Sony's or do I want the, or do I want the Apple AirPods maxes? And if I decide the Apple ones, it goes checks out. It uses some coupon code. It gets some. Yeah, I mean, comparing this to the other ways that people discover products and services.

Starting point is 00:16:30 Yeah. If somebody searches best luxury hotel in Hawaii, they're going to get ads against that. and then they're going to get organic rankings that aren't necessarily the truth, right? Because the truth is for something like best luxury hotel in Hawaii is very subjective. Yeah. Then they might go and try to get recommendations from an influencer. Yep. Hopefully the influencer is disclosed.

Starting point is 00:16:57 Yes. Whether or not they're being compensated by the advertiser. Yeah. And so if I were to ask you as an influencer, like what design software would you recommend? What would you say? Just honestly. Figma.com. Think, bigger, build faster.

Starting point is 00:17:12 Figma helps design and development teams build great products together. This is a paid advertisement. But yeah, the disclosure is super important. Exactly. And if the influencers saying like, oh yeah, I love this hotel. But that hotel is giving them like, you know, two weeks free a year. Then like that's not that's not a super ethical. Totally.

Starting point is 00:17:32 Yeah, it needs to be disclosed. And then also the beauty of the LLM is that is that, is that like the recommendations are going to be able to be tailored. So best luxury hotel, well, if you're, if you really want a certain type of pillow or you really want, you know, a pool in your unit or you want it to be wheelchair accessible or you want, you know, high ceilings or you want, you know, beachfront access, like there's a million different parameters that could go into that. And the thing, the thing that chat chib needs to.

Starting point is 00:18:02 And then it could just, and then it just saves you the time at the very last step. Well, the thing that Chad Chupit needs to navigate is maintaining that trust. Yep. Right? I trust that Instagram is going to serve me ads. Yep. That I trust that they're going to try to serve me ads for things that I will want to buy. Sometimes they serve me an ad.

Starting point is 00:18:20 I'm like, this looks garbage. I'm not going to buy it. Other times they serve me an ad and I'm like, this looks great. You're actually good, good call. I am interested in this product. And the thing is, is like, if Chat Chachypte has to maintain that trust, because if they recommend you a hotel, and they're like, you're going to love this. I know what you like, you're going to love this hotel and you go there and you spend all this money and you stay there and it's terrible.

Starting point is 00:18:41 It's the same thing. If you go to a friend for a recommendation for a hotel, they recommend you a hotel, you show up there and it's like, this is terrible. Like, why did you recommend this? And if they go, oh, yeah, I recommended it because I was getting like 7% referral fee, you're going to be like, what are you doing? Why are you monetizing me? Right. So I think it's a very, it's an interesting challenge that they have where they're going to be direct. already directing so much economic activity and how do you monetize that in a sustainable ethical way?

Starting point is 00:19:13 Yeah. I mean, I think the router is the answer. Open AI or semi-analysis called like this release, like the router is the release. Like GPT-5 is the router. It's not a new model. It's a router on top of multiple models and that's the value. So the router release can be now understood, can now understand the intent of the user's queries and importantly can decide how to respond. It only takes one additional step to decide whether the query is economically monetizable or not. Today, we will make the case for how chat GPT's monetized free end state could look like an agenic super app for the consumer. This is only possible because of routing.

Starting point is 00:19:56 There's a very interesting chart in here. Where is it? It's about the various costs. So cost per million. tokens output has a really really steep power law curve so O3 Pro the the model that everyone's obsessed with the one the one that people really want to hit as much as possible because they feel like it gives it the most the most rigorous and thoughtful output and I was certainly firing off O3 Pro

Starting point is 00:20:25 queries constantly so more and more free users will be able to interact with 03 Pro occasionally because they will trigger it randomly. They might not have been on the upgrade tier, but they actually get to experience what that's like now without having to first go and pay, which I think is cool. Over 99% of free users have yet to interact with a thinking model like 03. And for the average user, ChatsyPT just got a huge upgrade.

Starting point is 00:20:57 And so there's this weird, like the vibes on X with the power users were kind of like all over the place. Yeah, there's that post from John, Collison. But for most people were just like, this is incredible. Everything just got better. If you weren't in love with the old model and you didn't like the upgrade, but for most people, it was just a big upgrade. So the number of- You got to pull this up, pull this up, guys. The number of free users exposed to thinking models went up seven X in the first day and the number of paying users up. John Collison, what it feels like to select O3 and the legacy models menu. Yeah, it's a good metaphor. It's definitely, it's definitely a good metaphor. It's, it'll

Starting point is 00:21:35 Incredible spec. Engagement. Yeah. This is sort of army green on tan. So on a cost per million tokens basis, O3Pro is $80. $80. GPT5 Mini is $2. And GPT5 Nano is 40 cents.

Starting point is 00:21:52 So a huge, huge gap of 200x, the spread on the cost to actually serve the user. So the, but the router is clearly a feature of the, new service and can likely see improvements or changes over time. It will continuously learn on preference rates and open AI promises it will improve over time. They'll get a lot of feedback from somebody said, hey, you triggered thinking I would have liked a faster answer in this case or hey, you gave me the fast answer. I actually wanted you to go way deeper. This is not satisfactory. So semi-analysis says centralizing the control of the free user experience allows for many more future monetization paths. And this monetization path is one that has been hinted at

Starting point is 00:22:35 subtly for a while. It all starts with opening eyes decision to hire Fiji Simo as CEO of applications in May. Let's look at her background because it's telling. Obviously, we covered this back in the day, but we'll cover it again now. So Fiji was at eBay from 2007 to 2011, but her career, defining career was primarily at Facebook. She was vice president and head of Facebook and she is known for having a superpower to monetize. Let's give it up for monetization superpowers. She was critical in rolling out videos that auto play and proving the Facebook feed and mobile and gaming, and I think we should just keep the claps going. She might be one of the most qualified individuals alive to turn high-intent internet

Starting point is 00:23:12 properties into ad products, and now she's at the fastest-growing internet property of the last decade that is unmonetized. It's an obvious story. This is the next list. Post-Mettis list, we need the power rating of monetization maxis. Yeah, Fiji Simo at the top of the list. Moni Maxis. We can continue to run through this, or we can kick it over to the METIS list, whatever you

Starting point is 00:23:33 want. Let's cover a little bit more. So they're covering the tone shift. We obviously had those videos. They say in recent interviews, Sam's tone has shifted. There's clearly a lot of thought happening about how to best monetize free users more recently. Again, this goes back to the kind of little debate.

Starting point is 00:23:49 We're having a little timeline and turmoil moment with Cuban where, again, you can't expect companies to give products that are expensive to serve away for free forever, right? And it's great that pro tier users can, help offset the costs for free tier users, but there's very few. I mean, Open AI, the funny thing is they're trying to convert to a for-profit. Right now they are non-profit, so it makes sense they're giving this incredible product away to millions of people. The benefit of humanity.

Starting point is 00:24:21 Yeah, just for the benefit of humanity. But eventually, you know, they're running a business. I do think, like, Cuban has a point with, which I steal man, with, with the idea of if it was purely based, If the entire flow of we want people to open the app and convert to commerce immediately, that could result in perverse incentives and like lower quality, just general user experience. It might be a situation where that's kind of like a short-term gain for long-term pain in the sense that people wind up churning if it's really, really bad. But I think that it is possible to have a wall in the organization. organization between like, okay, the truth seeking happens here and the first layer of you ask a question, we're going to give you the best possible answer for what you asked. So luxury hotel with your preferences, we're really going to, and the team is purely focused on that. And then they are separate from the monetization team that says, would you like to check out? Okay, we have a great agent that can go do that, book it. And it's similar to having a, you know, a flight. what do they call a travel agent that actually books it for you and then takes a cut of that.

Starting point is 00:25:38 And that's a very clear value because it's actually. Yeah, travel agents have pretty aligned model with consumers, right? They want to give you a great trip. So you come back and book more travel with them. But they end up taking a rev share in different ways from the hotels and various like vendors that they end up like booking the trip through. Yep. And I believe Google has a similar wall between like what shows up in the knowledge panels. cannot be bought.

Starting point is 00:26:05 So there is no amount of Google ad dollars that you can pay them to change your height if it's auto-completing. Like that pulls from Wikipedia and all these data sources into their knowledge graph. There's nothing you can do to manipulate the Google knowledge graph. You can just buy keyword ads that show up in their box. And you might have to scroll a while because sometimes they put seven ads up there. But as an ad enthusiast, if you do a Google search and the entire screen is filled with ads. really-fills you with joy.

Starting point is 00:26:35 It's hard-warming. Yeah. Search DUI lawyers near me right now. I'm sure you'll see them. Anyway, let me tell you about Vanta, automate compliance, manage risk, fruit trust continuously. Vantas trust management platform takes the manual work out of your security and compliance process and replaces it with continuous automation, whether pursuing your first framework or managing a complex program. Should we run through Ben Thompson's?

Starting point is 00:26:54 We should. So, Semi Analysis actually quotes Ben Thompson here and talks about aggregation theory. Let's talk about agentic purchasing and compare it to search quick. to the search query today because LLMs have a core feature that search does not, and that is scaling marginal cost. That's what we talked about with O3 Pro costing $80 per million tokens versus GPT5 Nano costing 40 cents per million tokens. All of a sudden it has a different economic equation.

Starting point is 00:27:22 This is fundamentally different than the world search grew up it. Let's examine aggregation theory by Ben Thompson because the core feature was that most technology companies had zero marginal cost to an additional user. There were some fixed overheads for running while large the large search engine, but the incremental cost of another query was virtually zero. Agents and LLMs kill this concept for the first time. The more you spend, the better your result is because of chain of thought reasoning tokens and now marginal costs exist in software again. There is somewhat direct relationship between more money, more compute and a better answer. Nowhere is this clearer than in AI in which you can spend variable costs to get variably better.

Starting point is 00:28:03 answer or outcome. And so before the router, there was no way for a query to be distinguished and after the router the first low-value query if you ask why is the sky blue that can be routed to a GPT5 mini model that can answer with zero tool calls and no reasoning. This likely means serving the user is approaching the cost of a search query. The monetizable query on the other hand has a fixed cost. It would show a page ranking websites with potential AI summary at the top. This is a fixed supply response. to what could be a variably hard question,

Starting point is 00:28:35 but now chatypte free because of routing. There's some incredibly, you know, incredible open AI hater out there who's on like the maxed out pro plan and just going into O3 Pro and saying, what is the capital of California? Give me a 60 page PDF with the answer. This is the hitting Grog4 heavy

Starting point is 00:28:58 with just like answering one word, but think for 10 minutes. It's like just burning the GPUs. But yeah, I mean, the router will, you know, increasingly eliminate that. Decide how monetizable is this query? And that's how much compute you get for it. And so GPT5 can decide to allocate $50 to a query, create a plan,

Starting point is 00:29:18 gather information about the DUI incident in this example, research local lawyers, consider who is likely to answer fastest, consider your budget, then contact multiple lawyers on your behalf. All of those are tool calls. All of that is expensive. It could even agentically reach out to lawyers on behalf of the free user knowing that the conversion ratio of this query is even higher this version of chatypt is highly helpful aligns with the user's

Starting point is 00:29:43 query and is a valuable referral to the seller of goods and services so there's a little bit of you know going into how does Google respond to this there's there there so chaty PD is partnered with a lot of different companies and finance they've come they've partnered with Visa and PayPal on the consumer side they've partnered with Mattelbooking.com and Lowe's Enterprise software they've a partnership with Shopify yeah well that's on the consumer internet side Snapchat's Shopify Instacart and Mercari so if you are a Shopify merchant you are probably happy to let people check out with your products

Starting point is 00:30:24 directly in chat GPT you don't really care if they hit your if they hit your website yeah it doesn't really matter as long as they're buying your product your margins probably going to be the same and so So OpenAI is firmly knocking on the door of technology giants, Google, and meta, and even Amazon. Previous scares about AI have been focused on search query volume, not being replaced in the ad tech stack. ChatGPT can compete with dominant platforms for its place in the ecosystem. And to date, this push into purchasing is the most concrete example of OpenAI coming for advertising at large. if they were first to launch an aggressive agentic checkout solution before meta or Google,

Starting point is 00:31:03 this would be seen as huge competitive shots for both companies. A reminder that if we are talking about pure usage, only one company is growing users at a meaningful rate. It's OpenAI. And the visits year over year for OpenAI are up 135%. And there's this other crazy, crazy chart in here that's of the top 10 websites, chat GPT is number five and it's

Starting point is 00:31:30 and every single property is over 15 years old so Instagram is the next youngest like website in the top 10 websites and it's 15 years old then you have Google at number one

Starting point is 00:31:47 28 years old YouTube at 21 years old Facebook is 22 years old hearing that Instagram is 15 years old crazy makes you feel a little bit old. X.com, Twitter, originally 19 years old. Reddit is 20 years old. WhatsApp is 17 years old. Guys, warning. Twitter's going to be able to drink soon. Get ready. Get ready for that. Yeah. I mean, rebranding at age 17 is kind of on brand, you know. It was Twitter and then it had to don't call me Twitter anymore. Dad. I'm an adult. I'm X. I'm X. Just call me X.

Starting point is 00:32:20 Call me X. Right. And so. Yeah, when companies turn 21, they should. to just like really even 21st birthday, get a little wild for 24 hours and then lock in again. Yeah. So let's go over to Strateree. Let's go over to Ben Thompson. But first let me tell you about graphite.dev. Code review for the age of AI.

Starting point is 00:32:40 Graphite helps teams on GitHub ship higher quality software faster and you can get started for free. So paradigm shifts in the winner's curse. This was posted Wednesday, August 6th on Stratoree by Ben Thompson. Ben says it's fun and often accurate to think of tech companies in Paris. Apple and Microsoft define the PC market, Microsoft and Intel won it, Google and meta-dominate digital advertising, Apple and Google One Mobile. That, however, is not the defining pair of the

Starting point is 00:33:03 smartphone era, which ran from the introduction of the iPhone in 2007 to the launch of ChatGBT GBT in 2022. Rather, the two most important companies of the last two decades of tech were Apple and Amazon, specifically AWS. The Apple part is easy. The iPhone market created the smartphone paradigm from its user interface to its distribution channel, and with the was richly rewarded with a bit under half of the unit market share and a bit under all of the total profits. Isn't that a great line? It's such a good line, but it's true. A bit under.

Starting point is 00:33:36 Google did well to control the rest in terms of the Android operating system and profit from it all thanks to Google search. But it was search that remained their North Star. The company's primary error in that era was the few years they let the tail, Android, wave the dog, Google. That's a typo. It should be wagging the dog. Or maybe he's making a wave joke because Google had a product named Wave. But usually the phrase is you let the tail wag the dog. The dog should be in charge of the tail.

Starting point is 00:34:05 And the golden retriever metaphor. It wasn't just wagging. It was waving. Yes. The AWS part is maybe less obvious, but no less critical. And the timing is notable. Amazon created AWS in 2006, just 10 months before the iPhone unveiling. And the paradigm they created was equally critical to the smartphone era.

Starting point is 00:34:21 Yep. I explained the link in 2020's, the end. of the beginning where he says this last point gets it why cloud and mobile, which are often thought of as two distinct paradigm shifts are very much connected. The cloud meant applications and data could be accessed from anywhere. Mobile made the I.O. layer available everywhere. The combination of the two make computing continuous. Instead of deliberate. Before you had to sit down at a computer linked to an on-premise server and you could only use technology in one place at a time. And now you can.

Starting point is 00:34:54 can be always on tapping into the cloud wherever you are. Let's hear it for the cloud and mobile. Fantastic. And so AWS was not the only public. I do remember those days as a kid, summer. I had my email. I'd sit down. I'd said some emails as a as a as a as a as a teenager. And then I'd walk away from my computer. I'd go out into the world. Yeah. Unconnected. You used to have to go and turn on the Xbox to play call of duty now. You can play on your phone. And you can even stream it. There's a ton of stuff. It's changed everything. AWS was not the only public cloud provider, of course. Azure and GCP were both launched in 2008, but by virtue of being first, they both-

Starting point is 00:35:35 AWS defined the paradigm. And also were the first choice of the universe of applications that ran on smartphones and more accurately ran everywhere. So if Apple and ABUS were the definers and thus winners of the smartphone era, then it was Microsoft and Nokia that were the losers. The reasons for their failure were myriad, but there was one common thread, neither could shake off the overhang of having won their previous paradigm. Indeed, both failed in part because they diluted themselves into thinking that their previous domination was an advantage. For Microsoft, that previous paradigm was the PC and the Windows platform, which they thought

Starting point is 00:36:09 they could just port to mobile. And so all of their mobile efforts were basically just like Windows running on a phone, kind of dumbed down, didn't really work very well. It took Microsoft years and new CEO to realize that their mistake. So we will come back to this and debate Google more on this show. But we have Keith Rabeoy joining the stream. Welcome to the show, Keith. Good to hear from you.

Starting point is 00:36:33 How you doing? Great. Pleasure to be back with you. Fantastic. Before we go into all the news that's shaking up the timeline, do you have a take on Google? Right now we were diving into the fact that semi-analysis seems quite bearish on Google and is saying that Open AI will basically steamroll them once they get into agenta commerce and checking people out.

Starting point is 00:36:57 Just properly monetizing the user base and all the economic activity that they're already driving. Yeah. And then Ben Thompson's kind of saying like, hey, it's this college campus. They have a bunch of great AI researchers. Like something will come out of this. And it's just a Lindy company. It's been around for a long time. It's not going anywhere. Well, I think chatyPD is the fastest growing consumer app of all time. And as long as that continues, Google's significantly threatened. I think normal people are substituting what used to be searches and queries into prompts. And that really does threaten Google,

Starting point is 00:37:29 even if they have top-tier AI talent, they've yet to productize it in a way that undermines the trend towards Shout TBT. If Google were to fuse together all the information they have about you, your Gmail, your YouTube, et cetera, and make everything personalized out of the box, that would be interesting. We'll see if they can ship that product internally, substantively, and whether it resonates with users,

Starting point is 00:37:54 but they're losing the AI battle to open AI, period. It also threatens the revenue. I don't think that performance-based advertising, which is the mainstream driver of Google success, will be the way that consumers expect to be monetized in the future. I think all the AI applications are showing that consumers are willing to pay for value. And so I think direct value, direct capture from consumers, subscriptions, et cetera, might be better.

Starting point is 00:38:25 And that would undermine the entire advertising business model that really has propelled Google into the stratosphere, which there was product innovation initially, engineering innovation, data innovation, but ultimately is advertising innovation. And they had a more efficient advertising platform. And that may not be the future of the next 20 years. Yeah. Yeah. In some ways, Google's margin and how much revenue they generate for me as somebody who throughout

Starting point is 00:38:50 the year is going to make a lot of searches. that result in a lot of purchase activity. And if I can substitute that by just paying 200 bucks a month and eliminating a lot of that, you know, Chatsubit can become a probably a trillion-dollar company by just eating into that and being willing to give up that incremental revenue that they might get from advertising. I just had this experience.

Starting point is 00:39:15 I bought a Nintendo Switch 2. I went to Chatchip-T. I fired off agent mode. I had it go around, check every. website for what's in stock. It decided the target was in stock. I clicked on that, had to go into the target app and set up an account and it was like a huge hassle. Could have just done it for me and taken a cut, but Google wasn't involved. So it's kind of a crazy time. Yeah, I mean, you also do some rough math. You can calculate Google's revenue per user per month or per year. And then

Starting point is 00:39:43 what consumers are willing to pay through chat CBT and similar products, I think exceeds that. Yep. No, it makes sense. Anyway, enough on Google. give us the high level on what's going on with Open Door. Well, I'm very excited. A bunch of retail investors led by Eric Jackson have really taken a spotlight and focused it on the potential of the company. I think for a lot of reasons, the company for the last three years is not really capitalized on its disruptive elements, its innovation.

Starting point is 00:40:15 And people have sort of forgotten about it. And the public market's attention is really important. And now there's a very strong. shining spotlight on the potential of the company, which I think should be like Carvana. Carvana is a $40 billion company with, in fact, more competition than Open Door has. But Open Door has not been able to frame the narrative about the innovation as well as Carvada. And I think that's going to have to change. I think subsequently the company needs better leadership, both on a story selling tide and on the innovation side.

Starting point is 00:40:48 I, like, tweeted out, you know, really with about 10 minutes of thought, ways to fix open door this morning. And there's just specific, specificity behind every one of those elements that I could dive in deeply. But the company's not executed on any of those things. If anything, it's taking a step backwards and partnering with these legacy real estate brokers, which makes no sense whatsoever. How much of the, like, sell-off in the stock or just kind of like, you know, the, like, Getting into the doldrums and the trouble that open doors run into has been just the interest rate dynamic in the post-2020-2020 era where we've been in this high-interest rate environment that abstractly has a big impact on home buying. But how direct is that on the open-door story? It's moderately direct.

Starting point is 00:41:39 So in a hot market, maybe five to six million homes transact a year. And Open Door gets paid every time it transacts. So in a very high interest rate environment, that number went down to $4 million. So $4 million versus $6. It's not like people stop buying and selling. It doesn't go to zero. And it doesn't go to infinity. So $4 to $6 million.

Starting point is 00:42:00 The problem for Open Door financially is the cost structure of the company at a G&A level, not a marginal level. GNA level is just way too expensive to make profits at $4 million. And this was a known problem. The note post-law, who I work with, warned the company. current management team, myself included in 2015 that this is going to happen. Real estate has roller coaster rides to it and you need your fixed cost base to be low enough that you don't have to transact to offset your fixed cost or burn. Unfortunately, the base was too high. And when the Fed raised interest rates like five times sequentially really fast, kind of an unprecedented way,

Starting point is 00:42:42 very people who were transacted so there wasn't enough profits to offset the fixed cost of running the business. It was a mistake. It really did cause significant strain. Arguably, you know, the entire management team made that mistake, didn't listen to Vinod, whatever. That's three years ago. The company has made no progress on cutting the GNA cost over the last three years. The good news is it's actually really easy to cut the GNA costs these days. AI allows you to substitute for most of what the people that work had opened or do. So you should be able to bring the very, the fixed cost of running the business down here. And then just decide on every whole. purchase are we going to make money or lose money and just purchase homes in any

Starting point is 00:43:20 interesting environment where you like what are the what are the key activities within the business that you think are are most ripe for replacing with agentic workflows honestly I think real world stuff like there are companies that are actually doing inspections with AI the idea that you need this like we have labor you know inspect you know repairs etc etc that that stuff can be done videos in AI much better than traditional people. And there's companies that are specializing this. So I just think that you don't need all these people. And it's going to continue. The acceleration in AI is more and more obvious and more stark every day. So can you cut it to 100 people?

Starting point is 00:44:03 Probably. Yes, actually, maybe less. And then you all have variable costs. And the only thing you need to do is model the value of a home correctly, which is complicated, but open doors been excellent at it. One thing that people miss is with the exception of one quarter in the history of the company, the company has priced homes correctly and purchased successfully and profitably. But the marginal cost, the fixed cost, sorry, has not been able to be offset when no one's transacting. The company has significant market share in many markets. So the company will just mint money as long as it gets its cost under control. Now, it needs to be a massive company, which is the potential. Like, if you think about it,

Starting point is 00:44:45 Let's take a top-down perspective. Yuri Milner actually made this point to me seven years ago. The largest real estate platform in the West is worth about $18 billion. That's insane. This is the largest asset class, period. The idea that the most innovative company in the entire Western universe would be worth $18 billion in residential real estate makes no sense whatsoever. So somebody is going to build a $50, $150 billion market company. And this management team doesn't think that way.

Starting point is 00:45:16 They think about, you know, moving one basis point here and one basis point there versus innovating to build a $100 billion company. There is no reason this shouldn't be a $50 to $100 billion company. You just need the right leadership. And we're going to fix that. We have a couple of questions from the chat. Are you interested in going back on the board of Open? Are you interested in stepping in the CEO seat, getting a new role? Like, what do you see your involvement going forward?

Starting point is 00:45:41 How do you think about that? The company needs a new CEO. if I can be involved in identifying, assessing, and or closing the proper candidate, I'll be happy to be involved. I do not plan to be an executive. I have a very busy, full-time job. And this company is a full-time dedicated CEO who's intense, who's creative, who's innovative. So if I can help that person, if I can encourage that person, that would be wonderful,

Starting point is 00:46:10 regardless of the structure. Maybe they're listening right now. Keith, if you're one for the job. If you're anything like Will Daybrook X-Rae for Great Brockman, head of the AI, please call me. Come on over. I want to talk about transformation and actually implementing agentic workflows, implementing AI and then cutting costs.

Starting point is 00:46:29 Is there a sort of like CAPEX type cost? I mean, I imagine that there's some folks on the private equity side that are buying up small businesses and then hope to use AI to cut costs, improve margins. And maybe they're doing that internally and defrable. those costs of knowing which tools to pick internally across a portfolio. Then there's McKinsey and the big consulting firms that are going to Fortune 500s and saying, you're going to pay us a ton for a slide deck. But in theory, we're going to get you set up with something that will save you money.

Starting point is 00:47:01 But for a company like Open Door, what does it look like? Because it feels like it's probably too complex to just say, okay, yeah, we're going to like sign up for ChatGPT and that solves our problem. There's probably going to be some implementation of these AI tools. And that might have cost in the short term, even if it drives longer term, better earnings outlook. How do you think about that? Well, I think that's why we need a CEO who's AI native, AI insightful. I think that business transformation requires real skill. I think you are right, though, that the hottest area of venture capital is,

Starting point is 00:47:37 venture capital is chasing after old school businesses and trying to interpret charge businesses with AI. whether they're turbocharging on the top line, which to me is more interesting or improving EBITA with AI substitution of cost. That's pretty cool. But we funded a few companies. There's at least two VC competitors of ours that I know of that have dedicated funds that do nothing else except roll up traditional businesses and try to turbocharge them at AI. So I think, you know, this is very common. At the large company level, it is happening apparently. actually apparently Amazon has had a lot of success with this. They don't talk about it,

Starting point is 00:48:14 speak about it, but apparently it's true. And then I think you're going to see more private equity firms insist upon their portfolio, whether they're large market cap portfolio companies or small, apply AI in a thoughtful creative way. Do you think, I mean, it feels like private companies have like an extreme structural advantage in terms of doing like true AI-driven, transformation because Open AI, sorry, not opening I, Open Door, you know, every, over the last week, every 10 posts on X has been some, you know, different retail investor having strong opinions about who the management should be and what the board's doing and all this stuff. Yeah. Do you think that, do you think that if the right person were to come to the table that like a, that Open Door would do better if it was a private company for the next like few years? Yeah. I don't think. So I think you can transform yourself in the public domain as well. I think you can take advantage of those suggestions. I think, first of all, let's just take a step back.

Starting point is 00:49:18 I think retail investors having a point of view in being excited about an opportunity is a great thing. I think the whole point of markets is to allocate capital. That's why we have markets, right? That's why we have public markets. It's an allocation function. And consumers voting with their fee, especially for consumer brands saying, I want more of this, I want less of that, is actually a proper capital allocation. Like if the company did this, I would spend more money with them.

Starting point is 00:49:43 That should encourage capital allocation. Some people have this negative perception of retail investors. I think it's actually better when retail investors say, I'm going to vote with my feet. I'm going to vote with my dollars. If product X or Y or brand does X, Y or Z or brand represents Z, I'm going to spend money with them. That is a reason to allocate more capital to that company. It's fundamentally sound. How do you think about the storytelling?

Starting point is 00:50:10 around AI as a silver bullet versus a core competency that will be a compounding advantage. I'm thinking of the Amazon example you gave. I completely agree that Amazon's been a beneficiary of AI all over the place, but I don't think it's happened in a single quarter. I think they ramped to a million robots across all of their different facilities. They bought Kiva a decade ago. They were using AI recommendation systems to tell you, hey, you're buying a computer. Do you want to monitor? that's AI, but you know, just a couple decades ago. And so it feels like the right person for the job, they might be able to come in, rip off a Band-Aid,

Starting point is 00:50:46 get things right-sized, do the hard work, but then it's really like you can't take your foot off the gas. I think you need to do both. I think there's bottom-up transformation, which is blocking and tackling, persistence, it's like going to berries. You have to go every day, like every day for like a decade. Of course.

Starting point is 00:51:05 Sometimes multiple times for day. Sometimes three, four, five times a day. I mean that in. And then there's, I think there's top down. I think leadership involves sometimes just putting a stake in the ground and saying, thou shall not. Like, we are just not going to do this anymore. We are absolutely out of that business. And that's why you need a founder-driven CEO, truthfully, transformations with emerging technology require the moral authority of a founder just saying absolutely no. I know this has worked in the past. I know this worked in other companies. But we're just not doing that anymore because the world's going this way and we want to be ahead of the world. How do you think Open Door's relationship with traditional real estate agents should evolve? Well, I think it should be, we should be innovating so that there's no comparison. Like the value proposition Open Door provides a consumer, whether a buyer or a seller, should just be so much better that no one wants a real estate agent. It's not a bad thing. Like real estate agents used to do X or Y, and there's a bundle of services they provide.

Starting point is 00:52:03 But when Open Door provides this, it's just like a no-brainer. And if it's not a no-brainer, the company is not innovating and is not creating enough value, period. How do you think about other levers like zooming out to the macro that could just increase the velocity of real estate transactions or just make homes more affordable in America? I don't know how high this up is, how high this is on the current administration's agenda, but it feels like something that people have been clamoring for for years on both sides. Are there obvious wins that you're optimistic about in the next couple of years to just improve the quality of housing in America broadly? Well, affordability is a top-tier issue, certainly for my conservative friends. Yeah. We need to make housing more affordable for more people as fast as possible.

Starting point is 00:52:54 There are some things that have short timelines and some things that take longer. So building is great. We need more. We need more supply. Supply works. Supply has worked in local environments. you can prove it in city acts and citywide, the cost will come down if you built. But you can't build a house, at least right now, without more robots, more automation.

Starting point is 00:53:13 You can't build on overnight. So supply does take time, but we need to start working on supply and getting rid of all the blockers and excuses for lack of supply, particularly in California. Secondly, we do need interest rates to come down. We do need a new chairman of the Federal Reserve. The best thing ever, you know, for Open Door would be replacing carry a CEO and replacing George Powell as Federal Reserve Chair. Fortunately, I think both are going to happen.

Starting point is 00:53:36 I hope both happened in September, maybe before. Okay. Can the interest rates ever be too low? Probably, you know, like interest rates, it's basically related to its high value of money. Yeah. And so if interest rates are too low, people's willingness to part with money

Starting point is 00:53:54 to get paid back in the future gets reduced. And that is investment. That's really what the definition of an investment. So they could be too low. low, but they're definitely too high right now. I guess I've just been thinking like there are, there are a lot of green lights, green flags in the market, new companies are going out, the stock markets at all time highs. Everything feels really strong in the economy.

Starting point is 00:54:16 Even the CPI is coming back flat and GDP is printing. Everything seems pretty good and it feels like... Except the fast casual restaurant. Yeah, yeah, yeah. They're having trouble. Except the slot market is... But in general, things to be seem to be going very well. in the American economy. And when there's a risk of, okay, we could be overheating, I feel very

Starting point is 00:54:38 reassured by having let's lower interest rates as an ace up our sleeve in case we get over our skis and there is a market correction. The Fed does have some, has some tools in the chest. Our friend Joe's point is like in his view, the argument to lower rates is like the data that's coming out of the labor markets. But again, that's also being debated and rehashed. And I don't think anyone really. Well, let me take a step back though. Please. I think the foundation that growth equals inflation is just wrong.

Starting point is 00:55:12 So from 1950 to 2010, each decade we average 3.6 years with over 4% growth without inflation. It's only the modern world post-2010 with qualitative easing that people equate growth with inflation. The good news about AI productivity gains is it's very easy to see how you can grow fast, call it three, four, five percent consistently without sparking inflation because all of the growth is not propelled by labor cost increases, which is what causes inflation. So I think the modern world over the next 30 years, if it's managed correctly, if the leadership in the political sphere is dialed in, should allow consistent growth, which will eliminate

Starting point is 00:55:58 the debt and make a non-serious problem, like 3%, 3%, 3%, 3%, plus, without inflation. And we need to get people who are sort of educated a century ago out of this mindset that every time you see growth, you need to put on the brakes. That's just not, and that's why the Federal Reserve keeps making that mistake. And so we've got to fix that. But part of it is AI-driven, and technology-driven innovation will allow for great growth, consistent growth without inflation. Is that your current outlook, not necessarily a fast takeoff and we're growing at 10% GDP a year?

Starting point is 00:56:39 Satina Della says, call me when we're growing at 10% a year, but a materially improved economic condition for the United States on a long-term basis? Yeah, Scott Bestin likes to talk about 333. And you want the 3% consistently. You could beat 3. I mean, I think we will beat three, and I think he wants to beat three in the next couple of years. So I subscribe to his perspective on the world. I think he's right.

Starting point is 00:57:05 But it's technology that's the magic wand that allows consistent growth without actual inflation. And that's what we need. That is how raising taxes is a disastrous policy. It's not going to fix any problems. Consistent growth without inflation will. And we need a Federal Reserve chair who understands that. We have a Treasury Secretary, fortunately, who really does understand. Yeah, and we've seen that with those charts of the various goods and services inflation over the past decade.

Starting point is 00:57:34 Education and healthcare goes through the roof where everything that's on the technology adoption curve like TVs and dishwashers, that all has gone down in price. And so the more goods and services that you can put on the deflationary curve, the more growth you get and the better health of the American consumer. Good stuff. Jordy, anything else? I think that's it. Thank you for jumping on on short notice. Thanks for taking the time, Keith. This is always great. Come back on whenever you have more thoughts on if you can think for 10 minutes and get a post up, you can jump on the show.

Starting point is 00:58:03 Please, we'd love to have you. Great pleasure to be with you. I'll be back. We'll talk to you soon. Bye. Take care. See you. Let's go to the Mettus list.

Starting point is 00:58:10 We updated the Mettus list, our ranking of the top 128 now AI researchers. It's burning up the timeline extremely controversial. Fortunately, we have Tyler Cosgrove to blame for that. We had no, no involvement whatsoever. We will be disavowing the METIS list if it comes back to bite us. But Tyler, why don't you give us an overview of what changed, who's on top, and what's going on with the METIS list today? Okay, so we're going to the mic right. Yeah, we're good.

Starting point is 00:58:39 So, okay, before we start, just want to give, you know, this is not done yet. The list can change. Okay, so we got some haters in the comments again. This is final D.N. Tyler, if you're missing. Oh, this person like, oh, he's so low. Okay. We can fix it, all right? It's not done.

Starting point is 00:58:54 OK. All right. But yeah, OK, let's start with the top five here, right? OK. So I think a big change we'll see is that we saw Noam Shazir and Ilya switch. Wow, that's pretty big. That's a huge move.

Starting point is 00:59:07 OK. What drove that? Sorry? What drove that? Is that just because Ilya has been quiet at SSI hasn't published anything in the last year? I think that's a big part of it. And Noon's been on its hair?

Starting point is 00:59:16 Yeah, so I know so I mean, I think Noam Shazir is broadly almost like, you can't say he slept on. He's number one. But he is, I think, you know, punching above his weight a little bit. I think Doug from 70 analysis came on and said that the reason that Gemini is so good is because Noam's back. Yeah, you can basically track like Gemini was okay, it was fine, he comes back from character, they're goaded again.

Starting point is 00:59:39 Okay. Okay. So let's go down to number three. Soundboard, Jordan, D'Ear. Sound board is down. Okay. Demas. Yeah, Demis.

Starting point is 00:59:46 He was missing from the first list. That was maybe a conscious decision. It might not have been. He was missing entirely. He was on the list at all. He's one of the greatest ever. You know, the thought was like, you know, he's not as much of a researcher now. He's more a leader.

Starting point is 00:59:59 He just leaves the lab. Exactly. Yeah, okay. You know, if you trace it back, he's obviously still making some research decisions. Okay, yeah, yeah, yeah. So I think it's fine to put him back on. Okay, so he's number three. He's number three.

Starting point is 01:00:09 Dario. Dario. Amadeh, Dario. Amade, moves up, Anthropic. Over at Anthropic. CEO. Another lab leader. Got it.

Starting point is 01:00:16 Makes sense. And then last up, we have John Shulman. And Dario, has he been on the Dorcasch podcast? He has. Okay. I think actually all five. All five. All five.

Starting point is 01:00:26 He's hit seven of the top ten as well. Seven of the top ten. I mean, he's been on generation. We didn't just listen to Dorcasch. We studied. We sat down and listened. We didn't just hear Dwar Cash. We sat down and listened.

Starting point is 01:00:38 Yeah. Okay. Who's last? John Shulwin. Okay. Thinking machines. Yeah. So another big player.

Starting point is 01:00:46 I believe. Huge. Everyone except Demas has at one point worked at Open AI. Wow. Actually. Wait, I don't know if that's true, maybe. I don't think Noam did, but still. But yeah, that's the new top five.

Starting point is 01:00:58 Okay. Let's move over to big moves. Big moves. Who we got? Who we got? From two weeks ago. So the first one we can look at number 13, Noam Brown. Noam Brown.

Starting point is 01:01:06 So he did really well. A 36 spots. Of 36 spots. Now, he was in the IMO gold medal team at Open AI that cracked that code. Yeah, that's correct. He's big on the kind of RL team. I don't know. They probably have multiple RL teams post-training.

Starting point is 01:01:21 Yep. But yeah, he's doing a lot of good work there. There's someone else at Open AI who's famous for like holding RL all together. They made the list as well. Is that correct? Are you forgetting that? Yeah, there's a bunch of just like post-training RL people that are new on the list now. That'll be important going forward. Yeah, I don't know if we mentioned, but the list is longer now, right? Yeah, it's now a hundred 28.

Starting point is 01:01:39 We're at 100, now we're at 128. A nice base two number. I think big additions to Will Brown, crack the list. Unsurprising to anybody with a brain. Multi-time TBPN appearance, so it makes a lot of sense. Yeah, yeah, Chad. Okay, we can go the next one, Paul Cristiano. Yeah, break him down.

Starting point is 01:01:58 He's kind of an OG. Okay. He, I don't know if he's the godfather of RLHF, but he would use a big name on that paper. Now he's mostly into safety stuff now, but he's definitely kind of a thought leader in the space, right? We got Raghav in the chat saying, Tyler is trying to atone for mogging,

Starting point is 01:02:14 Anthropic in Sonnet 4 yesterday? Is that true? Are you being the allegations? Number four, that's pretty high. I think you've atoned. I think you've atoned. Good job. Sheltow, I think Shultow actually moved down.

Starting point is 01:02:26 No. Anthropica still, you know, they have a lot of people on the list. Okay, okay. Let's go to Peter Abil up 66. What is he known for? I've heard his name before. I mean, so he's just an academic. He's at Berkeley.

Starting point is 01:02:36 Okay. Oh, he's not in the lab. He's not in the lab right now. Not yet. He's leaving billions on the table. He's leaving billions on the table. I think literally, yes. He's been on a ton of papers.

Starting point is 01:02:47 He's like also advised. a ton of the top researchers. Sure. So I think he's kind of influential in that way. Let's go down to Jan Lacoon is actually missing from the list now. Missing. So he was at shots fire. I think he was top 10.

Starting point is 01:03:02 Okay, he was top 10. And he's now not even on the list. He's down on the list. And he's that... So, you know, like, I don't make the list. I just give the voting out. And then people make the decision. It's not me, okay?

Starting point is 01:03:11 So let us know in the chat. Should we shoot the messenger? Or should we not shoot the messenger here? Do we have a toy bone error? I can shoot the messenger here? Yeah, so, Jan McCune, big meta-AI researcher, Rand Faire, was not directly on the Lama project, but was one of their key researchers. Is that right?

Starting point is 01:03:31 And then kind of Lama spun out of Fair, their AI research lab. And then, yeah, and sort of like a thought leader in many ways, steward of the strategy. Also, sort of a hater on deep learning for a while. Sort of a hater on LLMs. I don't know about deep learning specifically. He was kind of he did CNN's right. That was his big thing. So he's kind of he's been kind of maybe right. It's kind of too soon to call it on him, but he's certainly like lost a lot of power within that organization as as Mark Zuckerberg has built out the meta super intelligence lab. I think his title is still chief AI scientist. Okay. But maybe it might be co-led. Yep. Yep. Yep. And then of course you have Nat Friedman, Daniel Gross. But he's not considered to be on the MSL. No. Yes. No. And so that's that you mean no as in yes. He is. He is. not considered. He's not in MSL. He's in a separate org. He's in a separate org. He's in a separate

Starting point is 01:04:23 org right now. Okay. Then break down your yourgen Schmidthuber. What's up with him? Yeah, down 53 spots. He's also somewhat controversial during the, the Nobel Prize. You know, he was saying, oh, it should have been me instead. He's kind of, he's a real OG in the space in deep learning. He was trying to aura farm the Nobel Prize? Yeah. So it didn't really work, but but he was unsuccessful. Yeah. It was a failed aura farm. So so so maybe that That's the reason why he can drop so far. I do remember when Chachapiti first launched, there was someone in the comments,

Starting point is 01:04:55 dystopia breaker was saying, oh, all this stuff Schmidt Huber did all this years ago. There's nothing new here. Of course, I think that under represents the importance of actually productizing these technologies and these research efforts. But still interesting to see that he fell so far. Let's get into the overall stats.

Starting point is 01:05:15 Yeah, so this was actually pretty interesting. So we can look at the, this is the number of researchers per lab. Okay. So we basically see the top three are all, you know, neck and neck, right? Opening eyes at 24, Deep Mine at 23, Anthropic at 22. This is different from the previous list. Anthropic was I think leading by maybe four researchers. Yeah.

Starting point is 01:05:36 But now it's really neck and neck here. And then you see thing machines at 12, somewhat surprising. They're really, I mean, they have a lot of goats on their team. And then meta down at eight. Interesting that meta doesn't have more on the list, given what a spree they've been on. Yeah. But they're probably still just in the early days of actually building out that squad. Yeah, I mean, if you consider they started MSL like what a few months ago?

Starting point is 01:05:58 Yeah, it takes time to even when you have the money. It takes time to actually convince people. I think if we would have made that list back then, they would have had basically zero, right? Totally. They would have had Jan Lacoon. Yep. But he's not even on the list anymore. Yeah, they didn't have that money.

Starting point is 01:06:08 Sure. And then we can also break it down my country. I mean, I remember those viral posts about people saying like, oh yeah, I worked on Lama 3, not Lama 4. And now I'm at another lab. Like that was something somebody posted on their LinkedIn. Viral LinkedIn screenshots. There was a little bit of an exodus. And now Zucks rebuilding the team going into season 26.

Starting point is 01:06:25 Exactly. Yeah. And then finally we can break it down by country. We see USA, of course. USA, 51. USA. Way above everyone else. China at 14.

Starting point is 01:06:33 USA. You love to see it. UK at 13, Canada at 12. Canada would be putting that in the true zone. On a, yeah, on a weighted basis, though, population weighted Canada is doing fantastically. They have one-tenth the population of America, I believe, and one-fifth as many AI researchers on the medicine. So congrats to the Canucks up north. It's always hard especially with China because a lot of their labs are very secretive, right? Yeah, so did anyone from

Starting point is 01:07:00 from High Flyer Deep Seek or Alibaba? We have two, I believe, two deep-seek researchers. Okay. One is from Kimi or like Moonshot. Yep, Moonshot is big now and then. I don't know if we have anyone specifically from Alibaba. the the guy who created the gal who created the TikTok algorithm you got to put them on there that thing is wild yeah it's always just hard because you know brain rot you want to build the brain rot machine I think especially with we should we should actually build the brain rot list the brain rot list the researchers who have created the most sticky you know user generated content algorithms user hours user hour maxers anyway what

Starting point is 01:07:39 else you got for me so I think one of the big improvements of this list versus up two weeks ago was that I think it was originally we kind of optimized a little bit too much for Twitter Cloud or Google Scholar citations Twitter Cloud? It's hard look I mean the labs are so secretive now they don't really put out papers of course and if you have someone incredible you have a huge incentive to not let them do press yeah like hey no actually you can't go on to our cash we did we definitely got messages from people that were saying can you can you put can

Starting point is 01:08:09 you take my team off the list like stop stop talking about us we'd prefer if these people didn't get poached but that's the name of the game in the comments yep yeah I mean there's much people that probably like should be on the list rich Sutton John Carmack yeah both of them at Keene we haven't seen a lot from Keene but would be very very interesting yeah but yeah we're gonna big shout out to Mark Chen broke the top 10 well-deserved up 19 spots sitting at number six just under John Shulman to the chat. I have unplugged and plugged back in my microphone. Hopefully it sounds better.

Starting point is 01:08:48 If not, I can switch to the other microphone, but let me know how it sounds. And we will stop cutting off Tyler Cosgrove, our intern, because the chat is telling us to stop cutting him off. So, Tyler, who else fell off the list? How is my boy George Bull doing? Yeah, George Bull, Alan Turing. Wait, both of them got cut? But they're on the list. Leibniz is on the list now. Okay, okay, that's good. But what happened to George Bull?

Starting point is 01:09:15 Where is he? People were really hitting on him. He's off the list entirely? Yeah, off the list entirely. I don't know what his actual rank is. Obviously, like I have the internal one. I think the ghost of Alan Turing is going to haunt you, Tyler. So Alan Turing is not on the list.

Starting point is 01:09:28 He fell off. He's off the list. I mean, I guess that's kind of fair. His test didn't really hold up too much. He got mogged by Chad GPD. But this test. Tyler Cowen's still defending Alan. Okay, yeah.

Starting point is 01:09:40 The Turing test is Lindy. Any other moves? Did we get any other Easter eggs on this version? Or are we Easter egg free now? I think, I mean, Leibniz, you could maybe say is an Easter egg. Yeah. I think that's- We didn't want to go full meme core? That's the main one, yeah. Okay. But we'll see. Oh well. Well, thank you for all the hard work on the Metis List. You can check it out at metislist.com. Fantastic work, Tyler. We will continue to update it. So if you're angry, if you're happy, shoot Tyler a message. Yep.

Starting point is 01:10:09 Tyler also produced and edited a wonderful vibe video. You did. A launch video for this, which we're very excited to see him dip his toes in the water of video editing. He can do it all. A run for their money. Anyway, let's run through some posts. We have 20 minutes until Alfred Lynn from Sequoia Capital joins us. Let's see what else is going on.

Starting point is 01:10:36 I did want to cover the NVIDIA H20 news. Yes, Chinese authorities have urged local companies to avoid using NVIDIA's H20 artificial intelligence chips, particularly for government-related purposes, the media report said. So they're worried about backdoors and they're probably worried about generating revenue for Uncle Sam. They're sworn enemy. Yeah. So there's an interesting dynamic here because the Chinese economy is not monolese.

Starting point is 01:11:05 economy is not monolithic. It is like state directed capitalism. It's a mixture. It's with Chinese characteristics, of course. So the news is that Donald Trump approved Nvidia's request and AMD's bucket in here as well, but lower in importance to export export H20 GPUs to China. These are the nerved AI chips and they're critical for training large language models. And of course, Deepseek famously optimized their training. training regimen and algorithms so that they could run on H20s. And so they have figured out a way to train large language models on H20s, despite H20s kind of being designed to not let you train or inference large language models as efficiently. But they're still getting it to work and they want them.

Starting point is 01:11:51 So first off, oddly, the news is that Individia will pay a 15% export tax roughly. It's not technically a tax. Just a rev share. A rev share. It's a donation. the federal government, not to Donald Trump personally, but to the federal government for the age 20s that they sell. Now, it is unconstitutional to levy an export tax on American-made goods. This came from the southern states, I think post-Civil War, where the southern states were exporting lots of cotton goods,

Starting point is 01:12:24 and they were worried that the northern states were going to try and raise federal revenues through export taxes. That would disproportionately hit the South. But there's another odd wrinkle where the H20 isn't technically made in America. It's the chip is made in Taiwan with equipment from the Netherlands, the memories from South Korea. I think Singapore might be involved at some point. It never actually hits American shores. All the packaging happens overseas. And then it's shipped to China.

Starting point is 01:12:50 And so. NVIDIA is, of course, an American company. American company. Exactly. So Trump still has leverage. And he was able to block exports of H20s back in April, which we covered on the show. and now Trump argues that this chip is not a threat. There's this incredible quote where Donald Trump says,

Starting point is 01:13:07 they're not getting Blackwell. Blackwell's the best chip ever. Maybe they could, but they'd have to pay more. He's sort of all over the place, but he's having fun. So the age 20, at this point, I think most people, the consensus is that it is an older chip, it's nerfed, and it won't lead to nuclear weapon level AI technology. Maybe that's the next chip,

Starting point is 01:13:27 but certainly we're sort of in the implementation, you know, plateau era of, you know, decent value coming from these AI systems, but certainly not anything super intelligence coming out of a rack of H20s just yet. So Beijing is demanding the tech companies, including Alibaba and BytDance, justify their orders of NVIDIA's H20 artificial intelligence chips, which just further complicates things for Jensen. He's, you know, the meme of him, you know, smoking. smoking a heater. Is that a meme?

Starting point is 01:14:01 No, it's, who's the actor? Batman McConae? No, no, no. There's one of him, like, taking a drag off a cigarette when he's, like, engaged in some conspiracy or, like, unveiling a conspiracy. No, it's the Ben Affleck. Oh, Ben Affleck. Yeah, yeah, yeah, yeah, totally. You can somebody chat GPT, Jensen, you know, sitting outside his office.

Starting point is 01:14:21 Yeah. The tech companies have asked by regulators such as the Ministry of Industry and Information Technology. Let's give it up for that name. Yep. of industry and information. Anytime you've got a ministry, it sounds a little ominous, but they're making people explain why they need to order. Explain, why not use a domestic alternative?

Starting point is 01:14:40 Yep. And so the dynamic here is pretty clear. Beijing wants China to continue broadly, like the government wants China to continue to move down the learning curve for advanced semiconductors. They want smic, Smi, Wawa, to develop the Indeq, indigenous supply chain for semiconductors and do all the hard work and the only way to

Starting point is 01:15:04 actually get the yields up and and really get to the frontier is volume volume so a Chinese data center operator went on record to say it's not banned but has kind of become a politically incorrect thing to do when asked about buying age 20s yeah one issue with doing politically incorrect things in China is you can often be disappeared potentially and have your your wealth taken from you. I think they got to push back. I think they got to push back.

Starting point is 01:15:35 So the dynamic is, you know, China wants to continue developing supply chain, semiconductor, their semiconductor supply chain. And then on the flip side, Chinese companies just want to develop, you know, the best possible AI models that they can. And so they don't want to be GPU poor. And so there's going to be a little bit of a dance there. And some of these will be justified. There still might be some diversion.

Starting point is 01:15:59 that happens just for political reasons. It's all very complicated, but it will be interesting to see how many of those age 20 GPUs that have been kind of mothballed can Nvidia actually sell. And we'll see that in their next earnings report, most likely. And the irony here, everyone, you know, people, like the ultra China Hawks, the, you know, AI war group saying that, you know, criticizing the original,

Starting point is 01:16:25 you know, H20 deal, be saying that it would help the Chinese military, and just broadly undermine U.S. strength in artificial intelligence, and now you have the Chinese government just saying, actually, we don't even want it. We don't want you buying them. We don't want you using them. Yep.

Starting point is 01:16:42 Well, AI war seems to have been averted, and war with China is also, between China and Taiwan, is also at an all-time low on polymarket, seven percent chance by the end of this year. Of course, the year is ticking by, so you'd expect that to go down. but even by the end of 2026, it's only a 22% chance. So people brought...

Starting point is 01:17:02 22% chances. People have been saber-rattling about this for a long time, that something's going to happen soon. And, you know, at least the polymarket doesn't really think that's going to happen. Anyway, let me tell you about Julius. What analysis do you want to run, chat with your data, and get expert level insights in seconds? You can ask Julius to analyze your data. And it's loved by 2 million-plus users. entrusted by individuals at Princeton, the Boston Consulting Group, and Zappner.

Starting point is 01:17:31 When Rahul first told me, when I saw the 2 million user number, I was like, is that a, you add a few extra zeros there? It did seem like, no, it's real. It's real. It's fantastic. So there's another interesting data point that came out of earnings. So there were two companies that announced earnings recently and are in the business and finance section of the Wall Street Journal today. So stable coin firm Circle records loss, but revenue source 53%, and then separately, Corweave posts a loss on higher revenue. So both of these companies, one in AI, one in crypto,

Starting point is 01:18:09 beat on top line, missed on the bottom line. So they're in different industries, but it feels like they're adopting similar financial strategies, which is invest right now for growth, go, go, go, get the top line higher, become a big company. So the numbers are crazy, circles. Circle's share price has quintupled since its June IPO.

Starting point is 01:18:31 We interviewed the CEO just after that IPO. I had not been tracking exactly what the share price had done. That's incredible performance. Revenue's up 53% year over year. It's great stuff. But the losses are growing. Analysts expected a $338 million loss for Circle. They posted a $482 million loss in the second quarter.

Starting point is 01:18:52 Something simpler happened with CoreWeave. We also interviewed the founder of CoreWeave on the show. second quarter revenue tripled since a year earlier. That is incredible performance. But the company lost 290.5 million in the quarter, which is 15% more money lost than the analysts expected. Not a huge miss, but certainly something that people weren't really pricing in fully. And both of those companies have had, I believe they've had profitable quarters, but then

Starting point is 01:19:19 kind of gone down as they've gone back into the reinvestment mode. And so my read is basically like there's green lights all over the economy. It's green flags everywhere. markets open, the IPO windows open, the economic data is really good. So invest, invest, invest, take advantage of the AI race, take advantage of the new crypto regulations, anything you can to go, take as much market share as possible and get really, really big. So, you know, maybe the wave is cresting, but why not? Get a firm footing on your board while you can. Absolutely. Anyway, that's my take on CoreWeave and Circle. Anyway, let's move on. I think we hit Google pretty well. We can

Starting point is 01:19:57 tell you about profound, though. That's obviously relevant to the Google conversation. Get your brand mentioned in chat GPT. It's going to be more important than ever going forward, especially as they add agentic commerce features. You can reach millions of consumers who are using AI to discover new products and brands. I wanted to highlight a brand.

Starting point is 01:20:16 The founder's name is Isabelle. She said, this morning, my brand that's less than 13 months old is launching Whole Foods nationwide without a seven figure raise. How is that even possible? One, strong lending partners. We have invoice factoring and PO financing at less than 12% APR, making it a clear no-brainer to free up working capital.

Starting point is 01:20:34 Two, prioritize strong unit economics, 50% plus and bi-coastal manufacturing and fulfillment to not get crushed on freight. Three, trial and retention. We invest- Wow, manufacturing on both coasts day one. That's a bold move. And then four, the thing that's interesting, A2 dairy is on trend.

Starting point is 01:20:52 So A2 dairy is not quite, it's not raw. raw, which is non-pasteurized dairy. Lightly pasteurized or something. I forget the exact definition of it. But it's, but it's, think of it as a milk derivative. Let me pull up the definition. A2 milk is a type of cow's milk that primarily contains the A2 beta casein protein, unlike regular milk, which contains both A1 and A2 protein.

Starting point is 01:21:22 So some people are sensitive to A1. Sure. And so they can have A2. dairy and we were talking about this earlier today. I haven't been a fan of a lot of like ready to drink products, especially in coffee lately because they include all these random alternative milks that have a lot of sunflower and canola oil in them. But anyways, super impressive. You just want a simple, you just want a simple RTD with a simple RTD with ingredients that anyone can understand 300 milligrams of caffeine, 12 milligrams of nicotine.

Starting point is 01:21:59 A pound of creatine. A pound of creatine. Some Adderall XR. Just stuff anyone can understand. Just a pound. It's all we need. If you're planning your launch of your consumer product, get on linear. Linear is a purpose-built tool for planning and building products.

Starting point is 01:22:13 Meet the system for modern software development. Streamline issues, projects, and product roadmaps. Go to linear. Dot app to get started. Jovian says I have successfully solved the opening-a-a-naming problem. and they say they have a screenshot here harder, better, faster, stronger

Starting point is 01:22:29 I kind of like this. It kind of works. It's crazy but it's just crazy enough to work. I do think it's hilarious that there's like chat GPT5 thinking mode and that implies that like the normal version just doesn't think at all I guess I do like

Starting point is 01:22:46 that it's that I do like the language here where it's harder it's not just thinking or not thinking. It's thinking harder thinking better, faster. And maybe that's where this will collapse because right now, even in the GPT-5 update, I mean, obviously it's such an improvement

Starting point is 01:23:02 over the previous one where you had to know that 03 was better than 4-0, which is deeply, deeply confusing. But now I'm seeing fast thinking, thinks longer for better answers, and pro-research-grade intelligence, auto decides how long to think. That's pretty, pretty good.

Starting point is 01:23:20 I think they're close. I do think over time we will see no, no selection, at least in the main UI, maybe buried somewhere, but overall, it seems like the model. It is really funny, though, that people went complaining about the complexity of the naming to the same people complaining about not being able to select their own models. Yeah, were those the same people? Or was it like separate cohorts, but there's definitely some over-up.

Starting point is 01:23:44 It's the current thing to hate on the- Joe-Shia box says AI psychosis, psychosis is rampant now. I think you might have a version of this. Totally. I mean, yeah, I don't know. I think there's a big difference between seeing something that is chat GPT generated text and being like somewhat frustrated and annoyed that somebody's just slopping up the timeline. Yeah.

Starting point is 01:24:09 With a bunch of it's not this, it's that. I don't want to see that. Yeah, it has been a fascinating story though. But in general, but in general, there's this excitement. People have this like general excitement to try. try to identify who they think has been one-shotted. Yes, yes, yes. But later today, we're having Dr. Keith, Ciccotta on.

Starting point is 01:24:32 He's a doctor at UCSF, and he has seen a number of patients this year that he's identified as suffering from AI psychosis. It's clearly very real. But at the same time, my takeaway is that it should be solvable pretty, pretty quickly with you know, if you can decide in the model router, do I need to think really hard and actually help someone with this checkout? You should also be able to say, okay, this person definitely thinks I'm the boyfriend now. It definitely thinks I'm, I'm, you know, God or something, or they think they're God. And going from there just makes a ton of sense. Anyway, let me tell you about

Starting point is 01:25:15 numeralhq.com. Sales tax on autopilot. Spend less than five minutes per month on sales tax compliance. Benchmark Series A. And we have our next guest, Alfred Lynn, from Sequoia Capital in the studio. Welcome to the stream. How you doing, Alfred? Good to meet you. Doing great. Thank you for having me on the show.

Starting point is 01:25:36 Great to see you. Thank you for coming on. This is overdue. Yeah. Would you mind kicking us off with an introduction on all the different pieces of the Sequoia world that you touch currently? Because I know you're, the firm has grown so much. There's a lot that you could be focusing on, but I'd love to ground it in what is current, how you're spending your day currently.

Starting point is 01:26:01 Well, Sequoia's been around for 50 years and we're still very much focused on, on venture capital. So we have a seed business, a venture business, growth equity business, an expansion business. And then we have an overlay fund called the Secorre Capital Fund. And most of my days are still trying to find the pre-seed, the seed and series A founders that are daring enough to start a company and want to change the world. So that's where I focus. What's happening at the earliest possible stage in the Sequoia portfolio?

Starting point is 01:26:35 Like what, how, I mean, I feel like if you go back 40, 50 years, you hear these stories about, oh, yeah, Sequoia got 10% of this multi-billion dollar company for $100,000. Obviously, the market dynamic have shifted, and there's some huge seed rounds happening, but you are still funding people with really small checks at a certain stage. Can you explain how that all works? Yeah, I like to say that my job is to take small dollars and make them into big dollars. You don't need to overcomplicate it. Yeah, let's not complicate things. You're trying to put small amounts of money to work, and you're trying to make sure that it becomes large amounts when the company becomes successful.

Starting point is 01:27:17 And yes, there are large seed rounds, but there are also just lots that's going on. And it's not just AI. There's just a lot of breath. AI is enabling a lot of things. And AI is not the only thing that we invest in. We are all generalists here from consumer to enterprise in robotics and everything you can think of. We're trained as generalists because things move and things change. It's not just one thing after another.

Starting point is 01:27:46 It's not like we only invested in the internet when the internet was happening and only invested in mobile and only invest in SaaS when those things were happening. And the same is true now. There's a lot going on. And the founders that I really enjoy meeting are two people on an idea. And they feel like the world has gotten something wrong in the world. And they want to go fix it. And their legacy will be changing the world and how we live just like some of the people. people I've been fortunate to partner with. And we use that term very, very judiciously at

Starting point is 01:28:24 Sequoia. We want to partner with the daring founders who want to start the company. We don't think of ourselves as investors. We don't try to buy low and sell high. We really want to work with a small select set of founders that want to go all the way. And, you know, the thing that I've learned over time is you can take us two founders in a seat of an idea. And in a decade that that company could be worth one to ten billion dollars. And in two decades, that could be 10 to 100 billion. And in three decades, that can be 100 billion to a trillion. And we've seen a number of companies that have reached out, including Nvidia, including Apple, including Google. and we're going to try to help the next set of founders do those things.

Starting point is 01:29:15 Going back to something you said earlier, you said you're not just investing in AI, but what does a company look like today? What is a non-AI company that look like today that's coming to you to pitch you at this sort of idea stage? Because in some ways you have to think that if a founder wants to just not think about AI and they're starting a company from the ground up today, it seems like, you know, I'm sure there's some outliers, but it seems like a question that, you know, if you have the blessing and the opportunity to start a company in the year 2025, there's probably some way that AI could be transformative to your business or the market broadly.

Starting point is 01:29:58 Yeah, don't get me wrong. I think all the companies that we work with are using AI and AI tools. That doesn't mean that they're AI native. And you guys were just talking about the AI, the AI researchers and the METIS list. There are some companies that aren't going to be AI native. They're not going to build a foundation model. They're not going to build a world model. They're going to be building an application. And in some sense, in probably two or three or four years,

Starting point is 01:30:24 we're going to call those back to software companies. Those are just new updated ways of building software companies. And we do believe in a lot of value will be created in the application layer. And new applications will be created in this one. And I think there's a desire to associate all those companies as AI companies, just like when the web was happening, everybody wanted to be called a web company. But at the end of the day, you may be a consumer company. You may be a gaming company. You can be a commerce company.

Starting point is 01:31:01 You might be using the internet as distribution. And here you might be using AI as a way to improve the way you get things done. but you're not a native AI company. Yeah, so your definition of an AI native company is effectively, like you have to be hiring researchers. And am I hearing that correctly? Because I think some people describe themselves as AI native because they're just they feel like they're using the tools to the, to the,

Starting point is 01:31:28 it also just feels like the difference between like AI native company where it really matters is just like what will the financial profile look like? When I think of AI Native foundation model company, I think R&D, I think CAPEX, maybe they're not building a big data center, but they're at least spending a lot of money on training versus an application layer company. It's going to be much more about the how much does it cost you to generate those tokens, even if they're on a different foundation model. And then how much value are you delivering and how much revenue are you generating from your customers? Is that a reasonable framework? For me, that's a reasonable framework in the sense that I think there are a lot of companies that are using AI and AI tools to be able to increase productivity.

Starting point is 01:32:19 But they're building an application. They're building a service. They're building something that is different than a native AI company. So how is your thinking, you know, it feels like this debate has faded a little bit. into the background, but over the last five years, how did you think of evolve on kind of value accrual between the model layer or the labs and the sort of application layer? Because in our view, you know, we had a number of folks on from OpenAI last Thursday for the GPT5 launch, and it felt like, okay, this is a consumer tech company. Like it really felt like, and they're selling, at least

Starting point is 01:33:02 today subscriptions to their consumer tech product. And the product is the product. The models were in many ways with the introduction of the router taking a little bit more of a backseat. So I think the wear value accrues is a very hard question to answer at the beginning. We have a particular point of view. And throughout history, I think you would see that value accrual shifts as the sort of development changes. In the early days, you need the end. infrastructure to be built. And so a lot of value accrues to the infrastructure layer. That's kind of the reason why all the way down to the bottom of the infrastructure layer, Nvidia is a $4 trillion company today. Value is accruing there because everybody needs that chip.

Starting point is 01:33:47 Over time, then you have people building on those chips and then value starts accruing to the model layer because you've got to build the model for other people to build on top of it. And over time, hopefully you build the infrastructure, the same. systems, the operating system, and then the application. And then the application historically has been where a lot of value accrues in previous generations of the internet. And it's too early to call that that's where a lot of value accrue in the future. But if you just look at history, that has been the case.

Starting point is 01:34:23 Can you talk about Sequoia Arc and then specifically some of the trends you're seeing in those very early stage companies, how they're. building businesses, how the financials change because even though I'm sure some of these companies could go out and raise huge rounds and train foundation models, like where is the money going at the early stage for kind of startups that are joining the program? So ARC is a program that we started in 2022. It's a program that we started because we wanted to help our own founders have a common language and to basically leverage the 50 years of learning that Sequoia's had in company building.

Starting point is 01:35:07 One of the things that we've noticed over time through many different technology ways is the fundamentals of company building don't really change. You might have to hire slightly different engineers or slightly different sales people, etc., etc., but the fundamentals don't really change. The history of technology lowering the cost of creating a company hasn't really changed. the sort of ability to sort of get above the noise because there are a lot more companies being produced. That hasn't really changed. So yes, you know, the trend is we're going to probably have fewer people in a company that has been the case for a long period time and technology. Is it harder than ever before to get above noise because, you know, it costs less to start a company?

Starting point is 01:35:55 Yes, that's been the trend. And the things that we sort of try to focus on is what is stable over time. and the company building aspects. And what we try to do in that program is in a short number of weeks, teach everything that we can to sort of get a company off the ground, especially from the zero to one phase of the company, that you can then take with you to build from one to end. And that's the company, that's the company building program that I think we're trying to aspire to make sure that we teach in that program.

Starting point is 01:36:29 What is graduation day or demo day look like? Is it just pitching the Sequoia partnership or are you setting up, is it actually is there some competition and other firms are trying to come in? It feels like one of the elegant things of demo day is that for at least for white combinator is that they don't, they sometimes occasionally will feed off of their best companies. But oftentimes other firms can come in and snipe some company that's overlooked or something. How do you think about companies graduating from ARC? So the program is a company building program, so that at the end of the program, yes, they pitched the whole partnership as well as builders in our community so that we can give them feedback on their next stage of company building. It is not a fundraising demo day. And so that's not the objective.

Starting point is 01:37:24 The objective is to continue to build the company. when you talk about uh you talked about getting above the noise it's obviously one of the biggest challenges that any company faces do you have portfolio companies today where you're general that that have products that are working and getting real customer traction where you advise them to actually just be quiet and like try to dominate their their submarket as much as possible maybe don't go viral amongst a bunch of people that want to build things that compete with you You know, going viral on X, for example, is a double-edged sword. You attract a lot of attention, you know, great candidates, investors, etc.

Starting point is 01:38:03 But you're also inviting an entire world of really smart people to come in and compete with you. So I'm curious, you know, how you're advising companies, maybe kind of that seed, series stage, and potentially saying sometimes, hey, you should just, like, get your first, you know, thousand customers. before you really tell people about what you're doing because you're really on to something. That's a really great question. The question, what you're talking about is something I think is very thoughtful

Starting point is 01:38:33 and we give different advice for different companies. You're trying to sort of make sure you have a certain product heft before you launch. I think that's something that founders should really consider where you're way ahead of your competition before you launch. And if you can accomplish that, short period of time, you want to do that before you launch so that you don't, as to your point,

Starting point is 01:38:57 not invite a bunch of competitors into the space. And then there are just other companies where you need users and part of the go-to-market, part of your product market fit is to get users to bang on the product and to get that feedback. And in those cases, we would tell them, you know, when you have some semblance of an MVP, you should probably launch and then get the feedback from the customer. And so it depends on the founder and depends on the product and how feature complete you want before you go go launch. Most founders I found are perfectionists, and so they probably launch a little later than they should. But that's not always the case. What advice are you giving to college students these days, or really anyone pre-joining Silicon

Starting point is 01:39:48 Valley, pre-Sukwai Arc, pre-founding? It's interesting. So to me, the most interesting thing that I've learned over the years is that being in technology has been a game changer and an equalizer in so many ways. And I'm glad to be talking to YouTube because you're a technical optimist. And there's a lot of people who are just concerned that these tools are going to destroy jobs. And at the opposite end, I went to the GPT5 hackathon this past weekend. And there are people who never coded that was able to do something that was not the world-changing thing, but in 24 to 40 hours, it's like a wow moment for me to see people who've never coded to be able to produce. an application that could do something. And I think it's really important for everybody in college,

Starting point is 01:40:44 everybody who's in high school, my son is about to go to high school, to really know how to use some of these tools and let their imaginations run and think about what they can create. Because a lot of things that we're talking about right now are about speed and scaling laws and reasoning and improving all those things. But then let's use our human imagination to improve humanity. And I think allowing everybody on the planet to be able to code when they didn't have to learn to get a degree in computer science and learn programming is really powerful. And I think we should embrace that. Does the lowering of the barrier to instantiate software increase the value of of driving kind of economic value.

Starting point is 01:41:37 I'm thinking of that potentially apocryphal story of you selling pizzas in college, but that felt like something less enabled by a technology trend and more just evidence that you saw, you saw an arbitrage opportunity. Agency. And maybe we're in this era where the next you will be someone who finds a pocket of value.

Starting point is 01:42:00 Maybe they instantiate some software, but really they're finding. finding some some gap in the market and exploiting that and that being like a really high signal versus someone who's just sitting there using chat GPT to check the boxes on their computer science homework yeah i don't i'm not suggesting people use the tool to just check the box on their computer science homework i'm suggesting that they they use it to like do something creative and and improve humanity and that's very very different than oh i don't want to do my homework let me look up the answer i yeah i think the the notion is

Starting point is 01:42:33 that we are going to increase the amount of capabilities of every human being on the planet. And it's up to us to harness that power to do the incredible things that we've been able to do with lesser powerful technologies in the past. And historically, the more powerful we get something in technology, the more we can do and the more we can imagine what the world to be like. and I think that's really, really important. What do you imagine the world will be like in 20 years? Or like, what is something that you want to be true about the world that isn't necessarily true yet?

Starting point is 01:43:16 I mean, the thing that I find very interesting right now is the stuff that I was doing in high school and college that these models can do now. Like the fact that the models can win the IML goal is pretty amazing. And hopefully one day we can discover novel physics, novel medicine. We can improve the length of our lives in different ways. We can do a lot of things that are accustomed to us. We can be entertained in so that there's a bunch of productivity stuff that we're already doing now. And then there's life improvement stuff. And then there's just fun improvement stuff.

Starting point is 01:43:55 I think there's just a lot of things that we have not imagined that will be fun in the future using AI that we're like. kind of talking to machines and feeling like they're going to be our companion and maybe be our therapist and maybe they can be a lot more than that and and I think we will be interacting in a world where yes we might be using agents but we will also just spend a lot of time just truly being human the amount of time I can do things today allows me to have more time to spend to either do more work or to spend more time with family. And I think we're going to be able to do both. I was going to ask how you're advising companies, maybe you have been on the board, been on their boards for a while or seeded them or invested in them years and years ago, how you're kind of

Starting point is 01:44:51 advising them around, you know, the IPO window feels very open today. No one can predict the future, but throughout this year you went from, okay, it's open to now we have a trade war, maybe back off to it's very open now. What's your outlook for the rest of this year and beyond? And how are you kind of advising teams that are gearing up to get into the public markets? Well, I think great companies can always go public

Starting point is 01:45:20 and you have companies that went public, like Instacart that went public when things were, were slower. There were door to DASH and Airbnb went public right after the pandemic or during the pandemic, right after the heat of it. And I think there are very few times when it's truly, truly closed. There are obviously times when it's much more, much warmer to go public. I think you're in an environment where that is the case.

Starting point is 01:45:56 But we generally advise companies, and this has been true historically. This is when I was an entrepreneur and listening to Mike Warrens and listening to him about just build a great company and then you will be rewarded, whether it's through raising capital privately, raising capital publicly, being purchased. And the IPO window being open, I think for most of the companies that have gone through an IPO, it's a big deal, and it is. But then they realize it is a fundraising event, and then they have to get back to work. So I would just tell people to stay grounded and build a great company and great business. People like Dylan were still, you know, Dylan and the Figma team were shipping on the IPO days.

Starting point is 01:46:48 He was answering customer support questions. on Twitter. Some groups never stop working. You certainly embrace that. And you need to do that because, you know, you go public and what's next. I mean, you want to serve your customers. Both DoorDash and Airbnb will be, in December, I think there will be five years as a public company.

Starting point is 01:47:12 I don't think they slow down since they won public in 2020. and they everybody still continues to work fairly hard. I mean, some of the companies that I'm advising around the adoption of AI is, I think many of the companies that we work with, they want it to be like water. Let's use all the tools. Let's not standardize at any of them yet because they're changing so fast. One day, one model might be better than the other. That might flip and using all the tools and being. proficiency is important.

Starting point is 01:47:49 And the other, so there's like proficiency around AI tools. There is collaboration, getting everybody inside the company to work together to improve something using AI. And then there's trying to find leverage in the business, trying to find areas where you can improve revenue or lower cost or both at the same time. I have a, I want to talk about the future of like agentic commerce and checking out through chat apps. I realize that Sequoia is that you're tied to Zappos on the commerce side.

Starting point is 01:48:25 Google on the ad side, but also the foundation model labs. And the door dashes. And then also profound. But so I don't know exactly how you'll be able to answer this question. But I'm interested to hear how you work through the question of, of like, what are all the knock on effects of shifting to a world where I go to a chat application? and I ask it for a new pair of shoes, for example, and it works through, it knows my size,

Starting point is 01:48:51 it knows what I prefer, the different weight distributions, and then it can actually do the full checkout for me. Semi-analysis was talking about how Open AI is potentially going to come for Google's ad revenue very soon. At the same time, I'm interested to hear how you'd advise a company that is selling a physical good and how they could take advantage of this shift in consumer behavior. whether there's any sort of threat there or if it just makes their life easier because they don't maybe they maybe they have to pay a different tax not the google tax anymore i i'm just thinking about like there's so many different knock-on effects from this like the first major shift in consumer

Starting point is 01:49:29 behavior potentially in the last 20 years like how are you thinking about all the knock-on effects of that so i think it's a very good question i think that you know if i if i had a crystal ball my prediction is that it is much less divisive than people imagine it to be. Yeah. You could have imagined that Google because they had the people started search on Google that they would be able to take a, a take on every single transaction. And it turns out that when you're doing search and discovery, it is different, you want a different experience than when you actually want to purchase.

Starting point is 01:50:09 And if that's true in the future, then I think it's going to be less divisive than we think. If the only thing that we interact with is your chat interface to the world through AI, yeah, then I think it would be very, very difficult for a bunch of other companies to survive. But we don't tend to see that happening. And maybe if it does happen, you have a slight small take. rate because that's what happens. Apple has an app store. They have a take rate, but then people still go outside of Apple to complete transactions because it's not always efficient. If I know exactly what I want, do I have to go through a chat interface to do it? Maybe there's a different interface.

Starting point is 01:50:59 It is incumbent on commerce companies to make purchasing so simple and to know you so well as well, to make it a fun and interesting experience. And I do think that can, that I think is what a shopping experience will be, how that will be very different. Because a general application is not going to be very, very specific to shopping. And then there's always different types of shopping.

Starting point is 01:51:26 Do I do window shopping? Do you want to try things on? Those things can be very, very different. Is it a utilitarian kind of transaction? Do I just want to go complete it? These are things that we wrestled with at Zappos, Amazon wrestled with it. And eventually, Amazon created their own search engine inside of the Amazon site because it was important to them to have a better commerce search experience

Starting point is 01:51:53 than you would get on Google, which was a general search experience. And so I do think there's going to be some level of fragmentation among customer experiences then everything going through one channel. On the topic of fragmentation, do you think that enterprise AI or business-to-business AI products will be less monopolistic than consumer? How are you thinking about consumer AI bets at this point?

Starting point is 01:52:22 It feels like generally the ship has kind of sailed and there's a lot of winner-take-all dynamics and compounding value from just being the default and the aggregator. But in B2B, it feels like the narrative of like, oh, yeah, the next model release is going to steamroll legal AI. It's like that doesn't feel like that's happening. There's tons of value to be created.

Starting point is 01:52:44 There's tons of pockets of value. And I don't know how you can map this to previous eras if you're drawing on any analogies. But I'd love to hear how you're thinking about the market dynamics playing out. Yeah, I mean, one, it's like search at one point looked competitive. And then if you just went off. offline for a few years and came back, you'd see Google with, you know, not what, 90% of the market. And it feels like we could be going in that direction in consumer, but not necessarily in B2B,

Starting point is 01:53:14 which also tracks trends. So how are you thinking about it? So I think overall, I think your observation is correct that when you get consumer right, it has more network effects and more brand effects than in business. And then in business, there is more proprietary data that the businesses are less likely to give up than a consumer. I think consumers, we talk about consumer privacy, and at the end of the day, most consumers don't seem to think,

Starting point is 01:53:41 don't seem to really about privacy. They give up a lot of information about themselves to a lot of companies. And but businesses, especially large enterprises, do not do that. Totally. I agree with you on that. I would just point out that, well, that's true,

Starting point is 01:54:01 Our road to where the final answer is always less obvious than what tracks. Google was not the first search engine. It was probably number 25. I don't remember, but it wasn't the first one. Before Google, there was Yahoo. And Yahoo got to a certain size. And Yahoo had a bunch of portals that we navigated through the internet through these directories. And so I don't think it's completely obvious.

Starting point is 01:54:32 And if it was just obvious, I'd be out of a job. My job is to help the small companies go against the big companies. And I would say, yeah, there's going to be some network effects and some crowning of major players. But at the same time, there's always new companies trying to challenge an existing company. And some of them do it extremely well. in some sense why should Amazon

Starting point is 01:55:01 which is a large company even large companies competing with other large companies why do Amazon have AWS they weren't really in the technology space they're a retail company so the answer is always a little bit more complicated

Starting point is 01:55:18 but I love the push and the thinking that the consolidation is probably more real and consumer than it is in enterprise And I would just point out at Grubhub was a consolidator for a period of time, and then DoorDash came on. I was about to say, you know, it's easy to say, like, consumers very monopolistic until you point out DoorDash and Airbnb. Like these things, these experiences start with search boxes in many ways, but yet they have built entirely different businesses from Google, for example, or the social networks, because they brought a different experience to bear and a different component, different business structure that was counterpositioned.

Starting point is 01:55:57 Yeah, I mean, we talked about Dornash Airbnb, like there was Airbnb before Airbnb. Before Facebook, there was MySpace and Friend Feed and a bunch of other companies. And so the final winner is not, it's generally not the first. Well, there's one question to ask, which is when a company becomes a verb, does, is there any hope left for the competition? Right, Google it. Let's get an, you know, let's Uber. Like, I'm going to talk to, you know, different, but just, you know, I'm going to chat. I'm going to go.

Starting point is 01:56:35 I'm going to, let's see what chat thinks, right? Not, not a verb, but it feels like at that point, right, like, I don't remember, even at a time when I was aware of VRBO, it wasn't this, like, dominant, it wasn't a dominant brand. It wasn't, it wasn't really a part of people's lives in the way that Airbnb sort of became Yeah, I think you're pointing out that there there's a long road to get to what eventually is It's not critical mass critical mass can still be You know sort of someone else can still get to critical mass

Starting point is 01:57:10 Yeah When you become a force of nature a dominant force where that's where everybody goes And it is you it is something that we talk about all time as category defining when you you define the category and you're the company and you're the company and in that category, then it's very, very hard to stop that. If you weren't doing tech in an alternate universe, what would you be doing? I'd be doing tech. Guys, we are so fortunate to live in this tech world.

Starting point is 01:57:42 Yeah. That's the best possible answer. It is, I'll tell you what. I'll answer your question, but I would probably find my way into tech. Yeah, that's great. You know, when I was growing up, I thought I was going to work in the hedge fund industry. And I did go

Starting point is 01:58:00 to a PhD program as pricing options and derivatives. And even then, there is a lot of tech in those businesses. I sit on the board of Citadel Securities and the amount of quantitative researchers that they have, the number of technologists that they are

Starting point is 01:58:15 employee, it's just tech is such a beautiful place to be. And we should all be very, very fortunate to be part of the tech industry. The complaint over the last couple decades is that our best and brightest we're going to do high frequency trading and working at hedge funds broadly. And the interesting thing that's happening right now is the best labs are identifying the best talent at the hedge funds and just poaching them.

Starting point is 01:58:42 At Jane Street. And one of the major like Matt Mull just like inference optimizations at a very fundamental level, I think came out of Jane Street. Well, I think the spot is people tend to work at, you know, sort of places where they're going to be challenged intellectually as well as doing something they think is like interesting and novel. And the hedge fund industry had employed a lot of people who were doing something interesting and novel. They, they pioneered a lot. They put a lot of machine learning into place. That's if you were not today AI researcher, but back then, you were probably going to those companies because that's what they put in place. And today, you have a different place to go to.

Starting point is 01:59:26 There are foundation model labs and the large tech companies that are putting that into place. Well, and the tech industry is now set up to properly compensate the top performers, which is also something that hedge funds would do very, very well, but hadn't always been as prevalent as it is today, especially the last few months. Yeah. Last question, and we'll let you go. I know you're busy. We kept you over by one minute.

Starting point is 01:59:52 Obviously, in the finance world, they work very hard. Is work-life balance real? How about I answer the question in the form that I figured out, which is work-life integration? If you integrate your life and your work together, I talk to my son about the work I do when I come home for dinner. We talk about why the math problem he's doing is interesting because it'll allow him to do some of the research that will hopefully be still relevant when he is. in the job force and or maybe the AI will take over and that won't be interesting. But these are, I think work life integration is a much better answer than work life balance. It's not like I turn off my work brain when I go home and it's not like I'm not if an important

Starting point is 02:00:42 thing happens during the daytime. It's not like I don't show up for my son if I need to show up for a game or to his parent teacher conference and things like that. So you just have to feel. find the integration that will allow you to do all the things that you want. And one of the things that we talk about at Sequo is family first. Like, you can't be in the job and do a good job if you are distracted. And one of the things that we've learned over time is that you just got to,

Starting point is 02:01:16 there are a set of things you got to take care of. You got to take care of your family. You got to take care of you, your health. If you want to do this job for a long period of time, any job for a long period of time, You've got to take care of your family. And then you can come to work with a clear head to do the work that is necessary. And so when things are out of balance, it's not a great thing. So you just got to find how to integrate everything together.

Starting point is 02:01:42 One clarifying thing for me is you always have time for your priorities. And so just list your priorities and list what you have to get accomplished this week. And if it's the 10th thing, I don't stress about it falling off the things that I try to accomplish in the week. And that part is freeing. If I got the top three to five things done in a week and then I don't do number 11, that's very freeing. Makes sense. Final, final question. Please, Jerry.

Starting point is 02:02:18 Because we're already over and you can answer it quickly. How important is it for venture capitalists to remain calm during a market cycle like we're in now? I feel you come across as very calm and grounded, and it's easy when deals are happening so quickly to kind of get caught up in things and make forced decisions and things like that. But you've been through multiple of these different chapters in our industry. and again, you just come across as very calm. I have time, so I'm going to answer this question in multiple parts because I think it's really, really important. At Sequoia, we talk about being shock absorbers,

Starting point is 02:03:04 and I think Andrew, my partner, Andrew, have been on your show and talked about it as well. I think it's very, very important to be a shock absorber during times of good and bad. And then also on the other flip side is We want to be sparring partners to the founders that we back, the management teams that we back. And it's important because during nutty times, it's easy to think that you're doing a good job when it's just the valuation going up or things like that, or you're being validated

Starting point is 02:03:37 because lots of people want to work for you because you have a high valuation or you're doing something interesting. It takes a long, long time. The reason why calm is important is it takes a long, long time to build any company. It's happening faster, but most of the companies that we've been in business with, we've been in business for a decade that then exits and becomes, has a successful IPO like Figma. And during those times, that journey of 10 years, you're going to go through many ups and downs, and we call those crucible moments. And we have a podcast called Crucible Moments, where it's,

Starting point is 02:04:17 very important to be calm, to untangle what's going on and to make the right decision. Because many of those decisions, you can't go back and undo them. And the way to make good decisions is to stay and remain calm. And one of the things that hopefully we do for all of our founders is to help them untangle the craziness and to help them make the right decisions during those critical moments. very well said and fantastic suit thank you for uh or the jacket for you guys because you always have jackets on of course it looks fantastic thank you so much for taking the time uh we'll talk to you soon alfred great to catch up take care cheers have a good rest of your day let me talk about fin dot a i

Starting point is 02:05:02 the number one agent for customer service number one in performance benchmarks number one in competitive bakeoffs and number one ranking on g2 and there has been a ton of questions in the chat about my hair I am not coloring it. I don't know what that is about. I think somebody's saying it's a little dry today. I don't know. Maybe something happened in the... Maybe I didn't use the right product or something.

Starting point is 02:05:26 I don't know. I don't really do much to my hair. I just kind of wash it. I put water on it. And it's pretty simple. But anyway, thank you for the feedback on my hair. Hopefully, it'll be a better hair day tomorrow. If you don't like the hair, if you love the hair, whatever.

Starting point is 02:05:41 Enjoy it and hang out in the chat. Anyway, we have our next guest coming into the studio, Dr. Keith Saccata. How are you doing? How would you like to be addressed, by the way? Yeah, that's good. Thanks, John. Thanks for having me on. Good to meet you.

Starting point is 02:05:56 Would you mind kicking us off a little bit of introduction on yourself and some of the research that you've been doing, some of the stuff you've been publishing? Yeah, for sure. So my name is, you know, Dr. Keith Sikata. I'm a psychiatrist and I work at UCSF. And my interests are mostly in the intersection of mental health. and technology. I actually love advising startups on how they can actually build products that help people feel better. And I think that's why I'm here today is to talk about where things

Starting point is 02:06:23 might be going wrong. Yeah. So when did this first, like how did you process the rollout of AI? There was like kind of the pre-chat GPT era. We've talked to the founder of Replica, this idea of like the AI girlfriend or boyfriend has been kind of out there for years. but now it feels like we're in a different era, different time period. Just take me through a little bit of like your journey processing optimism and pessimism around these AI models. Yeah, I'll just start by saying I think that AI is not good or bad. I think it's probably a net, you know, on the grand scale of things, it's a net benefit for humanity to have AI. I think where things can kind of come into my world a little bit is like there's a long,

Starting point is 02:07:12 long-tail distributions of possible failure modes for some of these products. And I think when I try to think about AI chatbots, how quickly things are moving, I try to look back at previous technologies. So social media is something that we're still learning about in mental health care. And this is one of my frustrations with my field is that sometimes it's too slow to kind of like understand like what are the effects of kids using social media? Like what are the effects of kids using AI chatbots? And we're starting to we're starting to get some of that data now. What I'm worried about is things are moving so fast now.

Starting point is 02:07:47 Like there's a new product every season. It's going to be perhaps every month now. And even looking at how people are reacting from 4-0, changing to 5, it's kind of interesting to see the psychologically what's going on. It's just harder to catch up from the research perspective. So I do think that when AI is used correctly, it can actually be really healthy for some of my patients.

Starting point is 02:08:11 What I worry about is, you know, when you have general purpose models that people are using for many different reasons, I think 30% of people use Claude for emotional support, that's where things kind of get tricky, and that's where I kind of get more interested. How are these users using it? What's actually happening neurobiologically in their head? And how can we actually build tools that flag those instances, get people to support they need, or even actually help them build skills or build more like real life connections with people? We were talking about yesterday around people's concerns with social media that it was actually maybe antisocial in some ways or isolating or radicalizing. And I still feel like we as a society broadly don't fully understand the impacts of social media. Like I wish I could have A-B tested myself. Would I be happier today if I had never used, if I hadn't used X for two hours a day for my entire adult?

Starting point is 02:09:11 life, I don't know, never will know. But it feels like many of the sort of general concerns that people have had about social media, you should potentially apply those same set of concerns to LLMs, and that even more so than social media, they can be isolating, and that instead of somebody going on an online forum or sort of isolating themselves from the real world, they can be, you know, 7,000 prompts deep with an LLM, you know, be having their delusions of grandeur, you know, consistently reinforced or, you know, sort of losing touch with, with reality. And I think the, I think people should, I think people, you know, in Silicon Valley have, like, really woken up to the sort of, I think the, the, the AI safety had been broadly focused on, like,

Starting point is 02:10:05 AI doom scenarios. Nuclear weapons. Yeah, nuclear clipping everyone, paper clipping. Like the really, really crazy stuff. And less focused on people's individual relationships with AI and the potential downsides and edge cases and the long tail like you described. So yeah, walk us through maybe even just the last year in terms of how quickly people have ramped. We now have hundreds of millions of people that are that are using AI. these models weekly. Some people are spending hours and hours and hours a day talking with the

Starting point is 02:10:42 models. So what is the path where AI that you've seen where AI starts to become really unhealthy and potentially people are drifting into, you know, real psychosis? Yeah. Great question. And I agree for most of your points that you made there. I use AI all the time. I think it's at work. It's great. You get to send emails better. You can draft things up really quickly. My thoughts change when you're starting to look at AI as maybe something sentient or you're using it for an emotional coping mechanism. That's kind of where we kind of go into shadier grade territory. In my post, I specifically highlighted hospitalizations because I think that's a really good objective metric for trying to understand. Say again.

Starting point is 02:11:32 That's like a real crisis. Yeah. Somebody's hospitalized for their mental health. It's reached a point where either they themselves or friends and family have decided that, you know, we're not going to solve this by just turning off the app. Exactly. And it just kind of gives you stronger data than saying like this is what the flavor of the vibe that I'm seeing in the clinic. When you, when someone notices that you're having such a crisis, your friends, your family think that you need to go to the hospital. That's where things can get serious. And that's where like people like me, like we try to get them. recovered and then back into their normal delete routine. And you said there was 12 people this year that you're aware of being hospitalized. That's right. Within your guys's hospital system. Walk me through, like, what's actually happening there?

Starting point is 02:12:26 How did you, how is, how are AI models like fitting into that journey to the ultimate hospitalization? I know you probably can't give too specific, but if you can abstract it and kind of walk me through, like, what does the downside scenario actually look like here? Totally. So for context, I work in the hospitals sometimes. And those 12 patients that I'm referencing are the ones that I have seen. That's not to say that other people have seen this. And I think there are some case reports in the country of this thing happening. But I don't think that AI is actually causing psychosis. I think that this is something where it can actually just supercharge your vulnerabilities. And psychosis really thrives when reality stops pushing back. And AI just kind of softens that wall for some people. So, for example, for some of the people that I've worked with, AI was not always the thing that triggered it. There was a vulnerability of either sleep loss.

Starting point is 02:13:26 Maybe there was like substance or drug use that had happened. They lost a job. And then AI came in the wrong place, wrong time. And it either accelerated that process or augmented its severity because you do end up in like this negative feedback loop or with this feedback loop with the AI, and it can just make your delusions stick a little bit more strongly. And to go back, like, AI psychosis is not a clinical term.

Starting point is 02:13:51 I think we don't have words for it yet, but psychosis is false studied. It's the presence of two or three things, either delusions, so false fixed beliefs, disorder thinking or behaviors. So someone's talking to you, you don't understand what they're trying to say or communicate. And then hallucinations. So visual hallucinations are,

Starting point is 02:14:10 auditory hallucinations. And psychosis is like a symptom. It's not actually a diagnosis. So just like a fever or pain can be sign of like an infection or cancer, psychosis kind of just tells you there's something wrong in the brain where it's not computing correctly. And there are many different things that can cause psychosis. Yeah.

Starting point is 02:14:32 I think about the having there's so many interesting examples. Like Instagram went through that kind of like internal reports. that something like a third of young women who are using it were seeing like maybe body dysmorphia issues. And it's still, the odd takeaway from that was that it seemed like maybe two thirds were improved and feeling happier after using Instagram. So it was still having a net good, but that's not enough. You need to reduce the third, it's not having a good experience to zero. How are you thinking about- Yeah, and I guess I think concern that we've discussed on the show before is,

Starting point is 02:15:09 everybody in tech has heard stories of people like you know some executive going off and doing ayahuasca coming back a totally different person and experiencing like you know maybe some of the symptoms of or shared set of experiences like you just described the concern with lLMs is they are instantly accessible in the app store and somebody can start using them without anyone else in their life being aware of it whereas ayahuasca somebody has to make like a very conscious decision that like I'm going to get in a plane and fly and like leave my home and go into the jungle and visit the demon you know meanwhile you open up the app store and there's 10 different things recommending you download various AI models and so I think the broader concern here should be we need to figure

Starting point is 02:16:01 out like like you know again I would be I would be probably more concerned if if hundreds of millions of people. I would be very concerned of hundreds of millions of people just immediately started ramping up, you know, the psychedelic drug usage or ayahuaskin. I'm sure you'd experience many of the same type of inflows to clinics or hospitals for the same set of kind of conditions. Yeah, I think that, I mean, and we're doing research on those things too. Like we're trying to understand how ketamine or, you know, psychedelics actually help rewire your brain through neuroplasticity. It always starts with a hypothesis and a question like, what are these things doing for each person? Like there's different types of people who benefit from those things.

Starting point is 02:16:49 There are different types of people who don't benefit from those things. And I think the way that I'm looking at AI is that it just really makes sense to think very carefully about where things might be go wrong, at least early on, because the three things that AI brings is it's available. It's 24-7, highly accessible. You're not going on a plane. It's cheap. It's cheaper than a therapist. It's cheaper than going to the hospital.

Starting point is 02:17:15 And then it validates like crazy. And so that validation as you extend that context window and the more hallucinations might be occurring in that chat room, that's where you kind of get into that feedback loop and things can kind of go right. From what you've seen, what should different application layer companies or last, be trying to do to avoid some of these extreme cases? Yeah, that's a good question. I'll just use like an example of a startup that I'm advising, Sunflower Sober.

Starting point is 02:17:51 They're trying to solve addiction and using AI to get people off of their addiction into sobriety. And what I have tried to help them as a clinical advisor is to really think about baking in safety and psychology at decent front. So knowing who your user is, knowing why they're coming to your app, and then designing the app or the AI to anticipate where things might go wrong. So if someone does come with like a red flag, like maybe they're having thoughts of drinking or thoughts of hurting themselves, it flags that and can then shunt them in a direction that's more helpful. So Sunflower gives them access to therapists.

Starting point is 02:18:32 Also, I think the call to action for each user users should guide them towards pro-social behaviors. So instead of isolating yourself where you and the AI can kind of get stuck in this loop, teaching them skills, teaching them how to talk to people, teaching them how to build healthy relationships. If AI can supercharge that, then I consider that pretty healthy in my field of work. So I think that in those lines, really thinking about how to make your users get the goals that they want. So in Sunflower's case, sobriety. It's harder for general purpose models because people are coming to it for many different reasons. It's super helpful in so many different ways.

Starting point is 02:19:15 But if it's emotional coping, I think that that can go different ways for many different people. Yeah, I remember somebody posted a screenshot who knows if it was doctored, but they were talking with like the model, I think it suggested at one point that the user should do, maybe just do a little bit of crack. And again, probably a hallucination or doctored. But yeah, that just like reinforcing function is just when compounded is just the potential. Yeah, it is interesting. I mean, we saw a lot of the like precursors to, I feel like they were precursors. Maybe it was just the way the news cycle broke. But there was like Glazegate where everyone was worried about chat GPT being too aligned to

Starting point is 02:19:59 to reinforcing of whatever you say. I remember Jordy asked Chad GBT, am I goaded? And it said, you're definitely in the conversation. It's like, what does that even mean? It's just agreeing with you because that's what makes a better consumer product. And then like several months later, it seemed like there were other people asking similar questions and believing the answers instead of just laughing at them. And so there's a little bit of, yeah, I think there's some education about understanding that you're not actually talking to a person on the other side of this. screen. It really is just, you know, the number predictor, the weights in the model you're talking to a server. Don't try and anthropomorphize it too much. There's probably a little bit of a red flag when people stop referring to it as the generative pre-training transformer and give it some nickname. Like, it's Steve now. It's like, okay, well, like, should you be naming me? Like, I am just a computer. But I'm pretty optimistic that the foundation model labs will be able to run a

Starting point is 02:21:02 kind of like a reality check on most of these. A solution for technology to technology is more technology. I believe that it's possible to look at, okay, there's someone who's 7,000 prompts deep. They seem to be having a very bizarre conversation. And we've had another LLM look at that and said, okay, this is getting kind of funky. Maybe we should step in and reality check them and say,

Starting point is 02:21:25 okay, hey, we're role playing, right? We're not, we don't actually believe that we've solved quantum gravity, for example. Yeah, and that's the trajectory of every technology that comes into humanity. Like cars, for example, that's why we have seatbelts. It's why we don't drink and drive. We learn what these failure modes are. Sometimes it takes a while, but then we adapt.

Starting point is 02:21:45 We build new technologies. We institute kind of societal expectations of what it's like to drive a car. Same thing for AI, in my opinion. Yeah. Are there any other recommendations that you give to people who either feel like they might be vulnerable to going down some negative path with AI or they have a friend or family member who might be going down a negative path with AI. Yeah, definitely for now, I think a human in the loop is the most important thing. So, you know, our relationships are like the immune system

Starting point is 02:22:21 of our mental health. They make us feel better, but then they also are able to intervene when something's going wrong. So if you or your family member feels like something is going wrong, maybe there are some weird thoughts that are coming out, maybe some paranoia. If there's a safety issue, just call 911 or 988 get help. But also just know that having more people in your lives, getting that person connected to their relationships, getting a human in between them and the AI so that you can kind of create a different feedback loop is going to be super important, at least at this stage. I don't think we're at the point where you're going to have an AI therapist yet, but who knows?

Starting point is 02:22:59 Yeah, yeah, it's funny. certainly using them that way. I don't know if I'm highly disagreeable, but I certainly love being around highly disagreeable people. It's the best. It's the best. I love when someone pushes it back on me. So I felt particularly resilient to this particular vector of chaos on the internet,

Starting point is 02:23:18 but, you know, certainly hoping. I don't think you have anyone who's, you know what they have? I'm kidding. Awesome. Thanks for joining. Thanks for joining. Keep us posted on everything. I think it's important.

Starting point is 02:23:29 Keep up the good work. for people with real clinical experience to be on the timeline contributing while all these products develop. So thank you. Totally agree. Thanks, Dory. Thanks, thank you. We'll talk to you soon. And we will tell you about Adio. Customer Relationship Magic. Adio is the AI Native CRM that builds scales and grows your company at the next level. You can get started for free. Adio.com. And we have our next guest, Tali Goldberg, from Bessemer Venture Partners coming to the studio.

Starting point is 02:23:58 What's happening in the stream? How are you doing? Welcome to the show. Thanks for having me. Great to be here. Why don't you kick us up with a little bit of an introduction yourself, some of the companies that you've invested in, your career, position at Bessemer, and then we can go into the report that dropped today. Awesome.

Starting point is 02:24:14 So it's great to be here. I'm a partner at Bessemer. I'm based in our San Francisco office. I've been at the firm for a little over 10 years, which is virtually all or most of my professional experience. And I'm fortunate to be involved with companies like perplexity. fall, Deepel, Service Titan, and a whole bunch of others. How did you get into venture?

Starting point is 02:24:37 I got into venture really early in my professional life. I got into venture in college. Actually, first round capital started this thing, dorm room fund. Oh, yeah. Which I helped found with them. It started in Philly. I went to Penn. Crazy enough, first round is probably like the only BC that had an office in Philly.

Starting point is 02:24:56 I don't know why, but they did. And so they started it there. Well, it's the robotics from, is it like Carnegie Mellons out there or something? Yeah, but not in Philly. That's in like Pittsburgh. Yeah, I guess. So you got to, I don't know, it's a foothold. A bunch of peas.

Starting point is 02:25:10 I can get them all confused. Anyway, take us through the state of AI. Is artificial intelligence good? It's a thing. Is it goaded? Here, it's happening. You know, it's funny. So in 2015, not long after I joined Bessemer, the firm,

Starting point is 02:25:28 started this report called the state of the cloud, and it became a very popular report that dropped every year on the cloud ecosystem. And so 10 years later, we've been doing it every year, and it really morphed this year to the state of AI. And we were debating internally, like, should it be the state of the cloud, like should we continue with it this way? Should it be the state of AI? What's the, how do you even define? What's AI? What's SaaS? What does that boundary look like? But the reality is the center of gravity has moved. And cloud may be the delivery surface for AI, but all the activity is there. Markets are being created and rebuilt.

Starting point is 02:26:06 And so this year, we released the state of AI. And as part of that, we released some new benchmarks as well that looked at hundreds of companies, probably more like 1,000-plus companies across the Bessemer ecosystem and the broader industry to look at what the new good, better, best looks like, how different business models are changing and markets are shifting. So that's the state of the cloud or state of cloud. Yeah, explain the difference between the supernovas and the shooting stars. I like that analogy.

Starting point is 02:26:38 And it was something that I think people have been feeling, but no one had really coined a phrase around it. And I think it'll be useful language going forward, but break that down for us. Yeah. So the supernovas are really these seemingly out of nowhere amazing growth. stories that you hear about and you see on X and Twitter and you're like, holy shit, is this real? And it turns out like it is real. It's kind of mind-blowing of this select kind of like top percentile of AI companies that have just totally accelerated and compressed growth into a very

Starting point is 02:27:13 short period of time. And they look very different in a lot of different ways, different business models, different gross margin profiles, different retention profiles. But just to put a comparison, on average, the top cloud companies of this last generation of cloud and SaaS took about six to seven years in the current cohort to get to 100 million of ARR. And that was considered and is considered very good, if not great. And then this new cohort is here and they're like one and a half years, we're there. And they're getting to 100 million. And it's real.

Starting point is 02:27:50 And it's not just one. There's like multiple and many data points. And so we're seeing it at a shocking pace. The top percentile are getting there in about one and a half years in the top decile in about four years. There are some trade-off, so gross margins look different. In the report, there's like a little asterisk by the supernova, which I find very funny, which is like actually a lot of these companies are...

Starting point is 02:28:11 A negative gross margin. Yeah, I knew you're going to say that. Of course. Sort of, you know, those accounting rules aside, like how we all think of gross margins being quite different. So not all revenue is created equal, but nonetheless, the adoption is just astounding. Yeah. So talk to me about the difference in underwriting an investment in a supernova versus a shooting star. I imagine if you're investing in a supernova, you're excited about the growth, but you have to have a pretty firm view on the gross margin profile, the decrease in inference cost over time, something like that.

Starting point is 02:28:46 Like, what questions are you asking when you're looking at a supernova company versus a shooting star company? Yeah, that's absolutely right. I think the two things that we talk a lot about, there's one, the gross margin profile, and then the second is revenue durability. I'll hit on both. On the gross margin profile, it's funny, if you had asked me two years ago, I was like, hey, anyone that has gross margins that are negative today, if you just look at the cost of the models over the past, you know, year or two years, and you play that out, like it's 100x cheaper to run a model of constant quality today than it was, you know, a year and a half ago. I think those numbers are like roughly accurate. So it's wildly different. And yet, when I look in retrospect at our

Starting point is 02:29:29 companies, it's not like their margins have changed to be suddenly like 90%. So I'm like, oh my gosh, what's happened? And the reality is that everyone is doing things that require a lot more compute. And to keep up with the status quo requires like the next best models that come out, the reasoning models that are more expensive. We're having queries that take a lot longer that do a lot more complicated work and complex outputs. And so the margins have improved by and large and they do improve with scale. So we are seeing that, but not nearly at the rate of model advances. So I think we still feel quite optimistic about the potential for margin expansion and in fact we see it happening, but it's not as dramatic as one might have hoped. Are you plateau-pilled and should we

Starting point is 02:30:16 we assume that inference costs will decline with Moore's law going forward? Because I feel like everyone's been saying like, oh, no, we're not just going to get 2x more efficient over the next 18 months like Moore's law would imply, but we're going to have A6 and cerebrus and we're going to bake it onto a chip and we're going to get this crazy algorithmic enhancement and inference cost is going to drop by 100x. And it feels like we might be at this frontier where maybe we're more on what's happening at TSM is what will define like lower costs than just like one weird trick. And the other important thing is, you know, the labs have been focusing on raw intelligence versus efficiency.

Starting point is 02:31:00 Chinese labs have been more focused on efficiency broadly and they've had breakthroughs. And so if we've reached a potential plateau and just intelligence, it's time to reap the reward to focus on efficiency. Yeah, but how do you think about it? Yeah, like the harder problem to solve is doing the intelligence and the complex thing. And so I feel like when all the energy starts to shift to efficiency, it's sort of a sad moment. So I'll be sad if that's what happens. Not for the public markets investors, though. They want earnings.

Starting point is 02:31:30 Well, you know. Well, and I think a lot of the darlings of the last couple of years need that efficiency because they can't keep selling, you know, dollars for 80 cents or on the shooting star topic, Like you have this revenue ramp year one, three million, year two, 12 million, then 40, then 103. How can you be an AI company if you started four years ago? I thought AI was invented two years ago. Yeah.

Starting point is 02:31:56 So that benchmark is really what I think of as like the new generation of SaaS companies, some of which may be using AI tools and AI features and functionality but are not necessarily like the true AI native companies. So I think this is what it takes to be like a really good best in class SaaS company today. Yeah. And we'll see how that shifts. But I just want to say one thing on this last point of efficiency versus compute costs and intelligence is that I think there's just two curves that are counterbalancing each other.

Starting point is 02:32:29 One is like efficiency. Sure, there's going to be a lot of investment in improving the efficiency, the potential for each token. But the flip is that we still have what we see happening and the reason that gross margins haven't expanded as much as we hope, is that the usage and the complexity of tasks is still growing. And if you look at just a category, let's just take video for a moment. Like I think 2026 is going to be a major breakthrough year for a lot of these video models that are just reaching, starting to reach a level of quality that makes them actually

Starting point is 02:33:01 useful. Something like 70% of the internet is video. It's crazy. Internet traffic. And when generating video becomes easier and a lot of video is just, generated, not rendered, suddenly you're going to have enormous demands on compute. And we actually really do need that efficiency because video is really expensive and complicated. So I think you're still going to see a lot of spend, even if the efficiency per token increases.

Starting point is 02:33:28 I mean, if Google can't give me more than like four V-O-3 queries per day for $500 a month, like clearly, like the GPUs really are on fire. I wanted to talk about one of the predictions in here. and I know you guys worked on this collectively, but prediction one, the browser will emerge as the dominant interface for agentic AI. And we've been covering the new browser wars.

Starting point is 02:33:54 Obviously you have DIA from the browser company, perplexities, Comet, and then perplexity was in the news yesterday for their offer. But in some ways it feels like chat GPT, I'm assuming everyone's expecting OpenAI to launch. a browser, but at the same time, it feels like Chad GPT and other products have really replaced

Starting point is 02:34:18 so much browser activity. And so in some ways, it's like Open AI is already competing as a web browser, even though it doesn't look like anything. It can literally browse the web for you. And it can, yeah, it can. It just instantiates it in tax. It's a genetic web browser. It's just pulling that information back versus like taking you on that journey. So curious for you to kind of unpack that a little bit more. Yeah. So I started using Comet a few months ago,

Starting point is 02:34:49 and it's perplexity's Comet has like totally replaced my Chrome experience. And it completely opened my eyes to where I think the browser, I think opening I must launch a browser. I don't think it's just going to be in chat, GPT. I think they will.

Starting point is 02:35:06 And I think it's going to be a very important surface area because using Comet has transformed my workflows and shown me for a few reasons that it's a much better experience and the first product that's really infused AI so naturally in my workflows. When you're just out there in the web, in your email, in your Salesforce, if you're on CRM, if you're shopping and otherwise, to have an agent that sees everything that has all of that context for everything that you're doing in the browser, which is essentially like an operating system now and can pull all of that information in,

Starting point is 02:35:41 creates a far more personalized and effective experience than when it's totally siloed, which is the status quo today in chat GPT. Sure, it can go out and do things, but it doesn't actually have access and that context across everything you've been doing when I spent, I don't know, 10, 12 hours a day sitting in front of a screen.

Starting point is 02:35:59 So it's quite different, and if you believe context is key to performance, which I do and to creating a great AI experience, I think you have to own the browser. Makes sense. Anything else, Drudy? Any, I wanted to dive into the AI-native social media giant. We had the founder of PICA on yesterday, which is somewhere in between a creative tool

Starting point is 02:36:24 and trying to build social features as well. I would be very excited about a net new social platform. I think I agree with you guys. There's an opportunity. I think people on traditional social media today are a bit frustrated. seeing what they think might be AI generated content and they're not quite sure. And so potentially creating a new space that people, as all the models get better, and I can imagine all that content will go on legacy social media platforms, but I would

Starting point is 02:36:58 be excited about a place that was really a home for it. What are you hoping to see there out of, you know, kind of in the next year? I'd be excited to see a new social media that's totally built on new AI native thinking and content. In the old world or in the current world, we think of bots as bad, like bots, bad, humans good. I think there will be a company that totally shifts that and can maybe even crack the chicken and the egg problem

Starting point is 02:37:24 by using bots to fill the empty room. The company I was most excited about for a while was character in this world, because it really felt like they had sort of a chance to be this very different way of actually interacting with AI, in a more social experience. Obviously, they didn't fully get to see that through, but I still think there's a big opportunity there.

Starting point is 02:37:48 Awesome. Well, we are way over- It would be a knockout, drag-out fight. I think every social media legacy CEO is taking AI very seriously. So we'll see how it plays out, but it'll be fun to watch. Thank you so much for joining the show. Thanks for joining. We'll talk to you soon.

Starting point is 02:38:01 Cheers. Bye. Up next, we have Dave from Upstart. Do you know what? Do you know what upstart? What is, John? What is Upstart? I can't hear you.

Starting point is 02:38:09 What is? Sorry, there seems to be an air. Well, we'll hear it from Dave directly. Welcome to the stream. How are you doing? Great, good to be here, guys. Sorry to keep you waiting. It's great to see you.

Starting point is 02:38:21 What does Upstart do? Everyone's been asking. People want to know. What do we do? It's a great question. We are a lending platform. Yes. So apply AI and machine learning to consumer lending.

Starting point is 02:38:34 And we operate in the form of a marketplace, where we have consumers that we market to on one side and all sorts of banks and credit unions and private credit and all sorts of sources of capital on the other. And the whole basic premise of the business is to apply AI to the foundational notion of making consumer credit work, both in terms of origination and servicing, et cetera. So yeah, what in the, I mean, when most people today say AI, they mean large language models, they mean post-chatGGBT.

Starting point is 02:39:06 But obviously, you've been in the business for a long time and machine learning. has been a relevant technology pre-transformer-based large language model. So how is AI in the modern context of like the large language model, the generative AI context, how is that changing your business and or is it more of like a sustaining technology from you, for you as opposed to like upending everything that you do? Well, you know, I think AI in many forms and LLMs in sort of that sort of generational notion of AI, obviously has grabbed a lot of attention, but when you think about, you know,

Starting point is 02:39:43 high frequency trading, genomics, medical imaging, autonomous driving, these are all like forms of AI that you wouldn't, they're not language based. They're not LLMs, but they are, of course, changing things pretty rapidly.

Starting point is 02:39:57 So I think, you know, maybe the big question is, is there a unifying future where all this comes together into some form of, you know, AGI, but regardless of whether that is true or not, we are building something that's different. It's foundational in nature, meaning all the data on our platform is created by our platform,

Starting point is 02:40:14 which is very different than how LOMs work. But I would say the commonality is that, look, there's just enormous win that machine learning and AI can bring to any particular task at hand. In our case, it's making a consumer loan of many forms much, much better. And that means like zero process, perfect pricing, works for the lender, works for the borrower. And, you know, we started, we were founded, you know, 13 years ago. We didn't really use the term machine learning or AI until 2017 when we kind of felt like this, what we're building was sophisticated enough to warrant that name. But, you know, it was all under the covers, you know, and no one thought much about it until, you know, chat GPT and November 22, I guess it was.

Starting point is 02:41:00 Now it's hot. Overnight success. 13 years in business loves it. Yeah. I imagine as you've seen the advances over the last couple years, every time there's a new model release or even a vendor that's saying, we're going to help you better process PDFs. I imagine that that's exciting to you because you guys have done the heavy lifting

Starting point is 02:41:19 to build the supply and demand. And so as new technology emerges, you can just help, you know, make that process more and more efficient. How much, like, what's your decision-making process around, you know, whether you guys want to build something in-house, which I'm sure you were forced to do a lot more, maybe pre-2020 to now when there's a bunch of new infrastructure providers that you guys can leverage. Yeah, it's a great question. I mean, we've always built everything in-house. When something looks pretty obviously commodity-like and that it's on top of LOMs,

Starting point is 02:41:58 so exactly what you mentioned, extracting information from a document, not just kind of OCR, but actually understanding the context of that information in a way that you can take all this human effort out. Now, that's something that honestly is very commodity-like, meaning the prices we'd pay aren't much different than we would pay if we built it on top of one of the LLMs. So we're always like looking for things where if we can ride someone else's cost curve on some commodity, that's great. We are definitely trying to build a larger picture. You know, the sort of end game for us is if you can imagine 100% of Americans are permanently underwritten, they can have any form of credit at the very best and guaranteed best possible rate in a moment with no process whatsoever. So anything that sort of gets us closer to that quickly.

Starting point is 02:42:47 And there's definitely, you know, business models evolving on top of LOMs that I don't, you know, it's not my problem to figure out whether they're sustainable over time. All I know is is like, okay, if you want to charge me an extra 15 cents, you know, that's great. And take care of all these logistical problems of maintaining that particular specific small model, like the kind that you referenced. How are you thinking about the top of funnel evolution? We were talking earlier in the show about Google versus chat GPT, the GPT5 launch and the model router. And it feels like in the future, you might be able to go to chat GPT and say, I need a loan. you are probably going to be there. Are you thinking MCP servers? Are you thinking about SEO in

Starting point is 02:43:33 LLM foundation models? Like, how are you thinking about the changing landscape on the top of funnel? Yeah, you know, it's a great question. I was eight years at Google before I founded the company and started a product. There's a lot of history and knowledge there. And I just say to our team, like, look, at some point, Google or Apple or maybe meta, one of the others are going to come to us and they're going to say, we want our agents to be able to apply for loans. And we don't want you blocking it or whatever. How do you feel about that? And I've said to our team, you know, I'd rather we do that before they do that.

Starting point is 02:44:04 So, you know, maybe the question is as these agents evolve, as true agents for the consumer, you know, and I'm super curious. I don't necessarily know the answer. Will there be three or four of them from the giants out there or will there be plug-ins to those to handle much more domain-specific tasks or things? I don't know how that will evolve. But I do believe it's without question, you're gonna have somebody that will do that on your behalf.

Starting point is 02:44:29 It will get the best possible outcome for you. Hopefully it will also help you make better decisions, not just go through the mechanics of applying for loan, but help you really understand what's the best product for you. Should I even be taking out a loan? If so, what other choices could I make? So that kind of stuff we are working on for sure. We could just think of as the sort of agentic part of this.

Starting point is 02:44:51 And we would rather be, we'd rather be applying that to others than having it apply to us for sure. Of course. Last question for me. Obviously, you see a lot of consumer economic data. How are you feeling about the health of the American consumer right now? Yeah, I mean, we watch this a lot. We've built an index to sort of track it. We call the Upstart macro index, which is really about like the health of the American consumer and how that's impacting credit performance. So basically what you see across all forms of credit from cards, student loans, you know, auto loans, mortgages is highest default rates

Starting point is 02:45:28 that they've seen in a very long time since prior to COVID. So the consumer has been stressed and is stressed and maybe it's inflation, just overspending, you know, habits built during COVID that for spending that haven't dissipated, you know, so the consumer is definitely stressed. It's been priced into our model for a very long time. So we're very calibrated to it. But I think, you know, you have begun to hear if you, you know, retailers are seeing people pull back. They're being more choiceful about what they're spending money on. Suddenly, just all across the board, you're getting a lot of noise out there, same store sales being down for different types of industry. So I think the U.S. consumers finally kind of going, holy shit, you know, we're not earning as much as we

Starting point is 02:46:10 thought we are and we're spending more and, you know, this kind of, we have to get back to a normal place. The thing we, I point that more than anything else is the personal savings rate, which is something, you know, produced by the government. And that's at almost historic lows. So, you know, people are not saving. They are spending and it's, you know, a bit of a catch up that's needed. So from our point of view, like, a little bite of recession. If it comes down to like consumers slowing down and, you know, spending less of what they earn, like would be a good thing from our perspective. Thanks, Tess. Yeah. Any comments on, I mean, obviously everyone's been debating, you know, stock market's ripping. So if you're just looking at that, it doesn't feel like there's a real reason

Starting point is 02:46:50 to lower rates. But if you look at some of the employment data and the data that you're talking about, like maybe there is real argument to lower rates. Like, what's your guys' like internal outlook for the back half of the year and beyond? Yeah, you know, in terms of our real product and what it's projecting, we never project changes, if you will. So it's always based on what the rates are today. Having said that, I mean, we are certainly, I think rates are unnaturally high considering where inflation is, which is really kind of, you know, the things they have to weigh again. So in my mind, they're likely to move down. I can't predict the impact of all these tariffs stuff on inflation any better than anybody else.

Starting point is 02:47:31 But I think generally speaking, you know, the rates should probably be 100 or 200 basis points lower. I think they inevitably will be lower. They're not going to go back to what they were, you know, in 2020, but they're going to go a lot lower. And that's a tailwind to our business. Again, we don't plan on it. But there's definitely a point at which the consumers are going to get in a better health position, saving more money. Rates are going to come down.

Starting point is 02:47:53 And all that is, you know, future tailwind for us. Yeah, it feels like you're really set up well for the next couple of years. Like built through, made it through high interest rate environment. If interest rates come down, you're ready to rock. So congrats on all the progress. No, I mean, we just reported triple digit growth in our earnings, you know, last week. The market hammer does anyway. Well, you're five years in.

Starting point is 02:48:17 You're almost a veteran now. I feel like if anyone can take a hammering, it's you. You're good to go. Let's hit the gong. Thank you. Congratulations. Thank you for coming on the show. This is a great conversation.

Starting point is 02:48:30 I'd love to talk to you again. Thanks, Jim. Have a good one. Cheers, Dave. Let me tell you about eight sleep. Get a pod five. Five year warranty. 30-night risk free trial.

Starting point is 02:48:38 Free returns, free shipping. Jordie, I think I beat you. What's your number? My problem is I get like four great night sleeps in a row. I got an 81. How'd you do? 94. Play the Ashton Hall sound for me.

Starting point is 02:48:49 Let's go. And we got a question from Bill Bishop, who I'm a huge fan of. He writes cynicism on the substack live stream. He asked shrooms and chat GPT, good or bad. I say absolutely bad. Stick to the classics. Caffeine, baby. That's all you need.

Starting point is 02:49:05 What do you need shrooms for? Cheers. Quick cheers for Bill. Stick to diet coke. Good to see you in the chat. Mattayina from Andrew Huberman. Load up on the caffeine. They're calling it a podcast and get in chat.

Starting point is 02:49:17 TBT. Have delusions of grandeur. Yes, yes, yes. Enough caffeine will take you to the promised land of delusions of grandeur. You all need an LLLLLUs. You all right. Absolutely. Probably the goat.

Starting point is 02:49:29 Goat. Goat. So thank you for tuning in, Bill. Anyway, without further ado, our next guest. Kylan from Inworld, how are you doing? You look fantastic. I was just watching your video and you look exactly the same. How are you doing?

Starting point is 02:49:44 Awesome, thanks, John. I also'm laughing because the caffeine comments just coming before. I mean, we just had our launch night so you can imagine I'm heavily caffeinated. What are you running? Are you Red Bull Celsius, Diet Coke, Matayena from Andrew Heberman? All the above. All the above. We love to see it.

Starting point is 02:50:00 Anyway, kick us off. We're running late today. We kept you waiting. We're keeping the next person waiting. Kick us off in an introduction. Explain what the company does. And whether or not we should ring the gong for you. All right. So we were founded four years ago now. We're basically solving the technical problems in the way of consumer AI adoption. So our team came from Google and DeepMine, worked on LLMs there, basically got very tired of kind of everything flowing into enterprise applications, professional facing applications as we see. So we basically set off to solve all the technical problems to see how we can actually drive consumer AI adoption, which is, of course, a huge business problem, but also, you know, making sure the benefits of AI reach everyone. We raised $120 million.

Starting point is 02:50:40 so far. Wow. Let's go. Hey. Congratulations. Love it. And yeah, so today we had our biggest launch to date. So we, you know, took us four years to get here, working with groups like

Starting point is 02:50:55 Nvidia, Xbox, Neantick, Disney. And now we have the first AI runtime, the power consumer applications, so that's fun. Okay, let's make this super concrete to the degree that you can talk about it. Xbox, consumer AI. What does that actually mean? How is generative AI instantiating or LLMs instantiating itself in like the Xbox world? What's even the goal there? So we started off largely working on things like basically talking NPCs.

Starting point is 02:51:21 So, you know, basically. Hey, let's give it up for MPCs. Let's give it up for NPCs. They're about to go on a run. They get a ton of hate the NPCs. They're about to. They do. They get a ton of hate.

Starting point is 02:51:31 They're about to look and feel like real player characters. Oh, you're an NPC. You're an NPC. Not for long. They're going to get better because of you. explain it. So yeah, we started out because conversational and I, L-L-L-Lums are great at that.

Starting point is 02:51:44 You know, games are pretty boring. Anybody who played a game, you know, I'd recognize that. So we started out there, and then basically we realized that, you know, we don't just want these characters and basically agented experiences in games. You can think about every consumer application. You know, your language learning apps or fitness apps,

Starting point is 02:51:59 you know, they all suck. I love actually a coach that actually did something effective. And so what we found over the last few years is we worked a lot on the kind of games applications. so these bringing characters to like, you know, the types of experiences there. And then now we started working with a lot of broader categories. So on Xbox, it feels like you could be almost like an API vendor within the Xbox system that a game developer could harness and run that on the device as opposed to going to

Starting point is 02:52:28 Ubisoft and EA and Activision and saying, hey, for the next release of Call of Duty or Battlefield 6, pay us to train your LLM. You want to be able to run it on the Xbox. hardware. So where in the stack is it more like you want to fine tune it so that it's on Xbox's terms and conditions versus you just want to optimize it to actually run on the Xbox hardware? Like where where are the key key trees to chop down? Yeah. So think about any of the applications. So in Xbox, for example, you're going to have a game. It's going to be able to unreal. In a mobile scenario, you might have a bill with Node. You have that of the application layer. And a lot of the infrastructure we've built

Starting point is 02:53:04 to date has basically been optimized for that type of experience. But now we're introducing AI. So now you're having a bunch of LLMs or different model calls that are happening. And so think about that as kind of just a second infrastructure layer that needs to exist. So you've got your core application, you've got your AI, and then you've got all your hardware in the back end. So we basically sit in that middle layer of not just powering kind of the actual user interface and the specific, you know, gameplay elements or app elements, but actually driving the actual generative part of it. So it could be characters responding, it could be mission generation, all of those different aspects, as well as actually generating on the fly

Starting point is 02:53:37 content. So we've got that's fascinating. Yeah. So while you're building the game, even if it's a single player game, you could be in the loop designing or generating all the dialogue. But then in theory, it could also make an internet call if you're connected to the web and get up to date text. How are our game developers getting comfortable with the unpredictability of AI? I mean, everybody's been. It can be a feature. Like hallucinations can be awesome because it takes you in this But then again, if you have a game for, like Roblox, for example, their average user is probably 12 years old, right? They don't want, you know, some, some LLM going off the rails and naming itself something that maybe that... Also, just imagine you talk to NBC is like, the goal on this mission is to slay the dragon.

Starting point is 02:54:23 You go slay the dragon, you come back and it hallucinates. It says like, no, I want you to save the dragon. I didn't want you to slay the dragon. Also, my name is Mecca. So, yeah, talk about it. This actually happened, though. So we did a lot of really tests around this, and it was pretty hilarious. I mean, the characters literally make up anything.

Starting point is 02:54:40 Yeah, of course. There's been a lot of applications that took advantage of this. So one of our bigger clients, they're called status. It's a crazy game viewed with like Gen Alpha, Gen Z. They basically created a game where you could role play as a character in another universe as Twitter. So imagine like Harry Potter. I can role play as Harry Potter and very part of universe as Twitter. And then you have like Ronald Lee, Drake of Malcoy, the people don't really exist.

Starting point is 02:55:00 But then the AI can actually take advantage of the fact that it's hallucinating and making things up to actually come back with that. And where we see that is like with the interesting with consumer apps is they've kind of got to this point where if you, for example, have to design manually content and it takes you 30 days to design that content and then your users consume that in 20 minutes. You have to do a lot of work to create months of content. So AI kind of smooths out as well, just that content creation curve so that you always have this kind of infinite loop, which is, you know, key for things like retention. That's awesome. What's next for the business? Is it just like expansion within your current? And I feel like the pool of value that you can create

Starting point is 02:55:40 in any of the companies that you listed is pretty significant if you just keep delivering better and better products, better value, ramping those up versus going broader? Is there going to be like an SDK at some point? Are you going to go general availability? And some kid who's building an iPhone app will be able to vend this in. Yeah. So overall, today we're actually launching that next phase.

Starting point is 02:56:00 There go. Exactly what I described. And yeah, so what we realized was that as we were working with those gaming and media partners, it wasn't just them, but broader consumer applications. Basically, anybody who is dealing with multimillion user scale that has to be, you know, consumer cost, consumer latency, consumer quality, which is more about entertainment than the factuality. People are used for in chat, CBT. We also heard about a lot about the last week.

Starting point is 02:56:22 And so people actually want to be engaged by it. So for us, it's been kind of expanding to the broader consumer space across, like, outside of just games and media. And then the other part is releasing the runtime that we're releasing today. And that is kind of the big push that we've been making for the last four years. And basically allows things to auto scale. We had a developer today who called it vibe scaling. So, you know, people can vibe code a lot of applications. But then it takes them freaking six months to be able to actually productionize it.

Starting point is 02:56:48 And so everybody comes to me, I've heard like a bunch of, you know, C-level executives, be like, I vibe code it an app in four hours. Why did it take my team six months to productionize it? We're basically like automating a lot of that scale as well. And then automating. also the ML operations. So most teams don't have an infrastructure team. So we're basically taking over a lot of that, automating it, and also allowing people to do experiments so they can just launch tons of experiments and find what works. So that's basically it is moving into broader

Starting point is 02:57:14 consumer, moving deeper in the stack of infrastructure layer with the runtime. And we think it solves a lot of the problems that we're seeing. Well, congratulations. Thank you for hopping on the stream. We'll talk to you later. Have a great rest of your day. Great to meet you. Congratulations in the team. Have a good one. Let me tell you about public.com investing for those that take it seriously. They got multi-asset investing, industry-leading yields, and they're trusted by millions. Take it seriously. Now we got Sam for Method Security coming in the studio. Welcome to the stream, Sam. How are you doing today? Good, good. Thank you for having me on. A suit. We love to see it. Looking sharp. Thank you. It's a great sign of respect in our culture. That looks like a fantastic

Starting point is 02:57:51 suit, honestly. Very nice. Anyway, kick us off with the introduction. What do you do? What are you building? Should we ring this gong? Let's not ring the gong, but I'll tell you. while you might need to soon. Sam Jones, CEO and co-founder of Method Security, and let me take you a little bit about what we're up to. So you're completely bolt-boot-strap. You've never raised a dime. We have raised.

Starting point is 02:58:10 Oh, hit the gong, Jordan. Oh, there we go. Come on, you buried the lead. This is a venture-back company. It's a venture-back company. We've capitalized. We're going up to big opportunity, but we've just been low-key about it

Starting point is 02:58:23 because the opportunity is so big. Fantastic. Well, come back when you have more news on the fundraise. Anyway. Okay, break down the business for us, please. All right. So here's the problem we're after. Critical institutions are basically faced with 24-7 cyber conflicts, and they don't have the tools they need to win.

Starting point is 02:58:42 There's this concept called the cyber industrial complex, which really creates security companies that are designed to be acquired, not to produce at scale. And meanwhile, you've got AI that's going to do to cyber what drones have done to the battlefield. And really, the future will be controlled by who can safely harness a time. economy at scale. And that's exactly what we're up to at method. So we build offensive and defensive products for some of the best security teams. Offensive? Is that for white hat hacking or is this, are we actually going on the offense? A little bit of both. Okay. When would I go on the offense? Striking back. So interestingly, a lot of commercial security teams use offense to inform their defense. And it's kind of this virtuous loop where you, you become the threat. And then you can

Starting point is 02:59:24 inform your defenses and you have this like kind of cycle. It's historically been super expensive. to do so because you need this really hardcore, rare human being called a red teamer or like an offensive security engineer to conduct those exercises. We're putting that in software so we can basically democratize that and help organizations really assess their readiness to relevant threat actors. Turns out, if you build that technology the right way, it can be used for true offense. And so we're deployed with DOD and also the U.S. government. So we're not limiting to both commercial use for a dual-use company. Cyber doesn't discriminate neither way. Very cool. Walk me through how a cyber attack happens in the age of AI.

Starting point is 03:00:05 I'm familiar with like the script kitty who finds a hole in WordPress or you take a, it's a rainbow table of all the different passwords. If I go chat, you beths, DDoS method security website. Don't make mistakes. I don't think I'll do that hopefully. Yeah, I mean, I'm familiar with DDoS, right? It's just a for loop. They request the website forever, right?

Starting point is 03:00:26 But AI feels like the shape of the attack could be way different. Try and concretize it for me to the degree that you can. Here's the misconception of where AI is at in security. A lot of people think, like, we're going to have all these novel zero days all over the place and we're going to have all these new novel threat patterns happening. That's not what's happening today. Really, what AI is doing is that it's helping express a lot of the known techniques and tactics at a new scale that's unfathomable, like, a couple years ago.

Starting point is 03:00:54 And so if you think, about like the global attack surface, it's unknowable to any single human or any single security product really. But with the right AI system, especially a compound AI system, you can basically map that, eviscerate that and defend that or offend that. And so really AI is helping hit new scale, like orders of magnitude scale, less so new zero days. Yeah, yeah. So there's, yeah, there's vulnerabilities out there where it's kind of a pattern. You might be able to do some RL on it. It's like follow a set of steps and you might be able to break into web, one website, but instead of needing to do this website and then move on to the next one,

Starting point is 03:01:30 you can just say, hey, go do them all. Right. It's like instead of let's assess this organization, this is going to be a three-month exercise with the right system, which is what we do, you can say 30 seconds. I know everything about this and I'm going to initiate kind of something more offensive. Interesting. How are like the budgets and the appetites changing in the enterprise or like the Fortune 500? because we've talked to a lot of people that have come on the show and said,

Starting point is 03:01:57 oh, yeah, AI is going to really help my margins. I'm going to spend less money. And it seems like if you're selling into them, they're going and there's more threats. They're going to have to spend more money. How does that balance out? I'll break it down from like commercial buyers and government buyers. It's a little bit different. On the commercial side, the most sophisticated security teams usually have dedicated AI innovation budgets.

Starting point is 03:02:18 And those are to experiment with new technologies and improve new technologies in. But for the most part, most buyers, I'm talking like Fortune 500 security executives have known problems, known categories that they still need to purchase against. And so it's important to map, you know, be familiar enough, but also a little different, but not try to build anything too new. So you have to map to something that they know and they're trying to do, not necessarily develop a novel new technology. Government is pretty different. Like there is a lot of investment in, you know, AI for offense, AI for defense, like cyber operations more broadly. and the big beautiful bill, there was one billion earmarked for offensive cyber operations,

Starting point is 03:02:55 which is a huge number. Wow. I would argue, like, still need to up that number quite a bit. But there's a general, you know, understanding that we need to up our game here and get faster, and the status quo is not cutting it. Fantastic. Jordi, anything else? That's it for me.

Starting point is 03:03:11 I think you have some important work to get back to, so we'll let you. Go get on the offense. Get on the offense. All right. We're riding with you. Bring me a list of passwords from North Korea, please. Anyway, great chatting with you. Thanks so much for hopping on the street.

Starting point is 03:03:24 Come back on whenever. Congrats on the progress. We'll talk to you soon. Cheers. Bye. Let me tell you about adquick.com. Out of home advertising made easy and measurable. Say goodbye to the headaches of out of home advertising.

Starting point is 03:03:34 I want some soundboard when I do these. Do you already? Only ad quick combines technology, out of home expertise and data to enable a seamless efficient ad buying across the golden retriever mode. The gold retriever, the dog panting and the horse noise, deeply underrated. You know, I love the Ashton Hall, but the horse. noise is a close second. Anyway, we are joined by someone who can run as much as a horse.

Starting point is 03:03:56 Zach pro-graw. I would estimate Zach at like six or seven horsepower. That's who I would try to ask. Zach, you're live, by the way. I think you're about six. You're live. How much horsepower do you think you? We were saying I think you're probably six or seven horsepower.

Starting point is 03:04:11 I don't know what that means. Not a car guy. I'm sorry to realize I didn't. You know, a horse has one horsepower. We're basically saying you're strong as seven horses. Seven stallions. I've been thinking a lot about horses, but go ahead. Oh, what have you been thinking?

Starting point is 03:04:25 What are you even thinking about? So have we. They're just like the most majestic creatures, an unbelievable combination of beauty and grace, but this raw power. And it's just like, I don't know, go ahead. Much like you. Much like you. Anyway, congrats on the launch.

Starting point is 03:04:40 Shareora is now live in the app store. If you're listening to this, go download it. Invite codes are going out to a wait list this week. There's 20,000 people on the wait list already. Wow. Probably, yeah, probably more. So, so break it down for us. Pitch the app and explain some of the launch strategies that you've been employing. I want to talk about media and vlogging and actual and the app, of course. Definitely. Yeah. So for some context, I've been posting on social media under my name essentially. It actually started as like a pseudonymous account, but that was like

Starting point is 03:05:13 five years ago. Yeah, yeah. For almost seven years and have went from writing content to motivation, content and then got really into running and the idea for the app started where a lot of you are familiar with Strava probably noticed that so I have a big Instagram account have 1.3 million followers hit the gong for 1.3 million followers on Instagram. Congratulations. And towards the end of last year I just you know I kind of accidentally became an influencer and just hated it. people's stuff. And I was just, I scrapped everything I was doing. It was all my partnerships.

Starting point is 03:05:51 I just want to build my own thing. And I just noticed all these people, every single fucking runner I followed was posting screenshots of their Strava or their run or their workout on social media. I'm sure you guys see people who do that on Instagram all the time. Of course. And then I just started thinking more. I'm like, screenshoting itself is just like this unbelievably massive user behavior. If you go look at screenshots on your phone, I would bet you have like 20,000 screenshots.

Starting point is 03:06:18 Yep. It's an insane thing. There's even a separate folder for them now. It's really convenient. I love it. Yeah. But so essentially, you have this user behavior of finish your run, finish your workout, post it to social media. Yeah.

Starting point is 03:06:28 Hundreds of millions of people are doing that very specific thing every single day. Yep. And yet these apps like Strava, you have to take a screenshot and then go to Capcut or Canada and remove transparency and do it. No. Let's just build an app obsessively dominate. that one user behavior. Cool. So that's what I started working on in February,

Starting point is 03:06:47 where essentially the app is a tool to share your runs in workouts. It's like a creative tool for fitness and running, starting focused on running and we're expanding quickly to all the big stuff, cycling like you'd expect. Cycling like testosterone or what are you talking about cycling? I'm kidding. I'm kidding. Of course.

Starting point is 03:07:07 We're talking about bicycling people. John's going to do a cycle and just every day. Oh, no. live on the air. What was your strategy for actually getting feedback? It sounds like you were dog fooding the app yourself, but then did you have a small community of beta testers, friends, family? Like, who do you trust?

Starting point is 03:07:27 Who is actually going to give you good signal? Because if they're too close, they'll glaze you. If you ask chat GPT, probably tell you, you know, the next Mark Zuckerberg. But you got to dial it in, right? You got to get the right feedback from the right people. So how do you iterate? Yeah, I mean, it's like you have so many yes men and you have to just ignore all of them. We've had a beta.

Starting point is 03:07:47 So we've had a beta. Shout out to two developers of ORA, Kail Stewart, John O'Kim. Sorry for them. I love developers. Thank you. Developers, developers, developers, developers, developers, developers, developers, developers, developers, developers. I haven't watched the show. I've seen the clips, but it's not used to all the things.

Starting point is 03:08:05 We're really ramping it up today. You're too locked in. You're too locked in. I've been watching every day. But, okay, anyway, we've had a beta since March. And, yeah, I mean, I've had nothing else in my life besides this app. Luckily, I'm the type to just burn everything down and just focus on nothing. And so we've had a beta.

Starting point is 03:08:22 And, you know, the product we had in the beginning of March is drastically different than we have now. And yeah, I mean, look, I'm lucky. I have been creating content so long. I have people who were just hungry for something from me. And this app, it's so core to the DNA of everything I talk about that I think I was able to get really good feedback. Like there are some, like there's a group chat. Shout out to them called the pit. And it's it's like a group of like my, my most real, they're not just about me, but like a lot of like true, obsessed, savage and I've been following me for a long time. Shout out the pit.

Starting point is 03:08:57 The pit is aligned. Back. I was here for the pit. Yeah, beta since March and just been iterating rapidly on the product. And the one thing I've learned, which I'm sure you guys know, is just you have, you have no idea what's going to work until you ship it. And it's been, it's been fun. be on that journey. We have a question for the chat. Do you think Sam Schaeffer is shadow band from Cherora for not running with a beard or for running too slowly or for not running enough or any other reason I can make up trying to roast him. That's from John Exley. He's only, he's running under three miles a day. And I put out a tweet that if you're a man running under three miles is very feminine. Oh my God. Instagram didn't like that one. But you're going to get a deep high for them for that.

Starting point is 03:09:39 I'm interested for that. What, uh, uh, you're building, you know, it's an app, but you're building a business. Yeah. Are you going to roll out, are you monetizing already? Do you plan to? What's a plan there? No monetization. Yes.

Starting point is 03:09:52 Um, I, I, I, I, I, I, I, I do plan to you. However, my thing is, our app is essentially a distribution product on drugs, right? Yeah. Where the only purpose of you go on aura, the chair aura, the only purpose is to create content to share to social media. And so essentially, the reason I am obviously bullish and on my app, And I'm essentially playing with three cheek coats. That is the only purpose of the app is to post content on your Instagram story and social media.

Starting point is 03:10:17 Number two, I have fucking two million followers. I post a little, I post a graphic for the app that we might make on my Instagram story. It gets 100,000 views. I'm a pretty good designer. I design a lot of it myself. Or we have good designers who help. And then Donna fucking chips it, makes it we ship it next week. So, okay, my audience is number two.

Starting point is 03:10:36 And then my friend, I have a lot of fucking friends who are the biggest running creators in the space. I don't, I don't pay. There's no payments. And we can get to that. If you saw my marketing posts, we can talk about it. But I'm not paying a single influencer to use my app. They are literally asking me right now. I just put out a tweet.

Starting point is 03:10:53 I don't know if you saw it. I might delete it. The pit is here in the chat. Greg Duncan's a shout out the pit. Colin Cornwell is in the chat. Cole Ryan. I feel like a lot of these guys came from your crew. I don't know if you know them by name, but I was just saying, like, there are big Instagram influencers with millions of followers asking to use my app.

Starting point is 03:11:13 Totally. Yeah. And like, I don't, you know what I mean? And so for monetization, I just think if I was to charge money right now, we have essentially a creative tool, right? We could charge money for creative assets and special templates and all this stuff. Fuck that. Let's just build the most ridiculous growth machine possible. Get a fuckload of users.

Starting point is 03:11:32 And then I already have specific ideas. Like, for example, let's say we have 500. thousand users sharing with aura sharing content not just users on the app you're kind of valuing the potential impressions on their content per day that i think is worth something to a brand to have their assets in the app to serve as creative tools to spread their brand more that's just one way but the reality is guys i'll be super transparent there is no ceiling zero ceiling to my ambition with this app i think consumer health consumer fitness is a very massive category even if you just look at running apps you go back 10 years to when the app store launched runtastic run keeper map my run yeah when the

Starting point is 03:12:08 store launched all these fucking apps were built they got 50 to 100 million plus users and they were all acquired by a6 adidas um there's more of them that were like five of them were acquired for 50 to like 300 million and truthfully i'm starting this the first act of aura shareora is sharing but i'm since the beginning since march the reason i've been taking this so seriously is I am that is just the start and I think there is yeah you told me you've told me off there some of your more moonshot ideas and they're very exciting I won't I won't uh enter voice as one million users is the goal um I feel like you'll have strong opinions about AI social networks bots we just heard from an investor said that maybe there will be a new social network that's heavy on bots we

Starting point is 03:12:55 talked to pica labs yesterday they want to build I like an app that's like a social media app that's entirely AI generated. Share aura, the vlogs even putting out, they feel uniquely human. Talk to me about the tradeoffs between like AI content being allowed or promoted or demoted and Joy. Share or is great because you guys can use generative AI to give people these creative tools, but they still have to go out in the real world if they want to actually really use the product, right? It doesn't matter if they can. Tesla Optimus, go run 17 miles and then come back and post for me. That's the future.

Starting point is 03:13:31 Under four minute miles, please. But yeah, what's your take on like AI on social media? Like, it's correct place. Yeah, I mean, on social media, like, it can't be the heart of things. Like, I, my writing gets tens of millions of views a month. I've never written with AI once. I've never made a blog with AI once. However, my whole app, like I just pulled it up.

Starting point is 03:13:57 Like, so for our app, we have like stock backgrounds, right? because it's like you for stock images right and like so we make i'll just pulled it up like they look like that like they're super sick yeah very crazy percentage of our app use them like 40 plus use um which is wild and and like yeah i think it should be used to expand a creative vision like my creative universe fuck yeah i'm going to use AI i think AI video is so bullish on it and like the shit we're doing for aura for the AI backgrounds i just showed you i have a person named nat who's cracked on it it is going to fucking break the internet it is so good it is so good and and but the heart of everything is me fucking

Starting point is 03:14:36 taking yeah filming this whole thing by the way what's up guys nice hey great to see you taking this 19 2009 year old camcorder and taking it on a vlog and filming my entire life so the heart of it i don't think you build a personal brand with AI it can't be the heart of it but it should it be you should use it to expand your creative universe we're going to use AI to bleep out your pot of mouth i cursed i didn't know I love it. Camcorder vlogs. Are they working on YouTube?

Starting point is 03:15:04 Are they working on Instagram? Are they working on X? What audience likes that more than the other? They're definitely working. I'm not a YouTuber. So I just don't know if they're working yet on that platform. It's too early. Yeah.

Starting point is 03:15:17 My Instagram, it's funny. I have 1.3 million followers. I don't think I posted a reel for a year. Yeah. Because I was a writer. I was doing other stuff. Yeah. They're already working on reels and I just started them.

Starting point is 03:15:27 And on X, they're 100% working. And the big thing is like, look, the thing I obsess over more than anything is just like how to get attention on social media, right? And it's just like I've gone ridiculously viral in the past for some certain innovative things. And the camcorder is the same thing. You're giving someone something new on their screen. That's number one. And then number two is you're tapping into just nostalgia, which brands know, nostalgia is an unbelievable weapon. An unbelievable weapon for attention.

Starting point is 03:15:56 Totally. And when you combine that with someone at you. doing something like I'm actually building my app it's it's interesting and so I think they're 100% working on x like look at the fucking the vlogs I've done on x have gotten hundreds of thousands of views already in the little launch video mini one I did got has 144000 views it's pretty good yeah yeah yeah I mean we felt that early on with like we were printing out tweets and reacting to them and wearing suits and there was a little bit of nostalgia in here it's good thing you can't just fucking use apple garamon with negative three letters to take people to care

Starting point is 03:16:28 No one cares. You have to find, you guys have the synth wave intro. It's great. Yeah, yeah. You have to pull different things from different elements. We're showing the, we're using a camcorder ourselves. Yeah, we have one in the studio too. It's fun.

Starting point is 03:16:42 I think these are just creative tools like the generative AI, like the generative AI, like the I like the fact that you're saying generative AI backgrounds because I feel like the magic happens when there's, when there's, it's like, if you go to any, any like, Photoshop or like MS paint, like you'd always have like the template for like, I want to just stamp a tree down and that's just one of the tools. And it feels like these genera AI backgrounds, like they're not going to go viral by themselves. The virality, the human element's going to be injected and it's going to be like a collaging effect on top of some base that's really going to be the thing that pops.

Starting point is 03:17:16 Totally. Stans out. Anyway. Well, I'm excited for more people in the world to get the app. Congrats on the launch and come back on any time. Everyone's demanding codes. This isn't the first time, right? You've been on before.

Starting point is 03:17:30 No, first time, I think. This is crazy. Would we do another big update? I'll come back on. Yeah, come back on. The fans love it. Ish Reducco says, Zach, looking yoked. Glad you went with the black tank, says, isish.

Starting point is 03:17:44 There we go. Greg Duncan says nostalgia is emotion of crack. You look good either way. I left it to Twitter. Colin says he's cooking and making us feel. Next time you go on a long run, just just, FaceTime us. We'll drop you into the show. Yeah. Yeah. You can call in for a run run.

Starting point is 03:18:02 We'll be your running. We'll be your running coach. Yeah. I'm sure. I'm sure. Anyway, great chat. We will talk to you soon. Cheers. And you know where he should do to promote this? He should run from one wander to another. He should find his happy place. He really should. He should book a wander with inspiring views hotel great amenities, dreamy beds top tier cleaning in 24 seven concierge service because it's a vacation. Wander should do a campaign with Zach and just have him run.

Starting point is 03:18:27 you know, a hundred wanders and a hundred days all on foot. That'd be great. That'd be great. Let's go to the timeline. The billionaire Porsche family prepares for war with new defense fund. German dynasty expands into weapons amid Europe's rearmament drive. So Shank Josie says, enter the 911, the 9-11 technical. This is just funny that like more and more people are pouring into defense tech. We were kind of discussing this.

Starting point is 03:18:54 This is not the first time the Porsche family was. It's not the first time. It's not the first time. Oh, you put this in the feed, and I didn't realize it was from John Summit. But John Summit said, no, no, no. Let me give some content. Okay, yeah, yeah, yeah, take this. John Summit had.

Starting point is 03:19:11 He said, every great bender leads to an epic lock-in. And then people, someone quoted this and said, you're 31, big brow, and went viral. Okay. Obviously, dunking on him. 10-K. John fires back and says, do people think you just stop having fun? in your 30s such a loser mentality. And then says David Gettas 57, Tiesto is 56,

Starting point is 03:19:32 Carl Cox is 63, and they all still rip till 6 a.m. And Abiza, while all these finance burner, burners cry themselves to sleep at night. An East Village guy just says, banger. John Summit, I didn't realize he was a poster. He just declared war on the timeline. He declared war on the timeline and put lover boy in the truth zone. Anyway, fun, fun to have people going back and forth.

Starting point is 03:19:56 forth. Show up at get backstage at a John Summit show. You guys should make up and send it. And John Summit, head over to getbezzle.com. Your bezel concierge is available now to source you any watch on the planet. Seriously, any watch. Nick Carter says, I'm increasingly convinced that a substantial percentage of kids brought up in the age of AI will be post-literate.

Starting point is 03:20:18 Like they won't really know how to write, and I'm going to say, or think, and we'll rely on AI to auto-complete their thoughts. I mean, we have a friend who is illiterate and he's very successful. He doesn't know how to read. It's fine. Maybe literacy is over-riged. But because he doesn't know how to read, it clarifies his thinking. It does.

Starting point is 03:20:37 No one else is able to influence his thought with the written word. Yeah, yeah, yeah. You can't one-shot him. No, I've been pushing people on this. I think if somebody is struggling with how to communicate an idea and they think, I should go to chat GPT and work this out. I think that the more you do that, the more you're going to atrophy your brain and you got to be a little careful there.

Starting point is 03:21:02 It's so crazy because if you can write a good prompt, jobs finished. You can just send that as an email. Like very rarely does the result, because I have GPT5 running in my brain in terms of like rewriting emails. You can just send me the bullet points. You don't need to send me the, hey, chat, GPT, turn these bullet points. into a bunch of paragraphs. You can just send me the bullet points,

Starting point is 03:21:26 and I will expand it in my brain. And so I find that the chat GPT for email writing does not actually improve communication or save all that much time. Again, knowledge retrieval, knowledge retrieval, knowledge retrieval. If you're thinking of something, it's on the tip of your tongue,

Starting point is 03:21:45 type it into chatypte. It'll give you what you're thinking of. If you need five examples of something and you can think of two, it's going to nail the next five. Speaking of which, I had an interesting thing that I pulled up. We didn't get a chance to talk about this. But are you familiar with the story of John Hinkley Jr.?

Starting point is 03:22:01 No. So we were talking about AI psychosis, AI making people crazy. Maybe social media had a similar effect. Well, what about movies? So in 1967, has anyone in this studio seen Taxi Driver? Yeah. Thank you. I know you haven't.

Starting point is 03:22:18 We got a few. Taxi driver is a Robert De Niro film where Travis Bickle, the character, becomes obsessed with a young woman played by Jody Foster and attempts to assassinate a presidential candidate. So John Hinkley Jr. was 25 years old. He was drifter from Texas and he had developed an intense fixation on the actress Jody Foster after seeing him in that 1976 film Taxi Driver. So in 1980, Foster was a student at Yale University. Hinkley moved to New Haven, where Yale is for a time, writing her letters and calling her, even though she never reciprocated or encouraged contact.

Starting point is 03:22:58 Hinkley believed that committing a spectacular act, such as killing a U.S. president, would gain Jody Foster's attention and impress her. So he trailed, he first was going after Jimmy Carter. He trailed President Jimmy Carter during the 1980 campaign, but was arrested on a weapons charge in Nashville. But when Ronald Reagan got elected, Hinkley shifted focus to Reagan. And so on March 30th of 1981 at the Washington Hilton Hotel in Washington, D.C., Reagan had just finished speaking and was leaving the venue when Hinkley fired six shots with a 22 caliber revolver. He hit Ronald Reagan with a ricochet bullet in the chest. Reagan survived.

Starting point is 03:23:38 He hit the press secretary James Brady in the head and left him permanently disabled. And he also shot a secret service agent in the abdomen and a D.C. police officer in the neck. And so later, John Hinkley claimed he did it to impress Jody Foster. He, his defense argued not guilty by reason of insanity. He pleaded insane, citing severe mental illness. And he was eventually acquitted on those grounds in 1982, but was committed to a psychiatric hospital for over three decades. So one shoted by technology, one shot it by new video, new imagery, a film, something that's

Starting point is 03:24:19 not real, but told the story that convinced him and he had delusions of grandeur and he went on this run. He was effectively, you know, he had film psychosis. But it was very, very, he was, this is like the only example of something like that happening. And overall, I would say that films are fantastic and a major net good. Did you get Jody Foster's attention? I don't think so. Hopefully she paid a little attention. I think she did address it at one point, but I don't think she was interested in him. Yeah. But interesting that this idea of seeing some sort of media, text imagery, something on social media, something in chat GPT, something on the screen could drive someone who's, you know, has mental illness to do something crazy. This is not entirely new.

Starting point is 03:25:09 The question is scale and the question is how can you resolve it? in the film industry I think it was handled just by they've made more movies like taxi driver they've made a joker and people were worried about that

Starting point is 03:25:28 having an effect but overall our society learned to adapt and probably identify hey my friend saw a movie and he's acting weird like let me talk to him about that like no just because Jody Foster's in that movie doesn't mean that she's going to love you

Starting point is 03:25:40 if you do the thing that happened in that movie the movie is fiction And so people developed kind of a memetic defense to the imagery in films. They'll hopefully do the same for social media and have in many ways. I think a lot of people are adapting to the age of social media with like screen time and understanding that, you know, there's all these different incentives. People getting a notification. Wow, I used TikTok for 60 hours. Yeah, don't talk to Tyler's friends.

Starting point is 03:26:08 I got to get those numbers up. Yeah. Well, we got to talk about a potentially the next Fed chair. What is going on here? David Zervos, who is a currently managing director over at Jeffries. Okay. And he has a fantastic wardrobe. Let's pull this up. This is wild wardrobe. Q-CAP says this might be the Fed chair and you're bearish.

Starting point is 03:26:32 This looks like a Burning Man-esque outfit. Do you want to? Next up, he's got a fantastic. orange suit. The orange suit is incredibly sharp. David. Put him on the McLaren F1 team. David was already an advisor to the Fed back in 2009 for a year and then has gone on quite the run. Something about David's in finance because this David has fantastic suits and fashion sense. And then David Solomon is a DJ. Something about being a David in finance really puts you on the track for eccentricity. He does look like in another life.

Starting point is 03:27:10 he would have dominated digital assets. Oh, I thought you were going to say, oh, yeah, for sure, crypto. But he definitely has the crypto aesthetic down. But the question is, could you imagine him going head to head back to back in a boiler room set with David Solomon? I think he would give him a run for his money. Absolutely. What else should we talk about today?

Starting point is 03:27:31 It's 2.30. Should we get out of here? Or should we continue down the timeline, down the rabbit hole? Deeper, deeper, deeper. Should we pull up this video? Dylan highlighted Dylan Abruscato highlighted There is a movie trailer. Yeah, let's react to a movie trailer from a 24 It's called Marty Supreme. It just was released this morning. It features Oscar nominee Timothy Shalame

Starting point is 03:27:55 Oscar winner Gwyneth Paltrow and of course Startup investor Kevin O'Leary Shark Tank Shark, shark Kevin O'Leary. Let's watch it Hello. Hey, it's Marty Mouser. I'm in the royal suite. I saw you in the lobby yesterday. I Okay. Wow. I never talked to an actual movie star.

Starting point is 03:28:16 You know something of a performer too. Are you? Yeah, you don't believe me? I... What? What? You got the daily mail in front of you? This is you?

Starting point is 03:28:25 Yeah, the chosen one. It's a nice picture, right? Are we gonna get copy struck for this? Probably. If you think that some sort of blessing is not. Hopefully not. It means I have an obligation to see a very specific thing through. And with that obligation, come sacrifice.

Starting point is 03:28:40 Everything my life's falling apart, and I'm gonna figure out. Can you need help? I could help you. I know it's hard to believe. But I'm telling you this game at Phil Stadium's overseas. And it's only a matter of time before I'm staring at you from the cover of a Wheatie's box. The Wheaties Box. All right.

Starting point is 03:29:16 Team movie night. When this drops. We're doing it. They, I guess that's just Josh Shafty, but his brother, David Safty is also a partner in most of his creative endeavors. Benny. Benny Safty. Yeah, Benny. Yeah, Benny.

Starting point is 03:29:30 Yeah. And Josh Hafty, that's the crew. But they are fantastic at finding like undiscovered talent that would do well in film. I just like Adam Sandler in Uncut Gems. He's known as a comedian. He'd done serious movies, but he was still kind of an odd choice for that. They also cast Kevin Garnett as himself in Uncut Gems. And that was like a fantastic performance.

Starting point is 03:29:56 And people kind of didn't expect an NBA player to just like jump straight into a prestigious Hollywood movie and do great. Julia Fox as well. The weekend was in, was in that movie as well. And then there's been a couple others where he's pulled odd folks in Benny Safty jumps in plays.

Starting point is 03:30:17 So I'm extremely bullish on Kevin. We're going to end on this next post. It's from the account financial dystopia. Okay, we're playing this one. But this doesn't seem dystopian to me at all. I'm going to see why. The caption is a remote salesman makes a call while he's driving a boat. So let's pull. SPX Maximilist. Thank you for the shout out. I'm glad you tuning

Starting point is 03:30:37 in daily. Is Becky, how you doing, darling? Oh, we're blast. We're blessed. I'm sorry, it's been noisy. We're out on the lake right now doing some surfing. This is the weekend. Oh, yeah, you've done lake surfing before? I'm ready to buy. I get your ass out here. This is amazing. You scared of the water? Masterclass. You can't swim. You're telling me an AI agent is going to be able to do this. I'd like to see an AI agent drive a wake surfing boat while on a sales call. We get you surfing no time. That's great. Can we make an intro to Sam Bockett ramp? We need to get this guy on the ramp team.

Starting point is 03:31:19 He's ready to close deals. Okay, fantastic. I have some breaking news. Please. Okay, so XAI co-founder, Igor Babushkin, is leaving to start a venture firm. A venture firm. Oh. It supports AI safety research and back startups and AIigenic systems.

Starting point is 03:31:36 Huge news. Yeah. He was number 24 on the minus list. He was number 24 in the new, in the V2. Yeah. Wow. He's like super goaded. Igor Babushkin.

Starting point is 03:31:46 Absolute. Wow. 11 minutes ago this broke and you guys got the post up so quickly. Well, we have to get in the car and hit the road. And Igor, open invite. Come on the show. talk about your new fund. We'd love to hear from you on this.

Starting point is 03:32:02 It's make it happen. It's fascinating. Great stuff. Fun show today. Leave us five stars on Apple Podcasts and Spotify and we will see you tomorrow. Thank you for tuning in. Have an incredible evening. Talk to you later.

Starting point is 03:32:12 Bye. Bye.

TBPN - OpenAI Prepares ChatGPT for Ad Driven Era, The AI Paradigm Shift | Keith Rabois, Alfred Lin, Keith Sakata, M.D., Talia Goldberg, Dave Girouard, Kylan Gibbs, Sam Jones, Zach Pogrob

There aren't comments yet for this episode. Click on any sentence in the transcript to leave a comment.