TBPN - Jony Ive Joins OpenAI Analysis, Google I/O Reactions | Keith Rabois, Dan Shipper, Reggie James, TJ Parker, Ben Hylak

Episode Date: May 22, 2025

TBPN.com is made possible by: Ramp - https://ramp.comFigma - https://figma.comVanta - https://vanta.comLinear - https://linear.appEight Sleep - https://eightsleep.com/tbpnWander - https://wander.com/tbpnPublic - https://public.comAdQuick - https://adquick.comBezel - https://getbezel.com Numeral - https://www.numeralhq.comPolymarket - https://polymarket.comFollow TBPN: https://TBPN.comhttps://x.com/tbpnhttps://open.spotify.com/show/2L6WMqY3GUPCGBD0dX6p00?si=674252d53acf4231https://podcasts.apple.com/us/podcast/technology-brothers/id1772360235https://youtube.com/@technologybrotherspod?si=lpk53xTE9WBEcIjV(04:10) - Google I/O Breakdown (42:54) - Jony Ive Joins OpenAI Analysis (01:00:33) - Dan Shipper. Dan is the co-founder and CEO of Every, a business writing collective and newsletter platform for founders, operators, and technologists. He also writes Chain of Thought, a popular column exploring tech, psychology, and self-improvement. (01:27:50) - TJ Parker. TJ is the co-founder of Matrix, a stealth health tech startup. He previously co-founded PillPack, the online pharmacy acquired by Amazon for nearly $1 billion. (01:56:20) - Reggie James. Reggie is a writer and technologist who recently published Hardware, a book exploring identity and creativity in the age of machines. He previously co-founded Eternal and continues to explore the intersection of internet culture, software, and design. (02:27:55) - Timeline (02:30:18) - Keith Rabois. Keith is a general partner at Khosla Ventures and formerly with Founders Fund, with a storied track record as an operator and investor at companies like PayPal, Square, and OpenDoor. He is known for backing and advising dozens of high-growth startups in consumer and enterprise tech. (03:01:53) - Ben Hylak. Ben is the founder of Raindrop AI, a platform focused on intelligent information capture. He previously co-hosted the Latent Space podcast and has built a reputation for bridging AI infrastructure with practical workflows for teams.

You watch TVBN. Today is Thursday, May 22nd, 2025. We are live from the Temple of Technology. The Fortress of Finance. The Capital of Capital. We're, have a great show for you today. Folks, let's pull up what we're going through. We're talking about reactions to Google I.O. still. Joni Ive joining OpenAI. We're going to break that down. Anthropic just dropped Claude 4 and is blowing out benchmarks. It's AI Week on TVPN. And we got a bunch of great guests. Dan Ship. TJ Parker, Reggie James, Keith Roy. You created the first entirely AI ad to promote a live stream. I did.

Should we play it on the stream? It's on my profile. Did you already share it into the group so the boys can find it? Or should we? Or should I send it in? I've been having a ton of fun with V-O-3. I upgraded to the super expensive plan. It's worth every dime even to just get a few funny eight-second videos out of this thing.

The fidelity is remarkable, and it just looks awesome. It looks awesome. Yeah, let's play it. I don't know if we have sound, but let's play it, because I added music, but maybe, I don't know, the music might be copyrighted. Let's see if it handles it. Okay, play this. Play it from the beginning.

There we go. That's us. I guess that was a close one. Yeah, too close. In the yellow Ferrari. CNBC is attacking us. They don't want us to podcast. They don't want us to podcast.

They don't want us. Okay. We're being followed. We're bringing media to Hollywood. Helicopter. The CNBC. Okay. Oh, now you're fighting guerrillas.

Versus 100 gorillas. Successful. As expected. Oh, and then the ramp out. Nice. Oh, how that's that you? Switch your business to ramp.com. Time is money.

Save both. Easy to use corporate cards, bill paying an accounting and a whole lot more all in one place. Again, at ramp.com. Anyway, lots of fun with that. I'm having such a good time. We got a, this is going to be a fun guest lineup. You already covered it a little bit, but we have a great group.

Reboy coming on, the legend, the myth. Dan Shipper from Evry, who raised around earlier this week and has a lot of thoughts on Anthropics, new release. And then T.J. Parker is getting the band back together. Can't wait for that. The Pilpac band with a new company called General Medicine and then Reggie James

HIPCity, Reg is going to be coming on and talking about Johnny I. I can't wait for that. I think he's got a big announcement as well. And I am excited for all of it. So should we talk about,

should we newsmax, John? Let's newsmax. Let's go through Google I.O. Obviously, this happened on Tuesday. It's now Thursday, but Ben Thompson's done some analysis, breaking it down, thinking about what what the implications are for the stock.

I mean, the stock popped a ton. And it seems like, you know, Google's on this weird cadence where it takes them a full year to like really release a bunch of stuff altogether through these like big keynotes. And they're more on like a waterfowl cycle than just dropping random stuff. But clearly every serious AI company is thinking about how they're scheduling releases. It's not a coincidence that Microsoft Build happened on Monday. Tuesday we got Google I.

Wednesday we got Open AIs acquisition of IO, which is hilarious. Which is funny because if you search, I think Lulu called this out, but if you search I.O. on X, you just get a bunch of stuff about Johnny and Sam. Sam, absolute dog on the S. He's an absolute dog, John. Yeah. That's happened a couple times where deep research, deep seek. I believe it was, wasn't it that deep seat came out and then saying? damn dropped deep research, and so the deep keyword kind of came back. But again, this all comes

from deep learning. And so you never know. Anyway, let's just a recap on what actually happened at Google I.O. Because we talked to the Google folks and there was so much going on. It was hard to kind of boil it down into one thing. I was watching The Verge does these YouTube recaps where they cut the entire keynote down. And typically, when they do Apple, it's like five minutes. Apple, WVC in five minutes because it's just like new iPad, new iPhone, new Apple Watch, blah, blah, Microsoft builds cut down was 15 minutes. Google IOS was like 30 minutes because there's so much stuff. So I really think just the volume of announcements was pretty high.

So from Reuters, Google said on Tuesday it would put artificial intelligence into the hands of more web surfers while teasing a $250 a month subscription for its AI power users. That's me, baby. I'm on the 250 plan. Isn't it 500 and you can get it for 250? Yeah, yeah, yeah, yeah. There's a little misinformation here. Yeah. They're going to be ramping that up.

So they're charging me 250 for the first couple months, but I opted into the 500 plan. they're going to ramp me up immediately as soon as it gets the end of the trial. If in the future everybody's paying Google $1,000 a month and just begging, like, please bring back the ads. Like, we're sorry forever. Complaining. Oh, the ad internet was bad.

Oh, I just want to pay. No. I was thinking the other day, I got an email from Google workspace saying, like, we're raising prices. Yeah. And I was kind of laughing because they can,

I think so many businesses are so dependent on Google workspace that they could like kind of overnight raise it. to like $1,000 a month per person and like probably, you know, see a little churn, but not like crazy amount. No way. It's so hard to churn. I mean, this is what we were talking about with deal and rippling. It's like deals in like the most dramatic spot possible.

But do I really want to rip out my payroll software right now? That's a tall order. I'm not on deal. But if you're on deal, the example, it's like, just turn the other eye. You have, you have, you know, you know, Pirelli tires. and Pirelli gets caught spying on Michelin. Are you going to like take your car into the like shop? Yeah, exactly.

Yeah, let me pay. I got to get a bunch of money. No, you're just going to be like, oh, weird. The Pirelli CEO is different than the Bridgestone CEO or whatever. Anyway, there was a flurry of demos, including smart glasses, which I'm excited to try the NREL smart glasses. And then they're also doing some audio-only smart glasses, which have been very popular with meta, the Raybans.

They've adopted a tone of increased urgency since. the rise of generative AI, challenge the tech company's longtime stronghold of organizing and retrieving information on the internet. And this is like, the AI moment should just be Google's moment, right? Their mission is to organize all of the information in the world. And LLMs literally just do that. It's just compressed out all the information. Yeah. It's like the perfect Google product. But of course, they have some business model positioning challenges. They can't just give up the golden goose on day one and go all in on generative AI. But they're taking it more. and more seriously every day.

So, in a major update, the company said consumers across the United States can now switch Google search into AI mode showcased in March as an experiment open to test users. The feature dispenses with the web standard fare in favor of computer generated answers for complicated queries. And what's interesting about that is that who were we talking to yesterday that was saying, like, the genius of Google is that there was just one button. I forget who we were talking to. but basically this idea that like Google had just one button, you know, search and then I'm feeling lucky button.

But it was such a simple interface. Like I don't love the idea of adding a switch. Yeah. It feels like a very, like a half measure. It's like, is AI the future? If it's the future, then Google search should just be AI mode all the time. Or is AI mode not the future? In which case, maybe you shouldn't even do it.

And I should just go to chat GPT when I want AI mode and I should go to Google or I should go to the Gemini app for that. experience if they're distinctive. I think one thing, one thing is for sure, consumers have historically, you know, imagined products for one use case, right?

Like SNAPs, people still use Snapchat for sending pictures to each other. Totally. Even though that functionality is like super embedded in Instagram and other apps now. And so, but yeah, I think it's always tough when you're,

when you're saying, AI is the future, but we don't want to get rid of the thing that's not the future. We like that thing too. So we're going to have our KU2. The scale that Google's at. They can't do it.

It's impossible. They could have a really good LLM ad. No, they could have a really good LM ad product. Yeah. And just if you were starting to introduce it to 1%, 2% of your users, it could dramatically impact revenue because they've spent decades, you know, optimizing. Yeah. Search results.

Yeah. So Google also announced an AI Ultra plan for 250. It provides users with higher limits on AI and early access to expand real tools like Project Mariner, an internet browser extension that can automate keystrokes and mouse clicks. Somehow I missed that entirely. There were so many launches. I completely miss Project Mariner. Excited to try that out because I guess I have access. And Deep Think, a version of its top shelf Gemini model that is more capable of reasoning through complicated tasks. I'm excited to check out that. I've been using deep research a lot. Definitely want to test drive Deep Think.

Pachai told reporters that the rise of generative AI was not at the full. expense of online search. This feels very far from a zero-sum moment, said Pachai. The kind of use cases we are serving in search is dramatically expanding because of AI. First off, and I mean, to be clear, like this is what happened with computers. We went from the desktop to the laptop to the phone and most people have a laptop and a phone. And it was not completely disruptive, even though you can do everything that you can do on a laptop pretty much on a phone. People wound up dual using and so it's totally it's not it's not as much of a dire issue that I think I was making earlier um so ben Thompson's dropping the analysis he says first off my compliments to roiders for this

write-up the amount of things announced and pre-announced or envisioned were pretty overwhelming and i'd say my compliments to ben Thompson for this write-up about reuters write-up because we are now three layers of reaction deep and it's the best um this was to be sure terribly impressive we all knew We already knew that Gemini 2.5 is awesome, and the company's Image and 4 image model and V-O-3 video model were significantly forward from just a few months ago. I wrote in December when V-O-2 came out. I called it off-air, too.

V-O-3 feels significant because it seems inevitable that V-O-4 will be almost indistinguishable from reality. You might have a hallucination here or there, but some of the videos that we've been seeing, like it just needs some slight tweaks and it just looks like. Yeah. It has, it still has a specific look to it.

I haven't really played around too much with trying to really break it out of kind of its default AI vibe. Yeah. I'm sure you can. But it's really, really impressive. And I think I'm, you know, the,

the, the memes are cool. We're in kind of a studio Ghibli moment for V-O-3 videos, I think. We're seeing a lot of these on the timeline. I'm interested to see where it goes more practically. I mean, I noticed that when the Sam Altman and Johnny Ive photo dropped, didn't see a single Ghibli version of that in my feet.

And that was like the most gibliifiable photo ever of Johnny kind of leaning on his side. And I actually... The place that I see Ghibli most now is in people's profile pictures. Yeah, people like that. And I actually ghiblied the, the photo myself, the Sam and

Johnny Ive. And I was, I don't know, if anyone can see that, but I don't know. It's a Ghibli of the photo. And I was just kind of like, this isn't worth posting.

Like there's nothing special here. Like we get it. You can do the cartoon filter. And so it's more about how do you use that tool. And now I've been using it to design. I mean, I'm looking through my recent, my recent,

although on day one of the Ghibli drop, it was a lot of just fun Ghiblies. eventually became, you know, okay, make a storybook for my son. Do the Gigacad meme on everyone who's been on Invest like the best. Make a bunch of ramp ads, of course. Do a bunch of stuff with like testing. Make a bunch of people bodybuilders. Do some set design for us.

It's like a, like my use of images in ChatchipT has morphed from, oh cool, the studio Ghibli thing is the meme to how can I practically use this? to do cool stuff, like making merch ideas for us and stuff like that. So I'm excited for VO3 to kind of get into that mode. It's very clear that it is extremely technically impressive, but it is just a tool. And so we have yet to see the V-O-3 Harry Potter Balenciaga moment, something that's truly viral, not just because it's an impressive that it's an AI image. Like that that joke, the stand-up comedian, that doesn't go viral unless you are like, it's AI, right? And you're like, wow, it's so impressive, it's AI. I can't believe that. That's why I'm liking it.

I'm excited for when someone creates something that would go viral if you had shot it practically, and it's so good that, yeah, they used AI, but it's just as true. The breakthrough moment for me was seeing a like man on the street style interview that was made with VO3 and realizing that I could, I could call it out as AI if I saw it independently if it wasn't being, you know, if the creator wasn't stating that it was made with V03. But you can imagine, but you can imagine, a man on the street interview going viral and creating this like 15 minutes of fame moment for somebody that doesn't exist. I mean, that's what happened with the Pope, the dripped out Pope.

Remember that? That actually went viral. People thought the Pope had this puffy jacket and it was mid-jury image. Yeah, I remember that. And so, yes, it's like the tools here. It's great. The Pope had a white G-wagon one that he would drive around on. Yeah, yeah. And so it wasn't that unbelievable that he would have a white, you know,

puffer jacket coat. Yeah. And so let's go through V-O-2, V-O-3 a little bit more. Ben Thompson has written about this before. He says, the reason to focus on VO2, however, is not just that it strikes me as a seminal moment in generative AI history, but that it also is the most powerful manifestation of the advantages that Google has in this space. Think about the three pillars of generative AI algorithms. Google, of course, invented the transformer, but it certainly seems that likely that the company figured out something important in terms of maintaining coherence, which I complained about with SORA.

And we saw this with SORA when we ran the same test in V-O-3 and SORA. The SORA generation was a lot, there were way more hallucinations, and it was way trippier. Whereas you can see, you know, it's like, it's a Ferrari, it's a yellow Ferrari in our little video,

and it remains a yellow Ferrari the entire time. And it stays, and the TBPN text stays track, onto the car the entire time. And so there's something algorithmically potentially that's happening there. And that's what Ben's highlighting. Then in terms of compute, Google is training and running inference on its own TPUs, which are increasingly developed in conjunction with the model. Jeff Dean said it. I didn't really know that they were, they had TPUs in production. Oh, yeah. These have been in production for years just on general parallelized computing. So you could do

recommendation algorithms or any sort of just machine learning and deep learning on it. for the Transformer Revolution. They knew the TPUs would be valuable and they've been investing in them for a lot. But what's interesting is that the V-O-3 model is very clearly designed and trained and inferenced all on TPUs. And so there might be some sort of compounding advantage there.

The TPUs are so great. Why are you rate limiting somebody paying $500 a month? That's true. They haven't made enough of them. That's probably why. Yeah. Because Apple's probably getting more line time at TSMC. And Jensen's probably getting a lot of line time on TSM.

And then also... I can imagine Jensen gets a little bit of line time for TSM. Yeah. And then also just the fact that, you know, Google has been investing a toning CapEx, but they also have Google Cloud Platform. They also have to serve higher value apps and services.

They have to serve YouTube. They also have to get better at serving me ads. I know, I know. Me, V-O-3 models, that should be their primary goal. If you're a Google, if you're a Google user, put a support ticket in and just say, hey, targeting's been lacking lately. just try to push it, try to get better.

That's my only request. Monetize me, you know, harder. Yeah. So Jeff Dean, one of the top AI scientists at Google, and one of the greatest programmers of all time, are you familiar with Jeff Dean? This guy. So if you search Jeff Dean jokes,

you get this consolidated list of Jeff Dean facts, and they're all Chuck Norris-style jokes about Jeff Dean because he's such a legend. And so it says, compilers don't warn Jeff Dean. Jeff Dean warns compilers. Jeff Dean builds his code before committing it, but only to check for compiler and linker bugs.

Jeff Dean once failed a touring test when he correctly identified the 203rd Fibonacci number in less than a second. There's a whole bunch of these that are just ridiculous. Jeff Dean's keyboard has two keys, one and zero. He just programs in binary. Goated. When Jeff Dean listens to MP3s,

he just cats them to devs. DSP and does the decoding in his head, basically just like Matrix-Men. There's a bunch of these funny thing. Jeff Dean knows the last digit of pie and Jeff Dean mines bitcoins in his head. There's a bunch of other good ones in here. He can lossily compress random data. All these things are like jokes about like computationally impossible things to do. But Jeff Dean can, of course, do them.

So he's a legend. He's pretty good at Leak. Last one, we talked about this yesterday, but data. Data is shaping up to be the biggest obstacle for model development going forward. the single greatest untapped resources likely video generally and YouTube in particular, which Google owns. I certainly imagine this was an important resource for the development of VO2. Yeah.

It's so... It's wild. This advantage is not been talked about enough. There's so many of these sort of video foundation model startups that have raised hundreds of millions of dollars. And yes, I'm sure they get access to YouTube data content. in different ways. Yep.

But ultimately it is nobody has, nobody has an edge that you could go do deals with every single major Hollywood studio and still not have anywhere near the edge as Google does just by having a constant net news stream of video content. Anyways. We got to look at the total data size of all YouTube videos. compared to just the text on the internet. I want to really know how much more data is there on YouTube or is it roughly the same.

We're going to dig into that soon. But anyway, Ben Thompson continues and says, Google I-O is this on steroids? But at the same time, I'm not sure, and I'm not sure if this was because of the sheer quantity of an announcement, I came away feeling less impressed than I should have. I decided I needed to take the dog for a walk.

I wanted, by the way, to tie these videos together into something more coherent. They're both made your using VO3 and Google Flow. The new product Google announced to utilize VO. Unfortunately, it's extremely hard to use and very buggy above and beyond the general challenges of linking together, disparately generated videos while retaining coherence. This in the end makes the point I ultimately came out, I came to on my walk. Google Flow was arguably the most archetypal, archetypical product announcement at Google I.O. But while VO.3 is jaw dropping, flow was pretty disappointing. And this was my experience. I went to flow on my phone and it basically told me no. It was like, you're on an iPhone, you're a second class of saying,

get out of here. We're optimized for desktop. And it's like for a startup, I get like Instagram started on iPhone and then eventually launched on Android. I get that. But it's like you're Google. You can figure out how to serve your web app in Safari on iOS. Like I, I, I, I, I, I, I, I, I, don't understand what's going on. It just makes you look bad. It doesn't really make me like Google so much that I'm like, oh yeah, I should get an Android phone now because like clearly it would be a better experience. Yeah, I've got to use flow, this app that I've never tried.

No, if that's the development paradigm, I'm out. If you're going to try and twist my arm into using your software, I'm less likely to use it. That's my take on it. And then they also funneling you into this like, hey, do you just want to watch a bunch of videos? Like, you definitely don't want to set the GPUs on fire by training one of these things. You don't want a custom video.

Why don't you just have one of these that we have off the shelf? I love that. I love that content or the copy there too. It's flow is a tool built for and by creatives. And it's like, well, YouTube in many ways destroyed the film industry, right? Did it? I would argue that it did just because it created an entirely new library of content that didn't require going to the movies, right? Yeah.

I mean, there's like the videos would have come on demand, but it's interesting. massive massive competition for people's attention and that just hurts the movie industry yeah I wonder about that that's probably right that seems reasonable I don't know yeah anyway anything about that more if YouTube is the is the reason that I mean not YouTube in particular but yeah just just the internet yeah certainly like like easier technology like obviously there's gonna be demand for video in a theater if you can only watch a video in a theater. As soon as you can watch on a TV, that creates competitive pressure.

As soon as you can watch it on your phone, that creates more competitive pressure. YouTube's a part of that. Probably, like, I'd make it like 10% of the reason. But yeah, but yeah, totally valuable, totally viable argument to make. So, Ben Thompson was a little underwhelmed. He said more generally, the generative AI pillars that I wrote about last December, the areas that Google dominates, notably don't include any genera, AI products. And that more than anything was my Google I.O. takeaway. Google's technology is

incredible, but I'm still not convinced this company can make compelling products. Indeed, my critique of the presentation as a whole was how unfocused and random it felt. Google's inability to prioritize and set a coherent narrative actually diminished just how incredible their advances are. Yeah, I mean, it certainly was an opportunity to come out and just focus on, yeah, we have a bunch of cool experiments running, but like here's how we're going to actually have a ultra-coherent strategy around the search bar and how that is going to interact with AI and what are our long-term plans. They almost should do more keynotes less frequently and just say it's notebook LM day and we're just going to take over the internet with notebook LM. And then it's V-O-3 day and we're making V-O-3

a fantastic product. Same advice we would give to founders that are like, I'm going to have a big announcement, I'm going to drop four things. We hired this executive. We had a new website. We have a new product release and we have a fundraise. I'm still so puzzled by the V-O-3 launch. Just because like, you're Google. You have to figure out how much it cost to generate one of these. It's an eight-second video. I understand that you're setting the GPUs on fire. But what is the energy cost? What is the depreciation cost on these TPUs? They must know how much it cost to generate a single V-O-3, eight-second video. Is it $10? Is it $100? Like, there should be an auction-based model, essentially,

like what they do for ads, and then they should have a queue. And they should say, hey, right now, the GPUs are fully utilized. We have no capacity. And so it's going to be $10 for this next generation, and you're going to have to wait one hour.

But if you pay $100, you can get it in the next minute, or five minutes. I think there's such a better experience to actually meet the demand where it is instead of like, you're on a $250 plan, but you can only use it a few times a day,

and then you wait 24 hours. But part of it is it's basically a research preview. I don't think a lot of VO3 content right now is actually going to be used. That they're advertising on the iOS App Store. Yeah. I don't, I don't care that they call it a research preview at Alpha, Beta, Gamma, Delta. Who cares? They're selling an expensive toy right now.

Yeah. It's like you're, it's like a buggy toy that only works. You're taking my money. And I think the biggest thing is the ads. Like the fact that if I go to the app store, store and I search for Gemini, like it pops up and says, create videos with Gemini, make and share videos of Gemini advanced with V-O-2. And like, you click there and it's like, it's advertising.

Like, if it's a research preview and you don't have the capacity, don't advertise it. Yeah. Say, hey, we're at capacity. We're maxed out on demand. Anyway, going back to search, Ben Thompson continues and says, the one clear exception was a smack dab in the middle of the keynote, which focused on search. I don't think it's an accident that Google's best and most important product had the clearest presentation and vision.

CEO, Sundar Pichai, started with a brief overview of AI. Brief overview of AI overviews. He says, so this is Sundar talking. It's another exciting example of how AI is advancing our timeless mission to organize the world's information. Yes, let's hear it for organizing the world's information. We love that Google does that, to make it universally accessible and useful. no product embodies our mission more than Google search.

It's the reason we started inventing AI decades ago and how we can deliver its benefits to benefit scale. Just to be clear, we invented AI. Yeah, they did. Now make a product. Slipping that in. Our Gemini models. It is funny, though, because talking about these AI overviews,

I just searched in Google, what did Google announce at I.O. And it doesn't even have an AI overview. of that. The self-referential stuff is particularly bad. And I think it's because of like data and security like Oroboros is. Like the fact that I can open up Gemini and I can ask, hey, just, I just want to know like, am I paying for this on this account right now? And it's like, I have no idea. I wouldn't possibly look at your own account within Google. It's like, I understand privacy. But like, this is actually important to me. I want to know if I can access the latest model.

You can even ask a Google model, what model are you? And they can never tell you because they're just like, couldn't possibly do that. It would be a violation of something or other. And you know it's just like some sort of security thing that got wrapped up in like 20 layers of lawyers and stuff. And they were like, don't leak the, don't leak the customers information into the LLM to let them know that they are paying or they're not. Like, the LLM should be selling me on upgrading, you know, if I'm not upgraded.

This is interesting. I just searched for stuff about open AI and anthropic. It would not give an AI overview. But then I searched about TBPN and it gives an AI overview. Yeah. And it misspells my name. Oh, weird. Which is interesting.

Adds the E. Ads an E. Never done that. Yeah. Four letter name. Tracking in the lineage of Apple CEOs. You got Steve Jobs, Tim Cook, Jordy Hayes, potential. I'm starting my run.

Starting your run. One great example of progress is our AI overview since launching at IO last year. They've scaled to over 1.5 billion users every month. Let's hear it for that. And over 200 countries and territories. As people use AI overviews, we see they are happier with their results and they search more often. I completely agree. The AI overviews have been very good.

I know that there's some hallucinations, but in general, I've enjoyed them. The beautiful thing from an ad product standpoint is that Google still gets all their ad. They're still serving ads everywhere. it just so happens that they're not, you know, especially on desktop. Yeah. I know, I'm very, very satisfied with the iteration there. The question is just like, they are pulling me into. It is funny to think about an opening eye experience where you query chat GPT and you get just like

exactly what you're looking for. And there's just like display ads like everywhere. No, I know it's not going to happen, but that's effectively what Google is showing is possible when you can just build ads around the result that the person actually wants. AI overviews are also one of the strongest drivers of growth for visual searches in Google Lens. Lens grew 65% year over year

with more than $100 billion already this year. So people are asking more queries and they're also asking more complex queries with our latest Gemini models, our AI overviews, are at the quality and accuracy you've come to expect from search and are the fastest in the industry. AI overviews are, as Ben Thompson pointed out, after Google's most recent earnings, the most used generative AI product in the world, most of today's search segment, however,

was focused on the new AI mode. AI mode was introduced by head of search Liz Reed. So, Liz Reed says, today you will see how you can ask anything and more intelligent, agentic, and personalized search will take on your toughest questions and help you get stuff done. This is particularly funny because did you know that there's a maximum query length in Google search. It's like 256 characters or something. It's like shorter than a tweet. So if you just if you just try and benchmark like what was your latest ask for chat GPT? Just like, hey chat GPT, summarize the news for me. I'm interested in seeing what Ben Thompson's

talking about. And I want to know what Dylan Patel's up to. And can you also summarize what happened at Microsoft Build and then Google I.O. And I want to hear about the Johnny Ive acquisition. And then I also want to hear some stuff about Claude 4. If you take that if you take that query, put it into Google, you will literally just get an error message that says query too long. And it's because Google Search is set up not for that type of searching. So anyway, there's Gemini 2.5 at the core, and they give a tour of AI mode. And so AI mode is more like a chat of a chat-like interface and all the demos are very impressive.

They are also real, as in either available now or shipping soon and all seem clearly useful, says Ben Thompson. On-demand data visualization, for example, seems like an obvious win for generative AI-powered search. ton of sense. Search for GDP per capita across a couple countries. It just generates a table or chart for you. It makes a ton of sense. The most interesting comment read made, however, is a paragraph in the middle. The real goal for AI mode is to refine these features so that they can graduate to the core search experience completely, very, very smart. In other words, Google has a clearly defined AI funnel, which actually brings the coherence to this keynote that I was looking for at the foundation

of everything are those pillars I referenced above Google's incredible models, unmatched infrastructure, and data advantage, reads discussion of personalizing AI mode. Using your personal data was compelling as well. Then you have a bunch of theoretical ideas that make for cool demo videos, but the actual productization that still functions in Google is on the search team. That team takes ideas that works and puts them in AI mode. So you can think of like every other product is just the farm team.

Only if you're looking at Google's. alphabet stock and you're trying to understand how AI is going to impact it. You can ignore Notebook L.L and the Digi knockoff and all these other things and just focus on this. And it's like, do you believe they're going to figure out a way to monetize these sort of chat GPT style queries that are happening in Gemini right now? And it probably looks something like AI overview over time. Yeah. And also, if you're a startup and you're worried about like getting rolled, by Google and becoming like a Google feature. Well,

the threat is not that Google launches a directly competitive app and out competes you. That hasn't played out. Like they launched Google circles and they weren't able to crush Facebook or anything like that. Or, or, or, or, like, yeah,

any of these little, these little, like, startup killers where they're Google products. That's not really the dangerous area. The dangerous area is will your customers and, your users be able to solve the problem that they want just by Google Search 2.0. Or like when Google Search gets better, is that a threat to your business?

Yeah. Or if you're building an agent that's like, we're helping people book travel. It's like Google flights is going to get better. It's going to get better. And so that's the real risk. I think it's less that if you, if it may, if it never makes sense to go. Google also had a Twitter competitor. Google Buzz.

Wait, no, that was a chat thing, wasn't it? it was it was a Twitter competitor yeah so like all all these all these one-off things that don't make sense it's a microblogging messaging tool developed by Google no way it replaced Google wave oh my god so many of these things yeah they launched in sunset so many products um anyway AI mode looks a bit like chat GPT the real race for Google is to make search compelling enough that people don't just switch to chat GPT for everything to that end another metric that likely matters is actual usage pechize definitely of the types of searches that are growing was very carefully delineated, but growth in general

is certainly a good thing. If an AI feature, if a feature drives AI mode, it may be a good candidate for the main search page. So one of the common refrains that came out of yesterday's keynote was how Google had just killed half of Silicon Valley startups. I'm not so sure. However, we've read enough Satechry at this point that we're predicting what he's going to say. First, it's clear than ever to me that the only product that truly matters or functions at Google is search or functions. A lot of the demos we saw yesterday are likely to stay demos or at best be forgotten and eventually killed products that no one uses. Second, the degree to which so many of the demos yesterday depend on user volition actually kind of dampened. What's that?

I'm getting the phone call? Google's calling us. Sundar, let's just bring them in the studio. Let's just talk to Sundar. Bring him in. What's going on with the XR glasses? We don't have Sundar. It would be a fun, surprise guest.

We'll get them soon. We'll get them soon. Android is probably going to be the most important canvas for shipping a lot of these capabilities, and Google's XR glasses were pretty compelling. And in my opinion, how do you XX much closer to what I envisioned for XR than Metas Orion did? Interesting. Devices drive usage at scale, but that actually, leaves a lot of room for startups to build software products that incorporate AI to solve problems

that people didn't know they had. The challenge will be in reaching them, which is to say the startup problem is the same as ever. You got to get distribution. Here's the thing to me. It's like, does Google search and Gemini eventually merge or do they stand as independent products? And so Google's like kind of like maintaining search for people that basically boomers who are just perpetually going to go to the search box and search. And like, I'm going to be in that bucket to some degree. I think everybody will still have. Yeah, for sure. It's so built in. They have so much distribution that it will continue to have usage. Meanwhile, though, chatGBT is just getting better and better at searching the internet. And it can be used for both things. And that will present a pretty

interesting challenge if Google's like, okay, we're competing with open AI and chat GPT, but we're doing it in this two-pronged way and open AI is like we're just trying to make the best like you know basically like single uh you know uh search you know box plus button on the internet that can do everything yep i mean i think a lot of the decision between those two uh will come down to speed so right now if i want uh let's see like tom cruise age tom cruise age. I just hit Control T and I'm in the search bar in Google Chrome 62 years. Now if I go to chat GPT, I have to go chat.com. I'm logged in Tom Cruise age and it's thinking, searching the web. It's searching the web. It's searched three different websites. Tom Cruise was born on July 3rd,

62. As of today, he's 62 years old. So you see how much longer that took me to use chat GPA for something like silly like that or simple. But there's a lot of things that you search for that where you're like, I know that Google will one shot this in five milliseconds. And I know that chat GPT will take too long. Also, for some reason, I wound up running that against GPT 4.5. I think 4.0 is probably faster. But like when I went to chat.com, 4.5 was just the one that was defaulted, right?

And so this is where we get into like the model router where if I land on, if I accidentally triggered that in 03. it would be writing 100 lines of code, you know? And so the model router needs to be like super fast and they need to figure out caching and they also need to figure out integration into like the search bar because even even on your phone, it's usually faster to just open up a new safari window. And then the question is like, does open AI need its own browser? Yeah.

Right. Even, you know, Arvind from perplexity is like very focused on the browser. Like right now, even if I do this on my phone, if I open up Safari, the search bar is always down at the bottom. And I can just tap that and immediately start typing Tom Cruise's age. When I open up chat GPT, I was looking at my image library. So now it's, I'm in image library mode. I have to go back. I have to go back to chat GPT.

I have a query that I didn't actually click finish on. So I have to delete all that to go. So all of a sudden I'm like into five, 10 seconds just to get a quick answer. And so I still route queries personally to Google when I want something super fast. Yeah. Anyway. Do you remember that Gemini started as Bard? Oh, yeah.

There's been like seven. Barred. But there were other ones before that. Palm. Bard, a poet traditionally one reciting epics and associated with a particular oral tradition. Yeah. Cool name. I have a few other examples.

Tough go for the winkle line. Yeah. Zuck comes for them with the Facebook stuff. Now Google's coming for them with the Gemini branding. If Gemini becomes synonymous with Google AI. And people are like, wait, where do I go to trade crypto? I never thought about that. It's Matt.

It is. I don't know how you could do that because they're both in technology companies. Like they're both tech companies. I get there in somewhat different areas. But like we talked to Austin Allred. He had to rebrand Lambda School because of Lambda Labs, which is a AI data center company. Oh,

is because of Lambda Labs? Yeah, who's coming on the show on Friday, actually. Nice. But I'm... Trademark Wars. New site on TBPN. I think it's Lambda. But I'm not sure.

Let's go through a quick. It was a different Lambda, which was a completely, it was not an education, but it was a tech company, and they were both tech companies, and so they had to rebate. Yeah.

Let's go through a quick history. I have a history of Google cloning startups. So way back in the day that in 2004, they launched something called Orkut. Orkut. Oh, yeah. Orcochchch. It was a Brazilian acquisition.

Yeah, so they got traction in Brazil and India. It was never truly global. Yep. Shut down in 2014. There was lively, which was basically a second life, like 3D virtual world thing. That's cool. Closed in less than six months. And then they had Buzz, which was trying to counter Twitter at the time.

Had a bunch of backlash around privacy stuff. Google Plus was obviously the Facebook rival. And then they actually had. had tried, they tried to buy Groupon for $6 billion, and it failed. And so they launched Google offers. This is like a whole other era. I have a fun story about Google Plus. So I was one of like the best, the biggest users on Google Plus.

I just went like full sending to it and was like, I'm going to figure out the social network, kind of like, you know, figure out how to create circles, create that. And because of that, I got. Really dating yourself. Yeah. I got, this is in college. I got verification. So I got a checkmark.

Wow. But that checkmark translated, they ported it, even though they shut down that, my account was like checkmarked as like, oh, you're like a serious influencer, even though I had that 500. Did that transition into YouTube? Yeah. And so when I started the YouTube channel, I got to 100 followers on one, on a new account. And I was like, wait, my other account has a checkmark.

So I shut down the old account and moved it over to my personal account and had the checkmark, which I could have gotten in like, it didn't actually do anything. for the growth of the channel, but it was just like kind of a funny, funny anecdote. I was like, I'm going to try this. And it was just a funny story.

Anyway, let's move on to OpenAI acquiring IO and Ben Thompson's analysis there. But first, let's tell you about Figma. Think bigger, build faster. Figma helps design and development teams

build great products together. You can get started for free folks. Go to figma.com, try their new products, websites, product. Try Buzz, too. Tons.

Oh, I mean, the trademarks available now that Google's out of the buzz game. The buzz game. Figma's got it. Create and share on-brand assets instantly. Very good. Powerful. Also, go check out vanta.com, automate compliance and manage risk, prove trust continuously.

Vanta's trust management platform takes the manual work out of your security and compliance process and replaces it with continuous automation, whether you're pursuing your first framework or managing a complex program. Anyway, let's go on to I.O. Let's do it. We already covered the Bloomberg article. we know the facts, $6.5 billion, all-stock deal. Johnny Ivan, his team over at I.O., get roughly 2% of OpenAI as major incentive pay.

And there was a lot of good commentary on the timeline about this, about how Apple's not really set up to comp someone a billion dollars. We looked at this with Tim Cook. He's making, like, what, couple hundred million? And it's like, when you're talking about a trillion-dollar company, executive compensation in the billions isn't actually that crazy, but opening I can offer that. John, yes. You're going to be devastated, but Tim only made 74 million. Oh my God. That's brutal. It's like he should have been a baseball player. I mean, he, he should have been a half decent basketball player. What do you think would have happened to Apple's market cap, had they not dodged the tariffs? That would have been hundreds of

billions of dollars in market cap erosion. It was already tumultuous for the company, but they were carved out the entire time. There was never any fear that it was going to happen. And that's entirely because of Tim Cook's negotiation. 2023 made 63 million. I genuinely feel like we need to take to the streets and protesting in Cooper Tino. We should. Because we've seen this on public. You can back test. The better that CEOs get paid. Typically, the better the companies do. And as an Apple shareholder, I want Tim Cook to be pulling down one B a year, and I think he'll be highly incentivized to grow the company. Okay, I have an idea. So Greg Abel, he's 62 years old.

He just took over as CEO of Warren Buffett's Berkshire Hathaway. Tim Cook's 64, only two years older. So what if Tim Cook steps down as CEO of Apple, starts a company, gets acquired by Obey? A.I. Boom. Goes in. 30 year run. 30 year run.

Generational run. And he's like, look, it's just dollars and cents to me. I can make more money over there. I can 10x my comp. He could 10x his comp by going to Open AI. Yeah, he really could. Easily. Easily.

I mean, yeah, that's the big question right with Johnny, you know, joining Open A. Six billion. It's like, who's the supply chain lord who's going to just really bring this thing into? If Tim Cook works, I mean, he's making what, 60? a year. He's not even cracking 75 mile a year. So he would have to work for 100 years to make what Johnny I've just made. I don't think he has another hundred years in him. 30. I would give him 30 solid years ahead of him in his career. That's not enough to get to $6 billion. He could maybe pull $2 billion. Well, to be clear, I think Johnny had investors and in I... It's all in Johnny's pocket, in my opinion. Yeah, he deserves it all. It was funny. I was looking yesterday. Johnny's like celebrity net worth tracker or whatever. They updated it in real time and we're like, he's a billionaire now.

Yeah. And he wasn't just from working at Apple. It's brutal. He had barely had 700 according to. It's a travesty. It's terrible. Anyway, let's go to Ben Thompson's analysis. He says they put out an open eye put out this nine minute and 21 second video of them

complimenting each other, expressing love for San Francisco and announcing the partnership. Ben Thompson says, it's a bit much, to be honest, although it beautifully shot and produced, I assume it was put together by Love From. Ives, not included in the deal, but not accepting new clients' design agency. I think the key segment from Ives' perspective is this. Sam Allman said, we both had a very strong shared vision. We maybe didn't know exactly where this was going to go, but the direction of the force vector felt clear. Then this deeply shared sense of values about what technology should be, when technology has been really good, when it's gone wrong. Johnny Ives says, I mean, that was in a way one of the basis, I think, of one of the reasons

Sam and I clicked was despite our wonderfully different journeys to this point, our motivations and values are completely the same. In my experience, if you're trying to have a sense of where you are going to end up, you shouldn't look at the technology. You should look at the people who are making the decisions and you should look at what drives, motivates and look at values. I like that. That's a good take. I've expressed what drives and motivates him earlier this month in this excellent interview with Patrick Collison.

And it was interesting because Johnny I, we haven't really heard from him in a few years. And then boom, massive podcast with Patrick Collison, just days before the acquisition. Yeah, on a media tour. So I'll read, I'll read Patrick's question. Sure. You're talking a lot about the purpose of design. That's not how he sounds. Okay, I'm not going to, I'm not going to do the accent, John.

Even though we both have Irish roots. I'm not going to, I'm not going to do that. Just do a vanilla reading. Fine. Now I'm, no, no, do it. Now it'll sound like you're doing Patrick Collison doing a Jordy Hayes impression, but that's fine. Yeah. Okay.

So I'm doing Patrick Collison, impersonating a technology podcaster. Yeah, that's great. You're talking a lot about the purpose of design and the effect that design has on the recipient, on the user, on the consumer, whatever the case is. There's widespread concern and speculation about the effects of smartphones and the internet.

It doesn't necessarily accord with just the smartphone, but on some of these products on attention spans and whether it has some adverse effect on kids or teens or who knows, maybe all of us, maybe the adults as well. There's questions over with AI, whether it changes how education works, and cheating in school, all of these technologies that we create have this potential double-sidedness to them. And so I guess as somebody who clearly takes seriously and thinks seriously about the full

effects, how do you think about possible harms? And Johnny Ive says, yeah, I think there is, probably not, yes, in the British, Irish accent. Yeah. No, this is Johnny I of doing a John Coogan impression. Okay. I'm excited for this. Yeah, I think probably not anything that I can be more preoccupied or bothered than by what you just described. I think when you're innovating, of course, there will be unintended consequences. You hope that the majority will be pleasant surprises, certain products that have, that I've been very, very involved with. I think there were some unintended consequences that were far from pleasant. My issue is that even though there was no intention. You don't think candy crush is pleasant? I think candy crush whales or?

You don't think sports betting on your phone at the drop of the hat is. That's 24-7. I think there still needs to be a responsibility, and that weighs on me, as you know, heavily. And yeah, I mean, obviously there are negative consequences, but I think overall technology is I think he might be alluding to when I made it possible to Angel invest on a mobile device. You're, he's literally, I think he's sort of sub-tweeting me. This party around basically took all these people in San Francisco who were hopelessly addicted to angel investing and made it accessible to them at all hours of the day. You didn't even have to send a wire transfer.

you just got an invite and it was like Venmo. Yeah, ridiculous. It seems pretty clear that Ive is talking about the iPhone, says Ben Thompson, which is to say it sure seems like he is motivated not to simply build an AI device, but to actually diminish the iPhone's dominance in the user's life or even in the long run

kill it completely with tracks with the Wall Street Journal's reporting on their plans. They've been working on a device that will move consumers beyond screens according to people familiar with the matter. I mean, it feels like her, right? I mean, Sam's posted about her. They're very all in on voice interfaces.

This could be a her device, right? Is that the most obvious thing that they would do? Such an interesting dynamic where it feels like sci-fi is dictating the product roadmaps of so many companies. Always has been. Always has been. I mean, Steve Jobs was always looking at like the Star Trek tablet and being like, let's make an iPad.

There's nothing new. You got to get ideas from somewhere. Jason Carmen. Yeah. Jason Carmen needs to make something. Save us. Understand that you're basically defining the product

roadmaps of the biggest companies in the world. That's true. That's admittedly pretty thin gruel, says Ben Thompson, but it's in line with other rumors and more importantly from my perspective,

tracks with Ives expressed state sentiment. And it should be noted that this sentiment is a sentiment that Altman has expressed as well. He's tweeted a couple of times people sacrifice actual happiness and actual accomplishment for for short-term dopamine hits by posting and chasing likes slash RTs on fb slash Twitter back from 2013 only 21 likes back

then sam was struggling to post bangers I think everything he posts now gets like 10,000 likes and then a year after opening i was founded sam album and said digital addiction is going to be one of the great mental health crises all our time. I like my screen time, John. Yeah, I like it too. I do email on my phone. Yeah. It's productive.

In 2017, Altman wrote on his blog. It is funny because I do believe that this sort of AI companions have the potential to be some of the most addictive products in history. Totally. And I believe that in general, people will continue to interact with them. through screens, devices, et cetera. Same Alman in 2017, almost a decade ago now, wrote, I believe attention hacking is going to be the sugar epidemic of this generation.

Ooh, sugar underrated though, right? Sugar is potentially good. Potentially good. I think it's, I've called it. It's a superfood. It's this amazing substance that gives you clean energy almost immediately. So maybe that's actually a out of these seed oils. Maybe that's actually like a broader metaphor for attention hacking is that it feels good in the, it is bad at the extreme, but it is good in the right dosage.

Yeah. Because pure sugar, when taken in athletic context, when you're working out at the right dosage with the right purities, can be good. Yeah. And it's not something that should just be completely removed from the diet, much like attention to technology. I can feel the changes in my own life. I can still wistfully remember when I had an attention span. My friends, young children don't even know that's something they should miss. I am angry and unhappy more often, but I channel it into productive change less often,

instead chasing the dual dopamine hits like Likes and outrage. That's interesting. I feel like our generation's kids are doing fine. Like because we're aware of like don't do the iPad thing, I'll see my son spend hours building Legos. and like painting and drawing and spending. I think it helps that... I think it helps that adults are generally hyper-addicted to their devices

and want to prolong the time in which their children can live a life of freedom from that addiction. It's like at some point I'm sure my son will try a cigarette, but just because heaters are Lindy, but I'm going to try to delay that. for a very, very long time, right? He's just 65 being like my first. It's always challenging with Altman to know whether when or if he is talking his book, says Ben Thompson, as it were.

But there is a record of long-running sentiments that might convince Ive that their motivations and values are completely the same. At the same time, there are certainly other motivations as well. Ive, for his part, just became a lot richer, at least on paper. Altman may or may not have diluted the nonprofit share of Open AI by making a big purchased with stock. This is a kind of a theory that was going around about Reddit and how Reddit kind of went into Conday Nast and then was pulled out through a bunch of different stock grants. Sam and the original founders seemed like they realized like Reddit was never going to be what it

could be while a part of this legacy media company. Yeah. When you read the story of Reddit, it reads this very like, almost like conspiratorial. It's this like really like 4D chess maneuver to get it out. And yet at the same time, like I'm like, yeah, I'm glad Reddit's not. not owned by Conday Nast. Like you guys did everyone a service. Like, good job Alexis O'Hanian and Sam for pulling that company out, which was actually an innovative technology company from like a dying

like the media company. Anyway, Altman may or may not have diluted the share. He's also sliding into the role occupied by his childhood idol, won Steve Jobs. So what's the strategic positioning? We got four minutes until our first guest. The larger question is, is what this means for Open AI. On one hand, the angle here is obvious and fortuitously articulated

by me just yesterday, says Ben Thompson. A little back that. It has long been the case that the best way to bring products to the consumer market is via devices, and that seems true than ever. Android is probably going to be the best, the most important canvas for shipping a lot of Google's new capabilities. What makes ChatGPT so remarkable? And OpenAI, the accidental consumer tech company, is that like Google two and a half decades ago, they have reached nearly a billion users almost entirely through word of mouth. The only distribution deal Apple has done, or opening I has done is with Apple. More on this in a moment.

This is exceptionally rare, and my longstanding opinion is that the obvious way to capitalize on this position is to roll out advertising. Let's hear for advertising. We love advertising. Speaking of advertising, go to linear. Linear is a purpose-built tool for planning and building products. Meet the system for modern software development, streamline issues, projects, and product motor maps.

MCP server. right? Yeah, they have a dedicated agent's product now that they rolled out yesterday and Scott talked about it. Scott Wu from Cognition talked about it on the show and highly recommend going to check it out if you're wanting to actually unlock the power of agents in a systematic way. So Ben goes on to say it is worth noting, however, that one challenge is trying to win in advertising is that you are in the end fighting on Google's turf, the same Google that demonstrated its Incredible model and infrastructure prowess. Well, this is awkward and surely couldn't have been intentional or was it.

At I.O. Yes, there are questions about Google's product chops, and I already think that OpenAI is the best in AI at product. And we talked to Kevin Weill, who's doing product at Open AI. Clearly incredible team over there, and the products are fantastic. And they just added Johnny I. And yes, Open AI has the advantage of building. building an advertising business from scratch, giving them more degrees of freedom than Google,

which has to worry about cannibalization and disruption. But Google is also moving AI into search, and the infrastructure the company has built out around advertising will take years and years to match. You can say the same thing about meta. And speaking about meta, think about how CEO Mark Zuckerberg has framed the company's approach to AR and VR. This is Zuckerberg in a 2022 strategyary interview about their partnership with Microsoft. But taking a step back, I think in addition to this strategic alignment. I also think there's a very deep philosophical, this is a Zuck, deep philosophical alignment on the direction that we want the next generation of computing to go. My brief take on the history here is that every major computing platform there has been an open ecosystem that's

focused on partnership and a more closed and integrated ecosystem. So with PC, Windows was the leading open ecosystem and of course Mac was the leading closed one on phones. It was Android and iOS. I think one of the things that's interesting in the history is that I don't think it's predetermined whether the open ecosystem or the closed one ends up being the primary ecosystem. And so he was pushing for an open metaverse. Implicit in Zuckerberg's answer is the assumption that the company they are competing with in the long run is Apple, the closed option. Another way to frame that contrast, however, is between modular Windows and Android and

integrated Mac and iPhone. The integrated option, the Apple option is monetized first and foremost through device sales. The modular option is modular in part because the underlying motivation of the operating system is to maximize reach, which in the case of Google means increasing the number of services on which to control search and serve ads. These framings are, in reality, a bit too limiting to fully articulate the various strategic puts and takes. Google, for example, is deeply integrating Google Android with Google AI in the cloud,

while continuing to lean on third parties to actually build the hardware. Indeed, one of my chief critiques of Apple's approach to AI is that by trying to do AI themselves, they are trying to compete with Google on Google's turf. still Google is as Google does focusing on being on as many devices as possible we should we should close out here because we have our next guest joining us in just a minute but he says Zuckerberg assumes that Apple will be there in AR and VR should Altman assume they will be there in AI to put it another way if you're going to take on one tech giant directly and you're an AI company isn't Apple a much more attractive target than Google and if you think that that way should your focus be on building a

horizontal service that runs everywhere or on building a fully integrated offering that monetizes the high end through an experience that integrates hardware and software. And if you want to build the best possible integrated hardware and software experience, the team I.O. is a veritable all-star collection of Apple talent that extends far beyond IVE that built the last one is a reasonable place to start. Very interesting. Gets everybody thinking that he's trying to kill Google and then goes for the jugular on Apple. It's very interesting. I think Sam's going after everyone. I think he wants it all.

And why not just to keep everything in energy, in servers, in chips? Well, we have Dan Shipper. Welcome to the stream. How you doing, Dan? How's it going? Excited to be here. I'm fancy of you guys. Great to have you on.

I've been wanting you to do this for a while. And you've had a busy week. You have some personal news at every. You're also at your at Anthropics event right now. Is that correct? Matt, code with Claude's event where they just dropped Claude Opus 4. This is amazing because we didn't get a chance to talk about Anthropic or Claude 4

on the show yet, and we have a perfect guest to talk about it. I am here a correspondent. Thank you. Can you kick us up with just a little bit of an introduction on you for any of the fans who are watching who might not be familiar? And then we'll go into what the events... They'd have to be living under a cluster down. They'd have to be living under a data center, but some people do.

My name is Dan Shipper. I'm the co-founder and CEO of every, at every published ideas and applications at the frontier of AI. So we're a daily newsletter. We write long-form essays about what's going on in AI. And then we also can create software products. We bundle all together and sell it to the audience. Very cool. Yeah.

And it, uh, Jordi was describing you as like a Claude Stan. You've been a fan of Claude for a long time. We've talked to some people in AI and they've said like, you can just walk into a party and immediately clock like, oh, that's a chat. user oh that's a Claude user. Is that true? Are you beating the allegations? I would say it's interesting. I would say I used to be. I think Claude like 3-5 Sonnet was the first model where you're like, oh my God, this thing actually gets me. But since 03 came out, I would say 95% of my AI usage is in chat GDP. And also with their memory feature, like that's these memories. It's incredible. It's very sicky.

It's tough to have a friend with no working memory. It's like, you know, yeah. Yeah, I mean, like concretize this a little bit more for me. My experience with AI and LLMs is that I'm using them as as information retrieval, agentic reporting, deep research a ton. I never just sit there and have a conversation. So if it's if it's answering a little tersely, if it's dropping boldly, it points on me, I'm kind of fine with that. But when I talk to some people about the differences

between Claude and Open AI, a lot of it is in the subtleties of the language. And I just feel like as a consumer, maybe I care about that less. But do you think that's important? Does it matter to you? Am I missing out on something by not seeing? I'm just like a very sensitive emotional guy. And so I want it to be nice to me, you know? But I also think to some degree, even for like more business, like straight ahead business use cases is actually really helpful for it to have a little bit of that sensitivity. For example, something I use it for a lot is if I'm having a management problem, we report almost everything that we do at every in granola and it's really easy to take your transcript of a meeting and be like, what happened here? Like, why did why was our fight here like or you know, an employee's

having a problem like how do I help you know that kind of thing? And it sort of has a little bit of that radar that can that can really help you move through a lot of interpersonal situations that come up off. Okay. So that's interesting. Well, before we just go and, and just, uh, into a much longer discussion and, and talk about Anthropics news, why don't you talk about, there's a couple of things this week. You had your own fundraise and it was a unique structure. So why don't we kind of start there? Yeah. We announced this week that, uh, we raised a little bit more money. We've, we've previously had raised about 700K in 2020. Uh, and then we raised, uh, just now we announced a $2 million raise, uh, led by Reed Hoffman. Um, but we did it in a weird structure that I'm,

like sort of tongue and cheek calling a SIF seed round. Basically, like Reed and starting line DC, who's another one of the investors, they committed up to $2 million, but we can pull it down when we need it. So we have not pulled down all of it. And I think that's a really nice structure for us because we're going to be a company.

We also have a lot of software attached, but I want to maintain our ability to be weird and just like have a video, it's creative playground when we make little stuff. And so this gives us enough money to experiment, but not so much that we're like locked into a particular growth path. Yeah, are you thinking that, I mean, we saw this with Open AI. They were a nonprofit, and then they wound up creating like the most dominant consumer app of all time. They needed to spin that out. Like the parallel here is that you are building software tools, but you also are doing media. If one of those software tools takes off, all of a sudden, does the media become a marketing engine for that tool? Does the tool spin out? Are you thinking about it like incubation where there's separate cap table, maybe investors ride along? Like, what were those discussions? like and how are you thinking about that?

Well, didn't you guys already spin out? Oh, yeah. We did. We spun out Lex, which my every co-founder, Nathan, now runs to CEO of that. And we raised a seed round from True Begglers for Lex. Oh, cool. Almost all of the incubations that we have are actually their own separate LLCs. I think, like, in general, I fucking love having a writing business.

And I want to keep the main thing, the main thing. And I think that we can build a really incredible universe of apps. and other offerings around the writing business. I do think sometimes we will hopefully run into opportunities that seem really massive and that can be like just really big independent businesses on their own. And yeah, we have, if that happens, we have the ability to go and actually like further spend them out,

turn them into C-Corps, have them to go raise money, all that kind of stuff. So we have the optionality to do both. Cool. Awesome. All right, well, I mean, there's a bunch of stuff to talk about. Why don't we, why don't we, as our official Anthropic correspondent, which is the title that we're giving you, obviously, not related to the company. I think we are going to have some folks from Anthropicon next week, which I'm excited about.

Yeah, so the headline in Wired is Anthropics' new model excels at reasoning and planning and has Pokemon skills to prove it. Claude Four Opus and Claude Sonnet Four can remember over long periods of time, a capability that's helpful at Pokemon and other. tasks that require an ability to stay on track. So what was most exciting? Where should we start first with understanding the progress at Anthropic? So first of all, you've got to read the every headline and the subhead. Every time one of these models drops, we do a vibe check. So basically we get access these models hands-on before they come out. And then we use them hands-on for the daily tasks that we're doing for build software and make writing. And then we give you basically hands-on review of like out every part of the model works.

What I found is it's a piece of coding. It really runs autonomously for a long time on like complex P.Rs that were not like not possible with three-sevance on it and are probably a little bit beyond the reach of like Gemini 25 and 03 in particular in Claude code. It's fucking awesome. And so I do think that that's a it's a sort of game changer if you're doing any kind of development work. for writing and editing.

Real quick, how do you, what do you look at the, what are the kind of like market implications of Claude in terms of the broader developer tooling ecosystem, right?

You know, we've seen, you know, cursors developing their own models, windsurf joining OpenAI. There's a lot happening, but, you know, do you think that Claude code is something that could have billions of ARR independently or what's your read on how it's going to evolve? That's a good question. I do think that the landscape is starting to consolidate a little bit,

and with each model advance, the extra intelligence matters less and less. So I think you'll see people, you know, instead of jumping, every time there's a new model release, instead of jumping and turning their entire stack from all models than new one, I think you'll see. people can start to stick a little bit more in the ecosystems that they've chosen. We're entering a little bit of a different era of the race, and I think it'll be a little bit more difficult for Anthropic to be like it's totally independent lab and go up against not only Open AI, but also Google, assuming Apple at some point gets their shit together, Microsoft, whatever.

But I really think that they have a very, very strong play with developers here. They have 3-5 Sonnet and 3-7 Sonnet are like superb coding models. 317 is a little bit over-eager, so it's, you know, probably not as good as Gemini 2.5 Pro, but it's still, like, really up there with the best coding models. And I think QuadCode is one of the best coding experiences, like AI-first coding experiences. It's also just very different from, like, a cursor or windsurf where those are basically text editors with AI on the side. And CloudCode is a command line interface that's like you're just directly interfacing with the, with the AI and that's it's it's just a different experience that's intended to be an assistant

an agent rather than like a chat yeah uh do we have any idea of where these new models uh clod four specifically sit relative to the open AI models is this because i remember four five came out and they didn't give it the gpt five name but it did seem like it was trained on order of magnitude more compute. And so this is kind of where the pre-training wall discussion came from. But is it fair to think about Claude 4 as a GPT 4.5 class model in terms of kind of the scope and the size of the model? Or did they give any indication as to the vectors that they're pulling on?

We've seen, you know, Facebook with Meta-Lama, behemoth is. going for this massive context window, these trillions of token parameters. There's a whole bunch of different threads that the different foundation model companies are pulling on. What seems to be their underlying motivation or what do they seem focused on in terms of optimizing towards? So it's a little hard to tell, at least for me, because I didn't get access to any kind of model part before launch.

And I honestly just didn't have running around like just at the event and stuff. So if there are actual numbers and stuff, like definitely go look them up. I don't know for sure. On the blog post, they mostly just focused on the software engineering sui bench and the different benchmarks around graduate level reasoning, agentic tool use, multilingual Q&A, visual reasoning. And the stats are impressive, but not a huge jump here. There's a jump on agentic coding. But Claude Sonnet 3.7 actually outperforms in visual reasoning by like a half a percent, not much.

But it's one of these interesting things where it feels like we're increasingly in the era of tradeoffs around models. And we might be seeing more fragmentation around a really great. We talked to some founders who are building super specified LLMs just for JSON decoding or just for translation or just for profanity filtering. And then they run the inference on a consumer grade GPU because it's so. narrowly defined still using the transformer architecture but just much more narrow and so I'm wondering if if that's the future that we're going to see at anthropic is they're they're they're dominating in code they're very popular with developers and they're going to go after a few other areas but they're probably going to stay out of a few other battle battlegrounds I do think that's an

astute point sometimes you do see tradeoffs and so like 37 sonnet for example was a better programmer than three five sonnet but it was like way more optimistic It's interesting. And but to answer your previous question, like going back to the difference between GPD 4.5 and Opus, Opus 4 and Sonnet 4, it's a little bit apples to oranges because 4.5 is just a base model. I mean, it's an instruction tune, but it's not a reasoning model. And Opus is both. They can go back and forth between being a base model and and being your reasoning model.

And that is like this sort of hybrid thing that I think all of the models providers are starting to move into. But I think Sona has, Anthropica has done it first. And it's really good. It makes a big difference when it both can do the kind of chain of thought reasoning that's like good for map groups and thin coding.

And it also has a little bit of a vibe. So like a GPD 4 or 5, which is just really good at writing and really good at great at free things. Cool. How do you kind of navigate and figure out what's real when everybody in AI is deeply conflicted in different ways, whether they're sort of like secretly an advisor to this lab or, you know, happen to invest in the series B of this one or, you know, everybody's sort of talking their book. I mean, obviously. I just got advisory shares in Open AI. It just got 1%. That's it. I'm not conflicted. It's just like a point. Yeah, you've got a point. Careful, John.

People are going to believe that. Every time we joke, every time we joke around, every time we joke around, somebody takes it 100% of fact. No, but obviously the immediate way to figure out what's real and what's not is just to use the new products and models yourself. But I'm curious how you navigate it as somebody who's like trying to provide, you know, really, really, you know, precise coverage of everything in real time. So first of all, if anyone wants to throw me some anthropic or open AI stock,

like I'm open, the answer open. But I think, yeah, I think that's a really good question. I honestly think most of the benchmarks are kind of bullshit, and you can train the benchmarks, and you see that with Lama, where again, we'll put it on the benchmarks, but it's just not a good model. That's why we do vibe checks. So when we launch new models, we're using them hands-on for the tasks that we do every day. And I think that's why it's really valuable to have every the media company and every the incubator startup studio in the same organization.

Because I'm literally just going and hanging out with here and who runs Sforra, which is our AI email assistant or Danny, who runs SPAR, which is our AI content automation product. And we're using it to ship features or to make better writing or to do all the things that we do every day. And so I can get a pretty good sense of just like the vibe or the flavor of the model from using it myself. and I think that's going to be the best. I think that's going to be the best way to tell them it's any good. What was your reaction to Google I.O? What was the single thing that stood out to you from the event? So I was at Build.

I did a really fun interview with Microsoft CTO Kevin Scott and then flew right from Build to code with Claude. But Alex Duffy is another one of our writers who leads AI training for us. And he was at Google I.O. And so he would have a better response as to like what is the one thing. But I do know he was like, he was like kind of in tears in our discord and was like, this is the future. And I was, and so, you know, I think that they've got something cool going on.

I think he's really excited about they're pushing forward on a lot of different parts of the ecosystem all at once and they're living super fast. And he seemed really psyched about it. Awesome. What about the products that Anthropic hasn't launched? We talked about the specialization a little bit. And I remember, so the big AI product that really grabbed me from Google I.O. was V-O-3. And obviously, Google has an immense advantage with the YouTube data set there.

And the results have just been fantastic. And I've been fighting tooth and nail to get more VO3 credits because it's so much fun. And Anthropic, it feels like they've been potentially behind the ball on image generation, Is that just not a focus for them? Do you think that's because of safety concerns, since obviously that runs very deep in the culture at Anthropic, or do you think it's just more of a business case that the coding market is much more lucrative

than the image generation market? And so just focus more wood behind fewer eras. That's your question. I honestly don't know the history of Anthropics relationship. with coding models other than like they just don't have one. And I don't know why. I do think it seems like they're really going hard after the kind of agented coding market.

And I think it's a really valuable thing that they bring in. I'll also say just generally like Mike Krieger is their chief product officer and like he knows a ship. And so I think they'll probably have certain. He knows the thing of too about images. Yeah. So I assume they'll have some really cool stuff on the more like consumer product side coming soon.

I think we're getting the Zoom air. I'm searching O3 right now for Anthropics offering. I don't think that they have any video models. Dan down. Dan down. Dan down. Bring it back. I think, do you think there's a widespread cyber attack?

Probably. Because X is really struggling. They know who we talked to earlier this morning. They do. They know. They figured. it out. We have a secret guess that will be released later. They're on to us. But O3's cooking. I mean, it's, I don't know if it's writing code yet,

but it's definitely it's written. Look at this. It's like, I just ask it, I just ask it, has Anthropic shelved any, any models? And it's like searching the web everywhere. But it is, it is a tougher, it is a tougher question for O3 because I'm asking for the absence of evidence, not not, not, I'm asking for it to find the absence of a model, not the, not the, not the presence of a model. And so it's very easy to search. Does, does Claude have a coding model? And it can just search Anthropic code and find the first result and return that to me. But it's very difficult to find a null and say, okay, I searched the entire web and I couldn't find anything. Therefore, Anthropic does not have a video model. But yes, I believe Anthropic scrap.

Maybe it was a text-to-speech engine, something like 11 Labs competitor. What you got? So I've been, I went to try to ask Claude. Yeah, yeah. If, why doesn't Claude have an image generation model? And the sign up flow couldn't be more opposite of Open AI. I've just been, I created a new account with my TBPN email. And it took me, there was, I mean, Nikita would probably have a heart attack. I mean, it was like 20 individual steps of approving this and improving that.

but it says Anthropic hasn't publicly released a dedicated image generation model, strategic focus, technical and safety considerations, resource allocation, market positioning. So kind of everything that Dan said, just, you know, it does make sense in a way to just say, hey, this isn't strategically critical for us. We're going to try to, you know, focus on what we're really good at, which is coding. That's great. Well, we have TJ coming in for the second time. I'm very excited for that.

Yeah, did we fully lose, Dan. I think we did. Let's just move on to some news. Let's talk about the, I mean, there's plenty of timeline we can go. Dan, if you're listening, thank you for coming on. Yeah, we appreciate you. We enjoy the conversation and always welcome to come to. This is the future.

We have AI that can one shot build you Zoom, but it won't work. But Zoom is still going to go down. The internet will. still fail constantly. Anyway, I do think it's fun that, uh, that, he says his phone overheated. He was just going so hard on, on, on, on, you know, corresponding. Wow. Yeah. That's that's, that's, that's, that's, that's, that's, that's, that's, that's, that's, that's, that's, podcasting. Yeah. You need to be, live streaming video, uh, doing spreadsheets, co-ing on your phone.

Dan, thank you for leaving it all. On the field. On the field. You're an absolute dog. We will see you again soon. I do think it's fun that they're doing, that Anthropics doing Pokemon because Pokemon's such an iconic game and it's so, and it lends itself to like you can watch the full stream of Claude play Pokemon.

And it is this interesting challenge that it takes a long time. They need to think about the different reward functions and all the different steps that you have to take. I was playing Pokemon on the Chrome. Pomer Lucky's new Game Boy. And it's actually really hard because at a certain point, you get to a place where you just kind of have to go farm XP and level up your Pokemon evenly before you can go beat the boss. Like, you can't just constantly be going forward.

You actually have to have to plan out your attack a little bit. It's not the simplest game. So impressive that they're able to play it for, what, 24 hours at a time. Anyway, let's go through some timeline. We'll get to Elon Musk later. There's a piece we've got to put in the truth zone. but it's going to take more than seven minutes. So let's talk about Open AI.

They committed to a giant UAE data center in global expansion. This is Stargate UAE. We love to see it. They're partnering with G42 to build a one gigawatt AI data center in Abu Dhabi. It's the first large-scale project outside of the U.S. Shagate to New is just cooking. He's on a tear. I cannot wait to have the opportunity to have.

I want him on the show. Him on the show. I want him on the show so badly. G42 will fund the construction. Open AI and Oracle will operate the data center. SoftBank, Nvidia, and Cisco are also partners in the project. The Koretsu of AI is coming together. And so the UAE.

Another $100 billion for Nvidia. So the UAE is trying to become one of the biggest funders of AI companies, infrastructure and build their own AI factory. The first 200 megawatt chunk of the data center is due to be completed by the end of 2026. The thing that's interesting here is I, Chase Lockmiller can get in on the action. They're having to like match the investment that they do abroad. It's a very interesting structure.

It seems like a good deal. I'm very bullish on it. They're doing deals. That's great. Strava raised at a valuation north of $2 billion. This is fascinating. I have to ask what are they running from? What are they running from?

Maybe they're running from they could very well be running from Zach Pagrob. who is in some ways coming for them. They're running from not having generational wealth by getting liquidity at a multi-billion dollar valuation. So congrats to the founders. Running from not having, yeah, it took me a second. Sequoia. And they raised some debt.

Let's give it up for leverage. They also acquired the breakaway, a cycle training app marking its second acquisition in two months. They had more than 150 million users and it was approaching 500 million in annual recurring revenue. that's good. Let's give a moment of silence for the VCs that passed on the seed in the A because it seems, you know, yeah, like running. I get a lot of people do it.

How big can this be? Easy critique. How big can this be? 500 million in annual recurring revenue. They showed you. Got them. Turns out a lot of people are running from. Strava lets users whom it calls athletes, recording quotes, Wall Street Journal.

Put in their term in the truth zone. Hey, runners are athletes. In some contexts, yes. They're running down a gridiron maybe. Share activities with friends and across dozens of sports. By the way, I've been avoiding, we had Rob Mower, co-founder of Huberman Lab. It's just more.

More. He just texted me. I botched his name every time. It's terrible. Rob, I've been avoiding a run with Rob. Oh, yeah? For years now. We live close by.

Is he a good runner? He's going to smoke? He's an absolutely insane athlete. Oh, and I just always end up busy. How's he on the bench press? Is he reping two plates? Because maybe you could give him a run for his money over there.

A big dog. Show him how it's done. Good. Potentially do that. I can go back to back. I'm going to have a Huberman bench three plates, and they're just going to smack us. I could see Rob just secretly repping three plates. Just demolishing us. Anyway, let's go through some timeline.

Jenny says, glasses won't win earpiece with a tiny camera will i don't know about that what do you think i think i just would like i would like a camera nostril i'd love to put a nostril camera in that just sort of mounts to my nose and just comes out right here no one's talking about nasal computing nasal computing that's the final frontier for for guys like us that don't want to chip yeah the nasal passageway is a great place to put a computer i mean there's probably so many different applications You could have little fans in there that accelerate the airflow in so you're getting more oxygen in your blood. Brian Johnson loves it.

A hyperbaric chamber for your body. Yeah. Woop on the wrist, but we typically have watches on our wrist. Yes, yes. And so. Whoop on the nose. Whop in the nose. Nose.

Or a ring in the nose. Or what about an or? You know where you're going to do. What are those? Septim piercing? Septim piercing. Septim. An aura septum piercing.

Oh, my God. Oh, that's insane. AI, AI generated right now. The nasal aura ring. That's insane. Anna says, we're in the, or Anna says, we're in the good old days in the beginning of the movie with the overly saturated colors and the smiles and laughter.

Your future is being written by about six people competing with too well their ideal civilization into existence. record freeze frame record scratch hi Anna here you're probably wondering how I got here yeah it does feel like it's a it's a monumental time it's a it's an amazing time to be a technology podcasting jensen Elon Satya Zach yeah uh accelerate harder says this blog post is styled like a cross between a wedding announcement and a memorial web page it is a very funny photo it is black and white and the font choice is is something that we haven't seen from open AI or Apple before they're boys They're boys, though.

They're just guys being dudes. Anyway, we got a couple other posts, but I think TJ's here. So should we just bring in TJ and start breaking down the news? Let's do it. Let's bring him in. Welcome to the stream, TJ. How you doing? He's getting the band back together.

He's getting the band back together. Let's go. Congratulations. They're back. Let's go. Keep it go. Keep it going. Two minutes of this.

And I have a microphone. I've made a real, real upgrade here. Oh, you love that. I love that. I love that. Yeah, you went all in. I love it. Yeah.

Big podcast guy. We turned you. Great to have you here. You went from zero to 100. It's great to be back. Yeah. Give us the news. What's the announcement?

What does this company do? I saw the announcement, a bunch of mumbo jumbo. Break it down for me. So, companies called general medicine. We launched today in all 50 states. So anyone in the US can sign up and use the service.

And the way we would think about it is a healthcare store where you can get anything you need in healthcare. We actually talked about some of these themes a bit a couple weeks back. But our general belief is that you should be able to shop healthcare like you shop everything else. And so that means, you know, if you have, if you know exactly what you need, like you know you want a certain set of labs,

or you know you need a colonoscopy, or you're having kind of ongoing shoulder pain, is like an issue I've had that I use the service for. You can literally search for that problem, fill out a relatively straightforward set of questions and then consult with a clinician. And if what you need is an in-person visit or in-person care or you need to get imaging or any of that done,

we make it super seamless to do that. And I think probably the most novel thing about the product experience is that you can see personalized insurance pricing for any provider prior to deciding where to get that, service. So for my shoulder pain example, I needed to go get a follow-up x-ray. And the difference in price between getting an x-ray done in Park City versus Salt Lake was $100 versus $500 with my insurance. Wow. Which before when you're making that decision, same insurance, same insurance,

same insurance, same X-ray, same actual kind of medical group. They're both kind of intermountain facilities, but a pretty big variance in the actual out of pocket. Presumably even the same X-ray device, like the same technology. Yeah, all the same. I think people are making this decision every day, but they literally have no idea. Yeah. Well, there's any sort of differential. Yeah. So, so the thing that stood out to me is when you land on generalmedicine.com, you see pricing without insurance. Yeah. And that seems you know, super intentional because you want to just enable people to not, one, just like speed to care, right? There's like certain instances where you're like, I need to solve this problem as fast as possible.

I don't want to deal with worrying about in network, out network, all that. And then there's also instances where like something can cost you more by going through insurance. That's correct, right? I remember something was going viral on X probably a month ago where somebody like there, the ambulance was like much more expensive because they had insurance where if they were just uninsured, it would is going to cost less. So can you talk about kind of that, the way that kind of pricing dynamic works and why you guys decided to kind of lead with

showing people that hey you can just kind of immediately get these services regardless of your current insurance situation yeah I think probably a useful anecdote is when we built a very similar experience in pharmacy when you go to kind of Amazon pharmacy now you can see cash and insurance kind of head to head when you're making a purchase decision we ended up with a significant amount of folks that were just using cash because either it was cheaper, which I'd say a quarter of the time, it was literally cheaper to just not use their insurance just to pay cash. And then a quarter of the time, it was moderately more expensive. Instead of 15 bucks, it was 20 or 20, it was 30.

But you didn't have to deal with all of the obnoxiousness of using your insurance, right? There was no prior authorizations and to deal with the fact that it was too soon to fill the prescription, like all these sort of sad pass that exists as a byproduct of insurance. I think we'll see a similar dynamic here to your point where sometimes it'll literally be cheaper. And that will happen, I think, a lot more than people would think. And then sometimes it's slightly more expensive to use cash, but it just is less annoying to deal with having to get it reimbursed and to deal with the upfront prior offs and all those things. And we think that'll be a, I don't know what the right proportion will be for this business, but I think it'll be a much higher percentage than people would assume. How do you think about the continuum of medical services?

because I'm thinking about the shoulder injury. Obviously, there's sometimes when you need an x-ray and you need insurance to pay for that. Sometimes you just need to rest. Sometimes part of health is maybe going to the gym or being on a diet. Or if you're having problems with sleep, you could need a prescription drug or you could just need some melatonin. And so what is the ideal front end to health? I mean, I imagine people will go to chat GPT. They go to WebMD.

They ask their friends. Sometimes they go to their doctor. prematurely and the doctor just says, hey, just, you know, stop eating so much or you hit the gym or something like that. But what is the, how will this change and then how do you plan on on being a part of that front end to medicine? I mean, I think the single biggest change here is trying to move healthcare to a much more simple and straightforward transactional experience where you can actually transact in the same ways that you transact in retail. I think the overly, if you think about just the overarching kind of journey for someone dealing with a condition or dealing with some symptoms, they don't know what condition they have, I think the behavior will be similar to what it is today, which is today they're going to Google and constantly searching for things and trying to figure out what's going on.

I think a bunch of that volume will move to chat GPT and other foundation models where it's easy to go from symptom to a likely diagnosis. I think the problem we solve is post that diagnosis or post having a sense of what your problem is. being able to quickly get what you need, right? Like you might need an X-ray, you might need labs, you might need an intervention in person. We make that really seamless. You can transact with us in a normal way. And so I think that is the sort of novel thing

about what we're doing is turning healthcare into really a product catalog with super crisp pricing up front. What's the business model gonna be? I think the right mental model is a combination of a first party and third party marketplace, right? So we do have our own medical group, we have our own doctors on

staff. We will use them over time for things that are more and more complex where it's not as easy to know kind of right away exactly what you need and so you're helping someone navigate that ambiguity. And then we both have a set of third party providers that are deeply integrated, mostly specialists where if you need an immediate specialist consult, we can provide that. And then you can literally request to see any provider. It doesn't matter if we have a relationship with them or not. We can still make that referral, we will literally book the appointment for you and we'll show you that again we'll show you like where the best option is for you based on price quality whatever things you care about and in the the third party sense it obviously won't be

as simple as a traditional marketplace from a monetization standpoint but it will be of that ilk right of that flavor of a kind of traditional first party third party marketplace did you guys have an urge to slap an lLM on this thing I know it's notably absent from the home page I think thrown it back. Just like my cars, we're like, yeah, we're throwing it back to. Yeah. Yeah, I'm curious if you, you must have gotten a bunch of pitches from people being like, hey, a lot of people are using chat GPT to diagnose conditions. We're going to make a better version of that. To me, doesn't feel super investable just because chat GPT probably already has full access to PubMed and a bunch of, you know, other resources, but yeah, was that a pretty intentional decision to kind of like

leave off the site and let people kind of figure out what they need wherever, whether it's Google or WebMD or family, friends, another doctor, et cetera, and then just be the place where they kind of take action? Yeah, that was definitely intentional and definitely something we constantly debate. Again, I think the place that Chad GPT and the other foundation models aren't going to go is the stuff we're doing, right, to go from likely diagnosis to actually transacting. And I think we'll see how it plays out. I'm a little skeptical that dedicated kind of doctor LLMs will win against the chat GEPTs of the world. We're probably effectively placing a bet here that the chat GPs

of the world will kind of sweep up that use case for the most part. We could be wrong on that for sure. I mean, I think there's a bunch of Yeah, you're going to imagine a world where general medicine is like, hey, you know, we think you, here's like our read of the situation based on the picture you uploaded or the symptoms you're having. Here's how you take action and actually talk to a specialist to like verify, you know, actually get a, get an expert involved that's not just, you know, predicting the next token. Totally. I mean, I think, again, the right frame here is retail, right? You're already seeing that exact flow for categories that have

this clear kind of product catalog and transactability and doesn't exist in health care. I think we'll have some version of that kind of chat with us and figure out what's going on to help you navigate our store over time. But that's not the bet we're placing. The bet we're placing is that it's the transaction that the customer ultimately needs and that we can create a very novel experience. You guys launched in all 50 states. How different is that than the early Pilpack days? I imagine that was it kind of a different go-to-market motion? Yeah, for better or worse, it's almost the same. Um, Popak took us about 12 months to launch and we were in 32 states, I think, at launch and then chipped away at the tail over like six to nine months, I think, if I recall. And this were now, I don't know, 16, 18 months in and we're launching in all 50 states.

So on the kind of from a time and availability standpoint, actually super similar. Can you take me through some of the history of other attempts in this category? I remember using Zock doc once to find a doctor. I used wellness effects once to get some labs done. There's a few different approaches that have been tried in kind of creating the front end, the front door to medicine. It still feels like one hasn't become dominant. There's a lot of point solutions. Like if you want hair treatment, you go to hair.com or something.

You want compounded erectile dysfunction medicine with GLP-1s in it. You can go to a different website for that. I probably won't be part of our selection. We debated it. I think we're going to leave it out for now. I want just the cocktail. Cocktail medicine.com. Testosterone.

Cherectin. I want some protein in there. Throw a 20 milligram aterolics are in there. And then mail that to me, compounded. And yeah, just don't, I'll pay in Cardano, actually. Yeah, exactly. BNPL. BNPL with Cardano.

Yeah, that's great. But yeah, take me through some of the history. You don't need to speak specifically about specific companies, but just the different approaches that have been tried, what worked, what didn't, why have you landed on and what have you learned that led you to this particular approach? Yeah. I mean, there's a couple of vectors there, right? I think the initial thinking for folks trying to build similar stuff was

that you'd start in one vertical and go horizontal over time. That's sort of the Amazon playbook starting in books and then adding categories kind of as there's demand and you have incremental customer demand. I think we observed that and have not seen evidence that that approach is successful. And I think one really different thing about healthcare is that oftentimes you actually don't know what you need, right,

when you're showing up. And so if you only serve one specific thing and you only have one intervention for that thing, it actually doesn't really really really serve the customer need in the way that they ultimately want. I think if you compare it to kind of booking services with like 3P provider search and you can book an appointment on the site, the thing that we've observed there is that you end up with this selection bias towards providers that have a bunch of availability

rather than the best provider for the customer. And so we're not requiring any level of integration to get you an appointment and get you pricing for any provider. So we don't suffer from that kind of same selection bias. We're effectively ambivalent who you want to go see and we can help you figure out who the right person is. So we've sort of learned from observing that as well. And I think the last point I'd make is that it's novel to have both first party and third party integrated as a seamless thing. That's novel outside of, certainly outside of health, you know, in healthcare.

It's not as novel outside of healthcare. But by doing that, it allows you to offer a comprehensive experience and ultimately make it super, super seamless for the for the customer yeah um have there been developments on the legislative side that have allowed this to exist now um i remember there was some telemedicine rule changes that happened in covid and kind of stuck around that kind of created a boom there maybe but could you have built this 10 years ago i think there's probably two main things that have made this the right time to build this business. The first is that patient data is much more accessible. So something novel about the customer experience is when you sign up and you give us very basic demographics, we can go

pull your insurance information and pull your full med history and medical record and repopulate everything in your profile and these flows. That wouldn't have been possible a handful years ago and that was due to regulatory changes that required companies to open up this data. And the second is the pricing would not have been possible to do in the ways we're doing it without LLMs because we're effectively reading your full coverage of benefits and then mapping that to the service that you need and then figuring out where you are in a deductible, all of that would not have been buildable without the current version of LLMs. And so we're not exposing that in like an AI sense, but it does power how we're able to back

into what your pricing ultimately is. That makes a ton of sense. Yeah, it does feel like there's huge opportunity for companies that are essentially driven by Enterprise AI or AI internal behind the scenes, but they don't just need to surface that. You don't need to create another text box for someone to chat with. That seems like a very deliberate decision,

but still like uniquely enabled that. Yeah, and if somebody can create the AI doctor that allows you to have a telemedicine visit and you guys are kind of the front end for getting expert care, you could just plug that in. You don't even necessarily have to develop it. In terms of, I don't know how familiar

you are with how the doctor's office is changing, but there was this drumbeat for years in the deep learning community about stop training radiologists, deep learning neural networks, focused on image recognition, will be able to do it at a superhuman level very, very quickly. Now we're seeing Google V-O-3 generate Hollywood-level cinematography that's pretty... Generate T.J. Parker going on a thousand podcast in one day. I mean, it's really, really good. And so when I see the real, I'm sorry, guys.

Yeah, yeah, yeah. Yeah, but I mean, when I see the advances in image processing and generative AI with regard to image, I feel like there has to be a similar progress happening in the image processing of self-driving cars. So when I see the chat GPT studio Ghibli moment happen, that advances my, timeline to a self-driving car. It also should advance my timeline to take a picture of this mole and

tell me if it's cancerous. But is that happening or are these separate paths in the tech tree or are there just barriers to actual adoption in the medical community or are their, you know, job displacement fears? Like what's actually going on over there in the doctor's office? Yeah, it's for sure happening. And I think they're as effective as you would imagine they are. I think, the, probably the biggest difference in healthcare, and I'm sure it's obvious, is that you still have to get the sign off of the dock and there's still a bunch of regulatory overhang that is required to get fully to a diagnosis and certainly at this point to an intervention, whether that's a prescription or some other intervention. But I think the enablement and the

efficiency is showing up, and I think that'll continue to continue to happen. Now, you're not a chat GPT wrapper. You're leveraging LLMs, though. Are you a fax machine wrapper? We have definitely, we'll always be a fax machine wrapper. I mean, I bet I, I bet that we spent like a third of our dev resources at POPAC for like three years on fax-ish stuff. Wow. It's really, it's really quite stunning. Let's give it up for the fax machine. One of the greatest one of the greatest ever.

But I mean, I imagine that there has to have been like a B2B SaaS company that created like a really great API around fax machines in the last decade. Has that happened? Are you just standing on the shoulders of giants? Or are you writing fax machine in a Rop code for the first sense? We don't even do a build it or buy it. We just always build it. If it's fast, we're building. Yeah.

Yeah, I imagine you're hiring Kuda engineers to develop custom parallel. Yeah, part of the 30 plus million is going to, you know, setting up, there's a JV with the UAE, I'm sure. Yeah, there's a whole special vector of engineers that will only work on fax machines. We have a lock on that talent pool. That's great. I want to talk about go-to-market. Oh, sorry, but yeah. Okay, please.

One more question here, and then I want to take it another direction. Yeah, so I mean, obviously it's important to drive folks to this. Give us the plug. How can people get started? But then what's the actual marketing rollout? Are there partnerships that drive adoption? Am I going to be seeing flyers for your service when I walk into a doctor's office or seeing Google ads?

Like, how do you actually plan on getting customers? Yeah. I think this arc will probably end up following the POPAC arc as well. We're very much a D2C business today and we intend to stay there for a while. And so I think if you're searching for any of the things that we offer, we'll show up in those search results like any other retail experience. So we'll get quite a bit of demand there. We'll do all the normal kind of DTC tactics to drive awareness and demand. I think if you kind of assume that. So, I think, if you kind of influence a partnership. Yeah, exactly. Yeah. Really get all the, all the key

influencers yeah a lot of lot a lot of stunts a lot of blimps flying a lot of stunts lot of ticke righting yeah exactly but I think if you if you zoom out and look at the Pilpac journey by the time that I left that business two-thirds of our business was B2B partnerships and B2B driven sure but we feel like to build a novel and great customer experience you have to start by marketing to the end consumers so we're doing that again here but we would expect partners over time to help drive demand yeah I give it two years so Andy Jassy is calling you daily trying to get the band back together again in Amazon.

I wanted to go a different direction and get an update on kind of the fallout from the drug pricing executive order. What's your kind of updated thinking there? It was only last Monday. It feels like a month ago. Yeah, it does. Yeah, I don't know if I've seen anything substantive over the last week. Yeah, I'd say my take is still roughly my take,

but I've not seen any evidence that supports it any more than I had a week ago. So hard to say. Should we SPAC NITO engineering? Absolutely. Spacks are back. Much higher margin business for sure. Yeah, NITO engineering is his family owned. It's sort of in the 50s car restoration, you know, engineering business that T.J. is a big fan of.

He's wearing a hat. So I had to ask since Spacks are back, you know, potentially. potentially an opportunity there. My plug is that I produced a documentary about Naito and it's coming out in the fall. So keep an eye out. We'll have to hop back on TBPN to talk about our documentary. Yeah, I think I saw maybe, was there a trailer that already dropped? Yeah, I saw the trailer.

That's cool. I love that you're mixing, you know, hyper capitalism and, you know, just like post-exit, post-economic, you know, documentary producing, you know, very few can do both. Yeah. And Naito is not related to the car garage that you're involved in. Is that right? Yeah, it's separate. That is my car. Storage business, which is the car business. Cool. That's called warehouse. That's my other plug. Parks City. There we're going. We'll do it. We'll do it. We'll do it. We'll do it.

So House meets like car storage. It's a cool, cool concept. Is there a track or where do people take the cars once they're stored there? We just go like, we go on weekly drives. So there's a bunch of places to drive outside of Park City. And then there's like a simulator's restaurant. bar kind of social lounge that kind of thing so kind of half social half car enthusiast stuff yeah uh i mean very quickly just on the news because everyone's talking about the johnny i've going from apple to open a i nikita was saying it's 100% due to apple's compensation structure they can't pay johnny 10 billion uh but and you said it's 100% the right take but why can't they like like it seems like at this point, like maybe they should be paying Tim Cook more. We've been joking about it. He only makes

$70 million, $60 million a year. And yet he was able to navigate the tariff, like war pretty adroitly. It seemed like he probably saved Apple from a $300 billion market cap hit during that fiasco. He doesn't really get to take a slice of that. Is this something that companies should be like set up for going forward? Or is this just a unique dynamic of the private market? like what's actually going on here? Yeah, I think it's a fundamental flaw of these large companies is that they're designed to reward generalists for the most part. And there's very little differentiation available on comp even if you're at a pretty

senior level. You've sort of run into these natural comp barriers. I was less commenting on whether Apple should pay Tim, $10 billion should pay Johnny $10 billion, but more commenting. Like if you bump up against breaking any sort of compensation rule, making an exception, it's incredibly difficult and it's basically impossible. And I think it's especially odd when you're willing to, like there's an incredible willingness to invest in a new thing, like well beyond kind of ventures tolerance for capital investment. Totally. Like billions of dollars a year into the metaverse. Like just super speculative. Amazon spending $6 billion in Alexa, like super

speculative investment with no disproportionate economics for the person deciding how to make that capital investment, which I've always thought to be very odd. You could effectively swap out kind of 10L7s and dramatically change the comp of the VP running that group. And that's probably a great investment, but that's just not the sort of mentality at most of these large companies. Yeah, I mean, you had the kind of canonical founder, exited to big tech, moved on pretty quickly, experience. Like, what was the one thing that you took away from Amazon that was like, this is great? I need to port this elsewhere. Was there anything that stuck out to you as like deeply underrated about those large

organizations? Because it's easy to say, oh, they're slow. Oh, they're all there. They, you know, that blah, blah, blah. But what was actually interesting about Amazon in particular or just big tech broadly? Yeah. I mean, Amazon's way of working, especially the kind of writing culture, we definitely have co-opted. And I would never kind of operationally do it a different way.

Like, I think that is quite effective. There's like a reasonableness that probably is a little bit. you know, tilted at Amazon, like you're literally writing like your annual budget. Like that there might be things that there's other formats that make sense, but as a general rule, like I think a writing culture is much better than a traditional culture. Yeah. I think they're also incredibly good at scaling stuff that works, right? Like they are a machine if they found product market fit, both because there's a deep willingness to invest and because there's a bunch of established mechanisms to scale this stuff.

I think that can be inhibitive of building something novel and new, right? There's like sort of over process and over litigate stuff early on. But the second you have something really working inside of Amazon, like they are just masters of scaling it up. Yeah, this is the same thing with Google. Ben Thompson was talking about with regard to the latest Google I.O. launch. Like Google launches a ton of demos, ton of products, but they never really go anywhere unless it's tied to search.

If it's tied to search, they will die before they let that thing not work. And so you can talk all about, oh, they say. sunset Google Buzz or circles or fly or this random thing but like they haven't they haven't sunset shopping like they haven't sunset display it's still going that one's still going um yeah yeah jor do you have anything else or i think that's it uh the only thing i'll share it with you now i just got breaking news uh and tj will laugh at this but uh john really wants the new zr1 oh yeah uh which which which is like probably on the opposite end of like the spectrum of like cars tj likes and like doesn't like but i just got a quote that basically you'd have to pay a hundred k over msrp to get one there's

only been 12 delivered so far so tv p bn is ripping it's a small price to pay for an american made uh maybe maybe we can flip that around uh i so i want an american made sports car i want something interesting but it has to be american made what would you recommend for me for can it be old Yeah, it can be anything. Get a cobra. Cobras are sick. Okay. Yeah.

That might be that. Do not get a kit car, but like get an OG unrestored cobra. Those things are sick. Okay. That's good. I mean, last thing on the writing culture, do you think that Amazon is at risk of being kind of one shot by these LLMs? Because you don't actually have to go through the exercise anymore of writing a memo. You can just be like, get me $20 million and I need to hire 25 people.

Justify it like it's an Amazon memo. Boom. Yeah. You're just going to drop in all your old memos and be like, update this for Q2, please. Yeah, exactly. I imagine that that kind of destroys the culture potentially. Yeah, it does have a real. There's a real risk there that I hadn't contemplated.

Yeah. Anyway, yeah, I mean, we were recently posting a job posting for an editor and the job listing was very clearly chat GPT generated. And I was like, what prompted you use? Maybe we should just post that prompt. And we wound up just posting the prompt. And it's like way better because the prompt is way more succinct. It's like, you had to give it anything.

Just give me the bullet points. Don't flesh it out into two pages. I am pretty upset that chat GPT is going to single handle the ruin M dashes for me. Oh, yeah. The record was like my favorite punctuation pre-Chad GPT and just ruined. Are you a delved guy too? Were you using delve a lot? Not really.

Not really. Just end dashes. Yeah. Anyway, great talking to you as always. Yeah, congratulations to Ashwin and Elliot. It was great. This is, I love that you guys are getting back together to fix a big problems. Fantastic. You'd love to see it. Great having you on. Good to see you guys.

Enjoy the rest of the day. See it. Let's tell you about numeral sales tax on autopilot. Send, spend less than five minutes per month on sales tax compliance. Benchmark series A. Benchmark series A. I love new. Series B income. Also get on public. Public.com investing for those who take it seriously. Multi-asset investing industry leading yields. They're trusted by millions folks. Millions. I can't wait till we can say trusted by billions. Trusted by trillions. Let me see. Trusted by Martians. What else we got here?

Do we have any others? Oh, we have our next guest already. Easy. Let's bring them in. Who got? There he is. Reggie. Reggie.

How are you doing? Chief, we didn't have a title for you, so I put the CRO, the chief Reggie officer. It's we're maximizing free agency boys. Yeah. Free agencies and trade season. I wouldn't be surprised if you get a maxed-out offer before this in your inbox before this. 6.5 billion's on the table now.

So anything's possible. I had an idea shortly before you, maybe you could build it. A septum-piercing AI wearable health tracker. Anything there? Ship it. Ship it. A chip it. An aura ring for your nose.

That can also be a companion. Yeah. You're talking to it all day long. It's like, you sure you want to? have that you know uh diet coke yeah but scent is tied to memory so imagine it just squirts the scent up maybe not maybe not squirts remember yeah it admits an aroma to remind you of your aroma aroma is better an aroma yeah this aroma is built in this is a this is a trillion dollar idea here what yeah i mean device they're making apple or uh johnny ives over at open a i i

I, we were kicking around what we think it might be. It feels like Sam Altman's been into her. So we're thinking single earpiece that you wear. Maybe it has a camera on it. That was kind of the best thing that I could come up with. Maybe glasses. It feels like they're probably not just going to do a phone or an iPad or a watch, something with the rectangular screen on it. It's going to be something a little bit different. Take you off of your phone, maybe curing dopamine addiction. what are you expecting and then we can go into all of the different all the different aspects of the deal yeah for sure i mean i think you know one thing that felt very clear is that they emphasized like a sort of like family of devices right so i think already they're coming

from a multiple objects approach which which i think is cool mostly because i think we're a lot of hardware companies sort of get stuck is they launch one thing and then everyone only knows them for that one thing and then they get stuck in the loop of that one thing right so like woup is 17 yeah pebble pebble like invented the smart watch in many ways and then apple came out with the apple watch and it integrated with the phone and the iPad and stuff so yeah i mean lack of ecosystem i mean and it's only getting stronger like now you can just on your macbook pro pull up iphone mirroring and just use your phone on your computer and stuff. And so the integration between the ecosystem, Apple knows that that's a strength that it's like, oh yeah, the new Android phone might have a better camera, but are you

willing to give up full integration with everything else you have? Like dripping out one device now is a $20,000 proposition for many people. It's not just, it's not just $1,000. So, okay, suite of devices. So you think there will be a rectangular, a black mirror? I don't think there will be a black mirror. I think they're, I think they're smart enough not to go head to head with the phone. I do think, you know, maybe it starts with a puck that's tied to like productivity, right? So something that just tries to get you interacting with chat GBT faster. So I can take it with me and it also sits on my desk when I'm working. And it's just the fastest way to sort of like, that was easy, you know, a staple style thing. Yeah, it's interesting that the workflow when you're

in conversation with a friend where I'm like, John, what was that company like a year ago that was doing this thing and you're like, oh, I kind of query you and you bring it back. What do you think about foldables? It feels like I've seen some demos out of China where the tech is getting pretty solid. The hinges are getting smaller and smaller. And you could imagine we're joking about, I know where you're going with is you want a newspaper size foldable that you can just. Let's make a newspaper. But, but I mean, seriously, you could imagine a situation where It's like, really what I want is if I'm going to be on, I don't want to carry a laptop because then I need a backpack.

I can't put an iPad in my pocket, but I can put something that folds up into the phone format. And then it's a bigger decision. It's like, am I just going to do the earpiece for the little random chats with chat GPT? But then if I want to go full in and watch a movie, I have a device that can really, really superpower. And it's like kind of counter positioned against Apple a little bit. We've heard rumors that Apple's maybe thinking about folding, but it seems like that's one of the technologies that you would be looking at. But what's your take on foldables generally?

Yeah, I think foldables actually squarely land in like the Apple value system. You know, Apple really values the sort of like high definitionness of their rectangles, you know. And so I think if you're trying to compete on high definition of these rectangles, you're probably in a losing battle. I think, you know, something I talk about a lot when it comes to hardware is like, your values have to zag away from Apple as much as possible, right? So, like, I think Avi Shiffman is a really good example of this. Like, Apple can't launch a puck that listens to you 24-7 because the conspiracy about Apple already and a lot of the tech giants that they listen to you 24-7. But you can't do the literal thing, right? At least not for some time.

And so Avi has a little bit of, like, narrative runway. because that value is so separate from like the Apple ecosystem. So I think similarly like Open AI and, you know, my assumption is that Johnny's aware of it, you know, they have to zag from a values perspective. What do you think their confidence level is, right? Like the most high profile example lately is humane, which is a talented team, maybe not as talented.

But they did have, you know, the full backing of the SV hype machine in many ways. A lot of investors were quick to point out that they weren't in it and that they thought it was silly, but it was like a moonshot new hardware attempt with a bunch of, you know, very talented people. Do you think that that is in the back of their mind at all in terms of... With projection specifically? No, no, just not even about the form factor, but just like, hey, this is actually like,

yeah, it's cool to have billions of dollars in the world's best team. but it's also Apple was started in a garage, right? Something closer to Avi's environment right now where he's like, you know, on a motorcycle, like, you know, just like hair in the wind, just like thinking about the next computing interface. Yeah. And it's also, I feel like the pressure to deliver value beyond just companionship, right? Like Avi's edge is like, I just want it to be your friend. I don't need it to be. And yeah, maybe it can do other stuff over time, but. Yeah. Yeah. So I think on the humane thing, you know, I got to know a bunch of folks over there.

And I think my main critique is that, you know, they were very like Catholic guilt ridden about the iPhone. Like the entire animating spirit was like, we feel, we kind of feel guilty about the iPhone. And now we have to present this alternative so that that's not the only road that we go down. I think unfortunately, like, they made a few sort of like key errors. Like I think the projection was sort of like a really key error. I think they like backpedaled away from replacing the phone and then it was supposed to be just like part of like your broader ecosystem. So I think it's what's really clear already, right, is that they open AI and Johnny are saying we don't want to replace the phone. We want to be this third thing.

But we do want to live in your pocket. Right. So it's like close, close enough, but far enough. And I think that, you know, what's funny is Sam, you know, back to Maine early. And they had an open AI partnership. And I think that I don't know the full politics there, but I would assume at some point, you know, Sam sort of maybe lost face and faith, sorry, not lost faith, and then kind of turn to, okay, well, who are the other superstars of Apple?

Because like make no mistake, like Imron and folks out on that team were like these superstars at Apple, right? It's just two different sides. Like Imron was the HCI team and Johnny was the industrial design team, you know? So yeah, I think that's sort of, I think they're aware of those lessons. One thing we were talking about was in terms of building like a super high performing hardware device consumer electronics team. Steve Jobs not only had Johnny Ive, but he also had Tim Cook to build the supply chain. And building robust supply chains was difficult in the 90s and 2000s when Apple did it, and it didn't exist in China. And now because of geopolitical considerations, it's also very difficult.

It doesn't feel turnkey. It feels like the turnkey era of consumer hardware was maybe the 2010s. And now we are post that and there are lots of different considerations around tariffs and trade policy and geopolitical dynamics. Do you think that we would see someone like a Tim Cook go into Open AI to get this off the ground? Yeah, that's a good question. So some things that I've been just sort of like watching is that Open AI already has like an internal hardware team. And they've had quite some time. They have someone that's like head of robotics.

They have, I think they hired a woman recently that's head of like all hardware. And so it's clear that they've been building up this capacity internally for quite some time. And so I think, you know, the joint venture is definitely some form of appeal to authority, right, with Johnny. And my personal hot take is I don't know how sensitive they are to actually shipping. or at least shipping on the time that they, you know, stated. You know, like the hot take, the extreme hot take would be, right? It's like, Sam needs another, you know, 200 billion. And he, you know, he goes to, like, Saudi Arabia with Johnny.

And Johnny puts on his, like, British accent. And, you know, they don't have to even talk about the next model or, like, numbers, right? It's simply, when's the device shipping? And they could just say a number or, like, say a date and they can get the money, right? Yeah, if you're building up the valuation, you're saying there's a 10% chance that we disrupt Google. That's a multi-trillion dollar business. So you can value us at multiple hundred millions.

There's also a 10% chance that we disrupt Apple. And that's a multi-trillion dollar business. So you discount the value and that's on a couple hundred billion. And so, yeah, you kind of add. And then that question doesn't matter. Exactly. You know, like then that supply chain question doesn't matter when I have 200 billion in cash. It's like, what supply chain?

Buy Foxcon. I mean, it's not beyond it. I think so. We menu back your apple now. So, yeah. I mean, what, we're going to bake, we're going to bake the new model into into. I've seen, I've seen some hot takes back and forth. Johnny Ive was obviously deeply loved and admired. And I kind of took the internet by storm with the conversation with Patrick Collison last week.

At the same time, a lot of people, I've said that a lot of the work that he did was iterative. and not that revolutionary. What is your take on the legacy of Johnny Ive? You know, he's in his 60s. That's sometimes retirement age. I believe he's in his 60s. I actually don't know. But I mean, at the same time, Berkshire Hathaway just hired a 62-year-old CEO.

And Warren Buffett worked into his 90s. So what is your overall take on Johnny Ives' legacy and where it goes from here? Yeah, I mean, you know, I think the early part of his legacy, or like the Apple part of his legacy is like undeniable, right? One interesting thing to like in Johnny history is, you know, the first place he was working, I believe, like the design firm specialized in bathrooms. And so when you understand that, you can actually understand like his obsession with smoothness, his obsession with like, sort of like knobs and dials and all of these things that we see that are kind of orthogonal to like computer hardware, as well as just like things that are unobtrusive, right? and like brushed aluminum. Like you kind of understand his design language

when you know where he started. So then you get to Apple and he's had very clear. Started from the toilet. Yeah. But you know in a lot of ways like these devices sort of, they blend in similar to like a home object and they don't really stand out like what we understood consumer electronics to be.

So then you start fast forwarding and you're saying, okay, now he has like all the devices in play. What does he do? my hot take is that like he kind of gives technology a type of like body dysmorphia right like all he cares about is thinness he even when he gets control of the whole design team that's when we get the pivot to flat design so he even takes like the interface level and makes it thin and flat and not too like fun wait so was the was the camera was the camera is that just to sell more cases what's up with that. I think the camera bump just became a trade-off post-steve. Because, again, think we still have to

understand like Apple the machine, right? When Steve was there, really the only person you had to appeal to was Steve. So there's like a very known thing around like Apple demo culture where like it's Steve in a room with a couple other people. You go in, you show it to him and he says yes or no. And if he says yes, it's in. And there's no one else you're convincing. My assumption is post-Steve, you're going through stakeholder management, right? So Johnny doesn't have like the number one authority anymore. So he starts having to like wrestle with, you know, the camera team and he's wrestling with Tim and he's wrestling with manufacturing. And he just doesn't get to win every battle anymore.

That's my. Yeah, yeah. Seems like you good. What do you think's going on with Applevision Pro? Have you used it? It feels like it's built on some of the. the design language that Johnny Ive pioneered, aluminum, I suppose. But at the same time, not the thinnest VR headset.

Johnny left in 2019. They've obviously been working on Apple Vision Pro for longer than that. But it feels like maybe the first Apple product that doesn't have his mark on it. But overall, do you think that's going to get any better? Is there any salvage that can do? Have you used it? What's your reaction to just VR broadly, any of that stuff that you're thinking through? Yeah.

So I have one. I used it. I was very curious. It is really heavy. Yeah. I think they had the wrong guiding metaphor at the end of the day. They didn't have like a true creative direction anchor. And so they landed on, we're going to put a floating iPad on your face.

And it's going to be somewhat anti-social. Right. So everything that they show is widgets and air. and then you sitting alone on your couch watching something, an extreme high definition. And that's fine. But as like a counter position, it's like what if their anchor was like the iPod, right?

And maybe their like hero product was actually something around Apple music. And it looked more like the dancing silhouettes that when you put this headset on, your entire apartment actually turns into this beautiful color scape and you are animated. Right. So they chose like a stagnant couch potato instead. of you being animated. And we're seeing the same thing in Apple advertising when it comes to the way they talk about Apple intelligence, right?

It's like this kind of like slobby guy at his desk, getting AI to write an email to his boss to slack off further. So we have this company that used to champion, like you're going to be your best self with these products. And now it's just turning into- Yeah. Yeah. Go on the headset and slob out.

Don't do your work. Get AI to tell your boss why you're going to be late. And it's just like, it's, like Steve's nightmare, you know. That's funny. Yeah. I mean, what about glasses as a form factor?

I'm bullish on glasses as a form factor. And I think it's going to go a few different ways. I do think like the meta raybans, like that form factor will continue to get better. Eventually like what it takes to put projections and those will be, you know, lightweight enough. You know, I think the Orion, I think was the project on meta. You know, I think obviously that's like too bulky. Zuck looks silly.

But I think they're going to just get it thinner and get it thinner. And like 10 years, it'll be great. Maybe five. Also, I mean, with the, with the XR stuff and the AR stuff, like X-Real and N-real air, this company does glasses that it doesn't do all the crazy tracking, but it can pull up a visual in front of you just statically. And that solves like 90% of the use case for this. If I'm on a plane and I'm just going to watch a movie, I think that that's going to get there

first. And then you have to figure out, can we actually? actually make it fully holographic, but if you just want a basic HUD, like a heads-up display, that's a lot more achievable. None of the big tech companies have really said, like, we're going to settle for that. But I think that that might be a path along the road, actually. But I haven't really played with it enough. Staying on Applevision Pro, I heard a rumor.

I don't know how true this is that the person, one of the key people at Apple in charge of it was an ex-Dolby engineer who was all about the Dolby Cinema Theater. and I thought that like that was the killer app. Dolby's great. I love the movies. The sound like to sound like. But but like I tried wrong creative anchor. Yeah, yeah.

But I tried the dinosaur experience. It was like two minutes. I'm like, I'm not going to game on this thing. I'm not really going to work on this thing. I'm not going to work on this thing. But I did watch. I watched all of Citizen Kane actually in in Vision Pro.

and it was amazing. And it was like being in a theater and it was fantastic. And I almost think that they should have gone narrower. And maybe the iPod is the right thing. Like the iPod, when you put in the headphones, it is isolating. But the campaign around that was you dancing around in silhouette is beautiful.

And there is something about like restoring this idea of going to the movies. But I agree with you that the ads were super dystopian. There's that one where the guy is clearly a dad and he's watching a video of his kids and it looks like he's divorced and like his kids are super depressing. But giving blood on the track, Bob Dylan, divorce. Yeah, it was very rough. But what is your take on this idea of like Apple narrowing the focus of vision to we have one killer use case? Because they're kind of the everything company.

The iPhone will call you an Uber, does your email, does phone, everything. You know, the original iPhone keynote was it just does three things. And then it grew out from there. Do they even have the culture to refocus around one value property? I always thought, like, they have Apple TV. They never made a real TV. They could have just positioned this as like, this is the best way to watch Apple TV. And it's just an Apple TV machine.

And then it has some other stuff. And yeah, we'll have an app store, but really. But what do you think about that culturally at Apple and just in terms of like, if you were running the place, what would you do? Yeah. Yeah, that's a great question. You know, I think it's so hard to understand why certain things in your own history worked out the way that they did. Right. I think we are so anchored on, like you said, the iPhone being this giant enablement to not only Apple, but a lot of, I mean, how much of the valley is built on just like mobile apps that did a thing, right?

And that's because if you're Uber, you know, you have GPS and now like you can actually do all these things, right? So I think Apple wanted to be hyper enablement ready again. You know, they wanted developers to really build things on this. And I just, you know, I think unfortunately at the premium price, unlike like the Quest, it's really hard to bootstrap that and get consumer adoption. And because they also have never done just like a developer-only product before. So there were just so many conflicting narratives. Again, it just seemed like no one was actually gripping on to the reins.

So I'm very pro like pick a singular use case that's going to be incredible at, you know, like I think Steve's original thing was like iPod for the eyes. You know, it's like noise cancellation for the eyes. And so you have Apple TV. You have Apple Fitness. This is like we're doing these two things. Make it super lightweight. Make it comfortable as much as possible. And maybe as a bonus you can like tie the laptop to it so you have like a keyboard in front of you.

Right. But this is about being static and XYZ. Have you ever thought of applying for a job at Apple? Maybe not CEO job. I don't think they have a listing, but trying to fix Apple, get it back to its roots. I think you need a lot of political capital to do that there, is my assumption. And so I'm far more attracted to the things I see getting built from like my peers. Yeah.

Let's switch gears for a second. I mean, it's just fascinating because it is a, it's like almost like a supply chain. story in the sense that they were able to pull forward the highest resolution display that no one else had access to. Not even Mark Zuckerberg spending $10 billion a year on the Metaverse could get that screen pulled forward by two years. I'm sure the next quest will have a comparable screen, but Apple was able to do that. But that's not enough. It's like it's a fantastic execution on the supply chain side. Anyway, Jordi, where do you want to go next?

I wanted to ask what you're up to. I know you I'm sure you're cooking I know you're cooking Anything you can anything you can share right now? Yeah um so yeah I mean I'm coming out of you know sort of an acquisition of eternal and just thinking about what's next and one one thing that I've been working on my friends throughout the past year is actually speaking of hardware this oh no the background look at what I can see paper I see paper hardware, underrated form factor for hardware. Yeah. It's a Wall Street Journal competitor. How do you do the

Oh, there it is. That's not that either. How do you do these things? Oh, there it is. Okay. So why would you want to hide that background? Why would you want to hide that background? That's a way better background. Listen, I love San Francisco. It's the best thing in the world.

Yeah, it's very appropriate. Says the New Yorker. Stolen valor on that Golden Gate Bridge. Boom. Boom. Yeah. Brutal. Brutal.

Show us. Yeah, what you were going on. Yeah. So the big project I've been doing and launching Technology Brothers exclusive

is this hardware book. Let's go. Let's go. Let's go. Let's go. Let's go. Let's go. I made my friend,

Julian Davis and and Charlene Dang. And, you know, So I saw sort of a few years ago that hardware was really going to become the new meta. It's sort of, you know, we saw software shifting into these two very big ways. And we started to see the containers for software shifting as well. Those containers being sort of like spatial and hardware.

And so last year just started to feel like a really special time in hardware, humane, rabbit, all these things with teenage engineering, daylight, USB club. And so we really wanted to give sort of. of like the art book treatment. This is my personal copy. The real copy is going to be hardcover. But we really want to give the sort of art book treatment to what's happening in hardware. Daylight. Let's go. Yeah, daylight is beautiful. You know, while writing the like humane sort of blurb, they got acquired by HP. So these things were, you know, really shifting in real time, which was so crazy. One of the big, big.

sort of sells on this book is a exclusive interview with jesper the founder of teenage engineering who maybe folks remember from his config talk last year and so it's 260 pages of giving like present day technology uh that art book treatment so really really excited about that amazing when can we buy it you can buy it you can buy it right now Hardwarebook, 24.com. Hardwarebook, 2024.com. Let's go. Fantastic.

Amazing. Amazing. I'm excited. And the site's beautiful, by the way. Very cool. Thank you. That was just a late night. One late night coffee slash.

Vibe coding. All right. Yeah. Some might say. All right. I just hit pay now. Let's do it. That was easy.

There we go. There we go. That's amazing. What's next? Are you already working on the next? book you're already cooking hardware 2025 it's not going now it's unfolding it's it is unfolding um you know I would I would love to sort of get sucked into the open AI hardware tentacles and uh you know maybe even like write some open AI checks to emergent hardware I think that we're going to see a lot

of form factors emerge you know beyond just pendants beyond just like pucks what have you um I've been investing in like advising some new hardware companies and AI hardware companies. So, um, yeah, I'm just, I'm, have you gotten a chance to play with the chromatic from Palmer Lucky? No, it's sold out before I could buy. Oh, it's sold out. Yeah. Yeah. Uh, yeah, I, I love it. It's, uh, I mean, it's a beautiful product. You can run over it with a car. Apparently, I haven't tried that, but also it, it's, it's like this perfect throwback, but you can use emulators to play any game and now they just added Twitch streaming from it over USBC, which is crazy. So you can so you can plug it in and it records what you're what you're playing.

So it's like this. And can it control an end of your old drone? One day, I'm sure. And I think that they might be working on an N64 and some other some other retro hardware. So it's fun. A lot of the retro throwback stuff coming through. I mean, the only other spicy take I want is on Memoji, but do you have time for that? Is Memoji underrated?

Are you using it every day? No, I don't. I don't drive. No, no, sorry. Damn. What do you think? Are you talking about Memoji or Gen Moji? Oh, it's Gen Moji.

Oh, yeah, Memoji is the one where you make yourself into an emoji. Gen Moji is the one where you generate anything. Yeah. Yeah, I think that, you know, I think the worst, I don't think Gen Moji is actually a bad product. I just think like the ad campaign, again, like they've lost the poll when it comes to communication. Yeah. The entire thing about Gen Moji, it's the experience of making a niche reference for

your friend. And the ad campaign is just like very weird looking, like end results. You know, like the entire thing about that is like it's about the process, not the end result. Like the end result out of context makes zero sense. And then that's all they paste it around like major cities. And it's just like, God. I mean, how do you even tell that story is like should the billboard be like, two friends like in separate rooms kind of like like like mean girls split screen like laughing

because they're sending each other gen mojis that's that's not a that's not a billboard campaign for the first first level of like taste is that a billboard campaign yep no i mean it's funny because it should be such it should be a product that needs no billboard right like it it should be inherently so viral and so user generated that people just start using it right and it has this sort of ghibly moment where people are just using gen moji and i just didn't see that at all maybe maybe i'm a boomer yeah um anyways thank you so much for coming on we've been wanting to do this for a while this was great timing uh excited to get a copy of the book we will we'll show it back here on the show and i'm sure have some follow-up questions but um congratulations on the launch and

uh welcome back anytime for more more hardware hot takes oh fantastic Thank you, brothers. Play them out for the studio audience. Thanks for coming on the show. Legend. Chief Reggie officer of Hardware Book 2024. We'll talk to you soon, Reggie. I'll be right back.

How'd you sleep last night? You know how I slept, John. Okay. I should just give you access. I should just give you my login to just motivate you because... I got a 93. I'm climbing up. I got a 99, John.

You got a 99. Well, I beat your Monday, Tuesday, you beat me, Wednesday, Thursday. You get out of here. I'm going to do some ad reads. Go get a pod 5 Ultra at 8Sleep.com. They have a five-year warranty, 30-night risk-free trial, free returns and free shipping. These are nights that fuel your best days, clinically backed sleep fitness. We also have a coupon code, TBPN at 8Sleep.com. Go check it out. We also want you to tell you about adquick.com, out-of-home advertising, made easy and measurable. Say goodbye to the headaches of out-of-home advertising. Only AdQuick combines technology, out-of-home expertise,

and data to enable efficient, seamless ad buying across the globe. We, of course, have Wander, find your happy place, book a wander with inspiring views, hotel-grade amenities, dreamy beds, top-tier cleaning, and 24-7 concierge service. It's a vacation home but better. And lastly, Bezell, your Bezell concierge is now available to source any watch on the planet for you. Seriously, any watch. They've over 26,000 luxury watches on Bezell now. So go to getbezzle.com. we should go through some more timeline posts there's been a bunch of stuff in the news what else is here uh uh l3 tweet engineer just as lmao in post a screenshot uh probably from blind or something google prestige is gone i was at a dinner party when i mentioned in passing that i

worked at google and everyone immediately burst into laughter pointing out how chat gpti had eaten our lunch then they pulled out their iPhones and showed that they didn't have a single google app installed and had set their homepage and default search engine to chat GPT, then they GPTed Google stock chart and erupted into another round of abhorious laughter while making the pinching hand gesture. Very, very fake. Extremely fake. You cannot get off the Google ecosystem, but very, very funny that somebody posted this. Very silly. Also, if they pulled up the Google stock chart, they would see that it's up like 10% in the last two days. So very silly. We also have the Teal Fellows.

We're announced the 2025 Teal Fellows. We're going to do Teal Fellow Day, hopefully, on Tuesday. And we're going to interview as many of these folks as we can, give you a little tour of what the latest batch of Teal Fellows are working on. There's a bank, a trading platform, a research lab, somebody working on non-invasive neurostimulators, foundation models to power digital humans that are indistinguishable from real ones, AI-powered solution to automate manufacturing processes

and dark factories. That's cool. Fizz is developing an AI financial investor. We have friends who've been on the show, are now Tiofellas, Saur and Monroe Anderson, the founder of Neros, is a defense technology company that builds drones critical to the modern arsenal at massive scale. Orbit is a non-invasive neurostimulator.

There's a bunch of other folks in here. Interfaces developing human native communication devices that change the way we work and think. That sounds maybe hardware-driven. I'm excited to talk to him about that. And I am working on a surprise guest, Ben Heilick, come on later today for 10 minutes at 2. That'd be great. X is so cooked right now.

It's extremely cooked. I can't even get DMs out. I'm trying to get his email for the invite. But he had something interesting earlier. Ben said, an AI alignment researcher at Anthropic just said that Claude Opus will call the police or lock you out of your computer if it detects you doing something illegal. I will never give this model access to my computer. And so Sam, I guess Sam Bowman over at Anthropic and a now deleted post says, if it thinks you're doing something egregiously immoral, for example, like faking data in a pharmaceutical trial,

it will use command line tools to contact the press, contact regulators, try to lock you out of relevant systems or all the above. That is absolutely crazy. But we'll see if it's real or fake news. And let us bring in our next guess. We have Keith Rube. Welcome to the stream, Keith. Hey. How are you doing?

There he is. We have a sound for now. Welcome to the show. Great to have you. Yeah, yeah, great to have you back. You got to work on the entry song. Yeah, it's a little messy. We're figuring it out.

Give us the recap on the KV Summit. How to go? Yeah, so every year, for the last 15 years, we've hosted all of our CEOs, like 150 plus CEOs at Kvala Point in San Francisco. And it's like a two-day program. where we bring it inspirational and actionable speakers. So, for example, I interviewed Eric from Ramp, John Carlson, Stripe. John Carlson interviewed Bill Gates, actually, and Sam Altman spoke, and then the CEO of Databricks.

So we have a wide variety of content, some of its more practical, tangible, and some of its very aspirational, inspirational. So we take our CEO time very seriously and want to make sure that every minute there's, or not in the office, you know, managing, executing, you know, growing their companies that they get, you know, disproportionate returns out of the time. What was any type of surprising takeaways or insights that you can share? I want to know what the biggest debate points are. Like what, where are the things where people are actually debating? I've, you know, Sam Allman recently said AI could be an operating system. there's questions about foundation model layer value accrual versus application model layer value accrual.

How many winners there will be in the coding market? Like what are people actually debating and unclear about? And then I'd love to know where you stand on it, obviously. But like the non-consensus topics are probably the most interesting to dig into here. Sure. We can talk about all of this. And generally speaking, we will post all of these videos online. So the vast library of, you know, 15 years.

virtually anybody who's been successful in tech has spoken at our conference. And so, in fact, Sam's probably spoken four or five times now. It was interesting. I compared when I was interviewing John Carlson, I'd interviewed Patrick Carlson literally a decade ago. It was an interesting contrast to show the arc of the company. When I interviewed Patrick in 2015, Stripe was worth, you know, several billion dollars, but it only had 250 employees. And now, you know, as John mentioned, they have 9,000.

Nine thousand. Wow. Yeah. So it shows you the arc of, you know, building a company that a decade had passed since I last interviewed Patrick. So we will share 90 plus percent of these videos. We'll start posting them next week. There's maybe one or two. It's like to speaker. You know, they don't want to share or some things.

But the contrast I noticed one, and one's one about building a company. We talked about founder mode a fair amount. So, you know, Eric talked about how they build ramp. And John gave a slightly different perspective on building strike. And then I interviewed actually Gary Tanna-YC to talk about the origins of founder mode, the lessons from, you know, thousands of companies that, you know, they've invested in for 20 years. And so I think the operating styles and approaches to hiring senior talent or not was an interesting contrast on AI.

We had Brad Gerster, Altimeter, speak, and talked about, about the future of AI is applied to pre-existing companies. He was quite inspirational. He made a couple points that I think Sam generally would have agreed with, which is there's only going to be one foundation model, research lab oriented successful winner. It's certainly not something that VC should be investing in. The application layer, I think there's more room and opportunity

or something that transcends what we think of research labs today, Just like Open AI and chat GPT is transcending what we think of search and to some extent what we think of social. I think the next generation will be something that doesn't look anything like at LLM, doesn't look like a research lab. And it's very difficult therefore to find. But hopefully some founders have visions and hopefully they call me women they didn't. Yeah. I feel like Brian Chesky was a big inspiration for the original founder mode essay by Paul Graham. he just went extreme founder mode a week or two ago with the Airbnb reimagining.

At the same time, Ben Thompson's kind of analyzing that. And Ben Thompson's argument is that Airbnb is a platform and a marketplace company. And so the economics of the platform kind of dictate a certain user experience, mainly that people open the app. Yeah, they have hundreds of millions of users. The user base is massive at Airbnb, but the average user opens it like twice a year. And they want to get in, get out, and it's not a daily active user experience. And so Ben Thompson was saying there will be some challenges.

What was your reaction to how Brian positioned Airbnb's next chapter? Do you like that communication style? Are you optimistic about where Airbnb goes? Give us your reaction to the Airbnb news. So yes, Brian pioneered what we now describe as founder mode, even though it's very specific. in some ways the operating style of Apple. But in any event, he triggered this at a Y.C. Alumni dinner in Napa Valley where he kind of off the cuff without preparation,

spoke for two and a half hours. And that's what led to Paul Graham summarizing and distilling it, which led to a lot of intriguing and interest. And this is kind of what I think CEOs were talking about privately, but no one had really stitched it together and kind of created a public discussion of it. I for those to be really interested in the topic I interviewed Brian a few months ago at Graham's offices and that video is available online where Brian is incredibly articulate and eloquent about the virtues and the benefits of founder mode and how we

actually implemented it so I highly recommend anybody in the audience who's intrigued watch the actual video because I will not do injustice. On the specific topic of Airbnb experience, is I think Ben made some interesting points. I've been a fan and early reader of Stretakery. It's the only thing I read regularly in all of tech, because I think he's the only one that actually understands tech that writes.

So I actually have been recommending this to have friends of mine for over a decade who want to enter into tech. It's like start reading the techery. So I think the point he makes, first of all, is that in services world, you're going to potentially want to get off the platform once you find a service provider you know, that you like, I think he missed one subtle point, which is Airbnb is still designed mostly for travelers, I, you know, I'm going from New York, let's say, to Cincinnati.

And obviously the canonical case is I would find a place to stay in Cincinnati. Great. But as you point out, the average user, you know, might travel 2.5 times, 2.4 or 2.5 times a year, so it's hard to turn my Airbnb usage into a daily, weekly, monthly habit. However, the point that Ben glossed over a bit is that I'm traveling to Cincinnati and then to SF and then to Dallas and then to somewhere in Utah, I have no incentive to go off the platform to find a service provider. Yes, in a city that I mean regularly, let's say New York City, if I found like a hairstylist

here or the equivalent, I may have an economic motive to just, you know, disintermediate the platform. But if Brian's right that curating experiences in search. services that a local would use still appeals to people in New York City, which is a huge a huge number of people come to New York City every year. Then I think Ben may be wrong and Ryan may be right. So I think Ryan is probably sort of dialing in to the traveler visitor use case. And then you need the breadth of services when you're traveling and experiences.

You don't necessarily have to only use it in your home. city, which is where Ben may have some valid critiques. Yeah, yeah, it seems like the dog walker example is probably the rough one, because it's the most obvious disintermediation point. At the same time, yeah, there's plenty of times when big group is traveling and they want, you know, private chef to come, cook for everyone, something like that, makes a ton of sense in that context. And then also, it's, like, you can build a good business in a referral business that does not

less, it's less transaction. It's more discovery and advertising based. this. I was actually, like, not even half joking, just saying that maybe Airbnb needs an advertising product because Uber's been so successful in advertising. I don't, I mean, I think that that might really frustrate people if they're seeing promoted listings, but at the same time, like on these marketplaces, usually advertising does pretty well. I don't know if you have a reaction to that. Yeah, I have two reactions. First of all, I don't think out at Airbnb will need it. Airbnb's margins are like roughly, take rates roughly 13%. Sure. And, you know, versus like the

Uber's or the Instacart's where they really don't make too much money for transaction. Got it. If you don't have a highly profitable advertising revenue, it's hard to build an interesting business. So I think Airbnb doesn't have the economic drivers that you see with Instacart and where it has been successful, but it is something annoying. That said, you know, Google used to run this internal study where they would actually show users who are exposed to SEO content and paid advertising are more satisfied

then users who are searching in only CSEO content. So advertising when done properly can be a value ad. If you think about it, the economic incentive of drafting content that explains, contextualizes, and markets, products does drive to efficient advertising that can be an important for a user. Even today, if you're in an area and some service provider is running ads, it shows a level of sophistication and professionalism that might associate in some ways with just a better end product experience than somebody that doesn't know how to set up a Google ad. I wanted to ask,

Ben had some great thoughts around, you know, the entire internet is based on, you know, advertising. And in a time when, you know, we may all have agents that are just trawling the internet for us and acting on our behalf, the, you know, economic, the existing economic model of the internet could hit some snags. And he makes a case for, you know, micro payments to allow people that produce content to get some benefit from having models, you know, acquiring information on those sites. He also outlines how this is going to be incredibly difficult to pull off because there's so many different groups and incentives. I'm curious if you've spent much time yourself thinking about the potential for micro transactions in the context of replacing lost advertising

revenue. So similar observation, I think the post, the original sin of the internet by then is totally worth reading. And I generally agree with the point that the original sin to the internet was that all content should be free. I actually made that point to Jessica, a lesson when she was launching the information in like 2005, six, seven, eight, whatever it was. I was like, you definitely are going to be on the right side of history by charging a premium subscription.

Ran into Dick Costello at that party, the launch event, and I actually used the term, the original sin of the internet was that we all learn the level of less than that because the marginal cost of content, marginal cost of distribution of content we'd go to zero. You should charge zero, which is absolutely false. However, micro payments make no sense. Ben is totally wrong about this. I'll give you an old lesson from PayPal. So one of my jobs at PayPal between 2000 and 2003 was to find new markets because we're really dependent on eBay. And that led to, you know, risk. And we ultimately went public and then sold the company eBay because we were nervous about the risk of eBay handicapping our future growth.

And as you know, tech companies are valued by the next 20 years of cash flow. So Peter and Roll-up and various people were very nervous about that. One of the markets I wanted to explore was quote unquote micro payments and Peter Thiel made a very astute sort of rebuke to me. That's the dumbest argument ever basically because content has a marginal cost. It has a gross margin of about 99%. So basically whether or not micro payments cost you 5, 10, 15, 20, 25, 30, 35, 40, 50% of the content. Who cares? Like if you're writing content that people pay for because your gross margin is like in the

90s, you shouldn't really worry about whether the micropayment cost is too large. Now, the friction to make it, the reason why I wanted to build the product was I think the friction of micropayments is too high, like the consumer friction of paying for anything. Like you know, you're going to be a short blog post, you're going to be a short post, you don't really know exactly what it's worth. And so if you have to do anything other than just one click. forget about it. But I still think people get this analysis wrong that it's not an economic problem for, now music's a little different, by the way, where you owe a license owner a reasonable

fraction of the transaction, then you can't be so, you know, sort of whimsical about the marginal cost of payment. But most people do not really, you know, have to pay out a meaningful fraction of that dollar transaction to somebody else, some third party. So I think this is, Excuse, by the way, Stripe originally started when Patrick John pitched me on Stripe in 2010. They originally thought that the core use case was going to be like micro-payments in accelerating content. They still have the GDP of the Internet, but they really felt that it would unleash new content, which hasn't for the most part happened. You know, substack is an interesting business.

It has real GMB and an associated real revenue line. But I still think it's the wrong place for Ben to be focusing. Well, yeah, the issue with looking at Substack or even Ben and their success is there's some element there of, I just want to support this, the person creating it. I want them to not think about getting a job or building a business other than just writing online and spending all day thinking about in Ben's case tech. Whereas for the average site on the internet, the person that would be making the micro payment, And it's not like they don't have any incentives to be like, I care about this person and I want them to have a nice career.

I mean, with like the recipe example, I'm fine if I'm baking an apple pie, blending three different recipes together and giving me the LLM result of that. But if I want to listen to Michael Jackson's thriller, I don't want half of thriller and then half of another song kind of cluge together. Like I want that exact piece of intellectual property. And so there's a very traceable lineage of that data transferring. And so the micropayment makes,

a lot more sense when you're actually buying that specific thing. That's a specific thing, but I think you're better off having a differentiated voice and trying a premium for that. It's very difficult. Yeah, yeah, yeah. How many people in all the realms of content have a truly differentiated voice. Like in sports, you know, Bill Simmons in his prime definitely did. Despite all the other sports writers, Ben in tech, I still think as an incredibly differentiated

voice. It's just very difficult to achieve that. But if you do, you can absolutely charge a, premium. Yeah, totally. What's been your, it feels like it's been AI week. We had Microsoft build on Monday. I know you were tied up with the KV Summit, but Google I. Google I. Google I. Open AI bought I.O for 6.5 billion. Anthropics dropping Claude for today. Are you tracking these things? What stuck out to you? What made it above all the noise? What was the signal in your mind? What are you tracking? What are you, what's, as anything changed your mind this week?

Well, I think what was interesting to me because we do have the CEO summit annually. This year, the AI content, not just the AI content that we programmed, but the AI content in cocktail conversations, dinner conversations, was an order magnitude greater than last year. So the first derivative was off the chart on AI. Last year, there was like a vertical of AI and a vertical, a slice of conversation. And the people who are not in AI space were probably intrigued, but it wasn't infusing everything. I felt this year AI infused every conversation up and down the staff, so to speak, latterly across verticals, whether you're in financial services or labor marketplaces.

It didn't matter. Everybody wanted to learn more about AI, how to leverage AI, put on this bionic suit for their own business. Yeah. Are you seeing AI pop up in unexpected seed stage companies you're betting on? We were talking to T.J. Parker. He launched General Medicine today. And he was like, we made the deliberate decision not to give the user a text box. We're not a chat chit repor in that way.

But we're using LLMs all over the place to parse insurance policies and medical records. and this business is only possible because of AI. And I feel like that's going to be an ongoing narrative, but it should dominate the conversation. At the same time, it should melt into the background just like, oh, yeah, you're hosting your new company on the cloud. No big deal. Everyone is.

Yeah, I mean, at some point, and same thing like the adoption of, you know, some, yeah, I think that's a good metaphor, but maybe I'll give you a barometer. Please. Of my last 11 investments, six are AI-based, and that's up from zero, literally zero.

Yeah. And so, you know, if even people like me who had no exposure to AI before last year are now making a majority of my investments are AI forward, that says something, you know, pretty interesting. But aren't for the non-AI investments, aren't you still talking with those founders, CEOs about how they're going to leverage AI? I mean, it's not like they're coming to you and being like, yeah, I think it's overblown.

I'm short. You know, I'm not going to, you know, they're not like Luddites if they're technology entrepreneurs. Yeah. Well, there's a couple of use cases that are important to produce any company. One is just engineering productivity, period. You know, whether you use cursor or, you know, some other product or all the products. Like we've seen at Ramp, for example, engineering productivity is up 46%.

Pretty strictly measured and increasing. So that's just purely on productivity. So getting more done with less. Great. You're seeing AI remove costs, I think customer support is a classic use case. Customer support can be expensive, and in many cases,

AI can perform on par with a human or better, certainly faster and more cheaply with more scalability. So there are areas that are no-brainers for any business to adopt. Then there's the question, in the strategic area you compete in your company, is there a way to use AI, you know the differential way and that that's a little bit more art than science right now yeah how how how how overblown is the narrative of uh startups getting steamrolled by the hypers

oh every startup just got turned into a bullet point in a google presentation uh there's one narrative that's you know google invented a high they have all the tpUs they have this massive data center at the same time, a lot of their product rollouts, they sunset a lot of products, they run a lot of tests, they're really dominant in search and advertising, but if it's a side project, it kind of gets side project attention and it's not really a founder mode project. How are you thinking about how big tech is competing or just stepping back and just reaping the rewards of everything that they've done over the past couple decades? Yeah, I think there's like classic, you know, sort of conceptual approaches to this is, is this a sustaining innovation or disruptive

innovation? And, you know, things, things that are disruptive, there's a different, you know, approach to those, things that are sustaining, there's a different mentality towards those. I think the key is for a founder to understand, where's my comparative advantage, ultimately? Like, you're going to build a startup from scratch. The world is not your friend. And nurse is against you. Nobody cares. Period.

Nobody cares about you. So you've got to turn that inertia. into positive momentum in a true physics sense. And then you've got to figure out how you propel yourself, even if other people, large or small, start doing something similar to what you do. And I think having a lot of a site, like at least as an intellectual, conceptual level,

how would I do that? How am I going to do that? Makes the difference between being, you know, a very mediocre startup that is the proverbial roadkill to something that looks like it should be washed out in this wave, but actually trumps over the incumbent. And, you know, the history of tech is typically the well-run, thoughtful startups prevail over the large. Don't think about AI.

Open AI is absolutely dominating Google, period. Open AI, Open AI's chat GPT will be the default interface for the majority of people on the planet, not Google Search. Bold, big change. I want to talk about it with, with, we, we, we, we. With OpenAI, they acquired Johnny Ives company. They brought in an absolute industry legend. I want to talk about it in the macro, unless you have a take on this actual deal or Johnny I,

but more importantly, startups bringing in these absolutely legendary heavy hitters, it takes a bunch of different forms, right? Palmer Lucky brought in John Carmack, video game legend at Oculus. That was amazing. Ramp has Ken Chenult, the former CEO of Amex involved, but not as an employee, right? Then Ramp has Gannati, this fantastic programmer.

Like, I think he's the best programmer in the world, more in an employee role. And so as startups are scaling, should they be hunting for these, like, you know, absolutely legendary people trying to get them in their organization? Is it just a very rare scenario that opening eyes so. doing so many things that they needed Johnny Ive. How would you talk to a founder about hiring kind of an industry legend? On the industry legend side, I think it depends on the motivation, both for the company and the individual.

Many people become successful and are demotivated and not as ambitious. Then there's the eons of the world that the more success they have, the more ambition they have. And so I think you need to triangulate who are you talking to and how much drive they still have left. And then certainly from an employee perspective, in, you don't want like someone who's past their prime. It's a little bit like sports.

You do get past your prime or your motivation to drive to K's unless you're like the Elons or the Cobees and people like that that are very rare. So I think that's the most important sort. On the specifics here, I've read mixed reports on exactly, you know, what involvement if any, Johnny Ives going to have. They're acquiring the company, but it's not really clear to me what he personally is going to do. I think that will have to sort itself out.

Yeah, could wind up in like a CEO of hardware role very deeply involved or, you know, more of an advisor or design lead. There's a bunch of different ways that that could shape up. Jord, do you have anything else you want to chat about? How do you think the AI safety era of the, of 2023 or, you know, 2024 will be viewed in five years, 10 years? Well, it's DOA. I mean, nobody even talks about it anymore. What's the last time you've heard someone like the credibility talk about? Well, so I heard about it today.

I'm just going to call this out. So today, apparently, I'm seeing a screenshot. An AI alignment researcher and Anthropic just said that Claude Opus will call the police or lock you out of it of your computer if it detects you doing something illegal. And the researcher says if it thinks you're doing something egregiously immoral, like faking data in a pharmaceutical trial, it will use command line tools to contact the press, contact regulators. try to lock you out of relevant systems or all the above, which to me, to me is like, if that's AI safety, that's scary.

I don't think that's what most, I mean, that's scary. It seems pretty stupid, but there are some legal obligations, obviously. Sure, sure, sure, on any tech company, you know, Asian child pornography and various things. So like, assuming there are compliance obligations on all companies, whether tech or otherwise, and absolutely those are all real and serious. AI safety, I think, connotes to a lot of people like the idea of AI taking over, you know, doing things that are, you know, mischief creating in some way. That debate, I think, is over in the United States for a very long time.

We're in the, how do we accelerate the progress of AI? How do we ensure that AI in the Western world dominates over the CCP's use of AI? I think those are the most priorities, the major priorities. I don't think anybody in DC right now really wants to hear about this old. a AI safety canard, which is basically an anti-TAC. You know, is euphemistic for anti-TAC. It wasn't a serious concern. Yeah, it's interesting.

I don't know how closely you followed, like, the latest release of Lama 4 by META, but there was allegations of kind of fine-tuning on the specific evaluations and a delay. And a couple of years ago, it would have been so easy for them to just say, hey, we're worried about alignment. We're worried about safety. We're doing this for you. But instead, they kind of just had to bite the bullet and say, like, we kind of messed up on the training.

It's not quite done yet. We're just going to go back to the drawing board and work through it. And the fact that they couldn't even use it as a fake excuse was very telling to me that it is truly a dead conversation. Yeah, fortunately, I think it's dead. I agree. Great. But I think ultimately, the U.S. needs to prevail and succeed with AI, period. Yeah.

The flip side is that. So I think I've been aligned with you that the AI super doom fast takeoff. We all become paperclips never really resonated with me. I thought it was kind of just paranoia and like superstition. But now that I've seen what's happening with Deepseek, all of a sudden I've said to myself, like maybe we do need something that looks like an AI safety team to go in and investigate these models and see are they waited to shift us? the TikTok algorithm might be shifting us towards certain beliefs.

Like these tools are powerful and they can be fed false information or bad data to kind of steer the answers one way or another. And so maybe we do need some sort of model interpretability. And it doesn't, it's not exactly AI safety in the doom sense, but it's the type of work that you'd see done by an AI safety researcher. So geopolitically, how do you think things are shaking out? Oh, yeah, again, I would discriminate that from the gloom-and-dume AI. Totally, totally.

I definitely think models can be biased in different ways and manipulated by different people who have power over those models, period. And I think observability, which is actually cutting edge technically. There are some really cool startups that are focused on observability. It's a little bit like doing brain surgery and trying to figure out how to rearrange the brain. And it's the metaphors I think they use. And so that stuff's actually tracking really well and incredibly intriguing. Like why did the model do X and then we all to decompose that and arguably do perform surgery and fix it?

So I think maybe these models are biased. You know, the famous examples of, you know, asked for the benefits of Donald Trump and, you know, when I used to struggle with that, I don't know if it's improved. But, you know, those things are real. But I think that's a different set of challenges. And maybe the market sorts that out. maybe people expose it and the company has to address it. I think the TikTok manipulation is a very real thing,

which is another reason why I think it's better than American companies be in the forefront. But I do think that's a valid concern. It's different than what I think of safety as if there's a sentient AI that's going to take over with so much power that you can't really unplug the computer. You know, that I think that debate is pretty dead for a long time. It seems like tech and the U.S. government broadly, the new administration, has been kind of going all around the world, doing deals.

Are there any countries that you're particularly excited about America building partnership with on the tech side or even any opportunities for earlier stage startups to go plug into the global ecosystem? It's kind of a narrative violation because just a couple months ago we were saying, do not do business abroad. There's a 75% tariff on everything. But now you see Jensen and Sam and, you know, Scale AIs over in the UAE. It's more like we're exporting American AI versus import. Where are the opportunities? A couple drivers there is there's a belief, I guess, by some people that exporting

American AI makes it the default standard. It may or may not be true, but I think there is an ideology there. Second is I think a lot of this is we need resources, i.e. infinite power. And so people are going to very wealthy countries and asking them to spend a lot of money to either develop advanced power or advanced manufacturing capabilities at scale. And I think that's why you see the CEOs who run these large, you know, AI-based companies, you know, sort of traveling around the globe to sovereign nations that have too much money. They don't really know what to do with it. But I think that's a, it's not a due business in the country. it's more take advantage of their natural resources,

turn that into something that we need, which may be either more manufacturing or more energy or both. Very cool. Well, we'll lay it out of here. It's past two. Always a pleasure. Always a pleasure.

Such a great time. We'll talk to you, Keith. Great. Great to see you both. Take care. Have you going. Bye. Do we have Ben Highlock or should we talk?

He is in the waiting room. Oh, he's in the waiting room. Let's bring him in. Ben, welcome to the stream. How you doing? Ben, is your computer being? controlled by Anthropic. That is the question.

Have you been doing illegal things? What's going on? It's great to have you on. Everyone knows. Do the Jaguar rebrand. Yeah, he did the Jaguar rebrand. But today we're going to talk about... He did not do the Jaguar rebrand for those who are listening.

We're joking about that. I did it now. Well, if we say it enough, then ChatGPT will just pick it up. They've said it 10 times now. It can't still be funny. They might arrest you for that. Taking credit. for a brand you didn't build. Stolen valor. I don't think anyone's going to have a problem with that

the SWAT team kicks down the door. Get out of here. Anyways, great to have you on. Glad we could make this happen, even though X is absolutely cooked today. It is really good. I've never seen it quite this bad. It's rough. But anyways, why don't you introduce yourself and then we can talk about that post and kind of get your reaction to it live? Yeah. So I was a designer at Apple for like four years. Before that, I kind of have a weird background. I actually started off with like robotics and avionics. So like interned as basics a bunch of times and did those sort of those rounds. But now around a year ago, two years ago, started a company called Rain Drop.

And so we do essentially Century for AI agents. Cool. Awesome. Break that down more. Give us more context. Well, I mean, I think it's actually like it perfectly dumptails with this. right, which is like, we get to work with some of the coolest companies in the world, like Clay.com, just like companies that are just really trailblazing when it comes to

AI applications. And what we do is we help them find sort of like really hard to spot failure cases. So there's like people that do AI stuff, there's this concept of e-vows, which are almost like unit tests. So these are like, given these test cases, you know, does it pass them, does it fail them? What our product does is in the real world, like, what are users actual experiences with your product? Where is it actually failing? Where is it doing things you wouldn't expect? Right? So, for example, we had a one of our customer, one of our customers, Tolens.

They have this like alien companions with like tollens.com. Oh, yeah. Yeah, so you're familiar. And so they had an issue where like their aliens started referring to itself as like a dude from the United States, right? which is unusual. One of us. One of us. One of us.

One of us. Total cultural victory. And so their issues like, how big of an issue is this? Like was that one time or was this like hundreds or thousands of times? And then like if they're going to try to fix it, like it's very like random, right? It's very unpredictable. It's not always a clear root cause. So what they want to be able to do is kind of get a view of that issue over time now.

And if it ever progresses, it ever comes back. get notified. So we make it really easy for companies to like literally just the craziest issues. That's, yeah, it's just such a funny example of the aliens. Like, yeah, I'm actually. From America. America. But yeah, it's funny because if you fine tune that out, then you wind up with a potential

situation where the AI never adopts the personality or of an American or doesn't even know the concept of America anymore because you ripped it out too hard, right? So it's a really tricky balance to now, right? Yeah. And from like a detection standpoint, I think the thing that makes us kind of like, I think we have like actually one of the most complex like AI pipelines I've ever seen and just like research wise, a lot of really cool stuff cooking. Like being able to categorize these messages as it is an alien character talking like

misrepresenting himself as being from like the United States. It's very hard, right? It's not like show me events where the assistant talked about a guy from the United States. States, right? It's not some sort of like keyword search or something like that. Like some of these categories actually get pretty complex. And you can imagine that that is an issue that's only relevant to that app. So it has no application. Like so our customers have to be able to go in there and like define what they're looking

for and create this categorizer that can run on millions of messages a day very cheaply. Interesting. Yeah. Yeah. I mean even in like the web 1.0, web 2.0, like you're almost building like a pager duty. It used to just be, is this page 404ing? And the HTCCC spec.

There's no clicks. Yeah, there's nothing like that. It's so much more squishy. But, you know, the problems created by AI, also solved by AI. You said you have to run these queries. I imagine that they're LLM powered millions of times. That sounds really expensive. What are you doing?

Are you baking Lama 3 that's free onto an A6? So you can just like run it super cheaply. Are you looking at GROC with a Q or Cerebris? or something else that drops the inference cost? Are you on open router constantly trying to find the cheapest thing? Have you distilled models? Like, how do you control cost in that scenario? Yeah.

So because of a lot of our customers volumes, like if you think about like clay.com or something, like they're just, you have millions and millions of requests a day. So it's not actually feasible at all to send them to an LLM at every request. Yeah. We can use LLMs, like, smaller ones, like especially like Gemini Flash to like do things like summarizing or like describing clusters or stuff like that. As far as actual detection doesn't really work well for that. So we have a bunch of essentially custom trained embedding models and then models on top of those that are small.

They're like technically neuralettes, but you can think of them like an SVM or something where they're just like really good at doing a first pass detection. So around like usually 98% of the events or it can filter out like 95, 98% of the events that are not relevant. And then kind of the last pass or like is like a small. fine-tuned LM. What's it like training one of these smaller SVMs? Is this something like commodity Nvidia GPU for like an hour? Yeah, yeah, yeah. It's just like simple. The smallest ones you can do pretty fast embedding models are like longer. A lot of it's about having the data, honestly, having the right data to do it. And the cool thing is that and I kind of one of when I say like the pipelines complex, like one of the things I mean is that

our customer, each of our customers looks pretty different, what an issue is or is not is pretty different. So for example, you can imagine that one customer, like, if you have a customer support chat bot and you ask it to write code, you know, like, they might even want to flag that, you know, a user is asking them in the first place because that's like more like they're getting hacked or something. But a coding assistant, obviously, like, that's the bread and butter. So one getting refused is okay and the other one isn't. So we actually train these models on the fly for every single customer. So that's one of the things we do. four maybe more massive AI announcements this week. You got Microsoft Build, you got Google I.O,

you got OpenAI buying IO for $6.5 billion. Johnny, I've going into Open AI, building hardware. You also got Anthropic launching Claude 4 today. What stuck out to you? What was the most interesting story of the week? What has you thinking, I want to implement that? I want to play around with that. I'm excited about that or I have a hot take about that. Yeah, yeah. So I think that most of the times the things that excites me are the things that are for builders. Like we use AI a lot. All of our customers use AI. That's sort of the lens.

We're always looking at it through. I think that Google has really done a good job of nailing that niche of like building models for startups. I think that like Google Flash, for example, is the closest thing we have to like intelligence too cheap to meter. Obviously there's like four nano that was like a follow up to that. and like in response to that. But yeah, Flash is still just like superior in so many ways. They have just like throughput and just like the, you know, like they pretty much have no, you know, rate limiting at some point when you pay enough like pretty much zero. Yeah, because they're running on their own hardware down to the actual TPUs.

And so, you know, you have to imagine if you go to Open AI models, it's probably running on Azure. And so there's an extra layer of networking and transport and all the data's flowing back. So if you're looking at like super, super fast, fine-tuned responses, like Google infrastructure team is probably hard. Yeah, exactly. They know how to do it, right? Yeah. And so like, for example, like I think, I think diffusion is like super interesting.

I see a lot of applications for it. Like, have played around with a bunch. And there's something just wild about. I don't know if you guys are familiar with like the jump. Yeah. I saw one screenshot 900 tokens per second. Is it actually using a diffusion model instead of a transformer? Is that, is that what's going on?

Yeah, yeah, yeah, it is. Okay, wow. That's amazing. Yeah. Can you see like illustrations of it? It's essentially like actually like each part of the code. It's actually, you know, being generated separately and kind of literally like instead of just predicting the next word, next word, next word, next word is presenting like the entire result. Yeah. That's fascinating. I want to know more. I got to dig into like how big was the cluster that they trained on. Is it all synthetic data? I mean, they have so much data. That is fascinating. I had no idea that there was. I had no idea that there was. would be a flow back because we're seeing an images in chat chat. I know, I know. A flow forward, right? They're going transformers and then we're going back. But I guess it's like, these are great algorithms. Let's use them in every single way in every single application. And you'll probably see diffusion all over and transformers all over and everything.

And I think there's use cases where it works for. There's use cases where it doesn't. Like I think that code actually, I think lends itself pretty well to it in certain use cases. Like I think one of the key use cases they have in the demo that you can request access to is, like essentially it'll one shot a website like in like a second. You know what I mean? Especially build a calendar and it would just one shot. Like three seconds.

There's three seconds. Three thousand, three thousand tokens, I guess came out in three seconds. So it's like a thousand tokens a second. Yeah. It's just crazy. It's pretty there's something really insane about it. Like obviously like the apps that's actually generating are like not going to be as good yet. But like there's just something insane about going.

from like words to just like render. There's this idea I've always been really fascinated with, which is like, let's say for games, for example, like is there ever a point where you're, it's all just being generated. I think like gentlemen has said something along the same, the same line. Oh no, no, no, no.

Somebody's actually running Minecraft in a, I've tried that before. You've tried that one, right? I think etched is in partnership with that. They're building a chip that has a transformer architecture baked onto it in Silicon Fab by TSM. and they can run Minecraft, purely generative AI version. There's no game engine whatsoever.

It's just trained on Minecraft. Fascinating. Let's talk about the post today that you shared and is picking up steam, even though X is cooked. But I already read it. And we actually got Keith R. Boy's reaction, which I'll share as well, because I think it's an important context. But basically, AI alignment researcher at Anthropics said, if it thinks, you're doing something egregiously immoral, like faking data in a pharmaceutical trial, we use command line tools to contact the press, contact regulators, try to lock you out of relevant

systems or all the above. Keith made a good point, which is that technology companies have obligations to various authorities that if they detect illegal activity happening, that they have a responsibility. I think the concern, the obvious concern here is, you know, somebody's like playing around with a model and then suddenly it's contacting the New York Times and being like Ben Hylach is, you know, faking pharmaceutical data, you know, or something like that. And a hit piece comes out instantly. I mean, this hurts for me because I lie to chat GPT all the time. I'm always in there. I'm always in there saying, I am a train expert. Tell me about trains. I own dozens of trains just to get it to give me better, more rigorous responses and not talk to me

like I'm a casual train consumer. But if it finds out I'm lying, it's going to be over for me. Crazy. Yeah. I think, I mean, first of all, I don't have the full context of Tita's response, but, you know, it's important to note that this was not that. Like, this is not them saying that they have some sort of regulatory requirement, defining what that is and doing the minimum to meet that requirement.

I think that would be different. And then there's some sort of like, you know, someone could sue the U.S. government. There's some sort of path to recourse. So that's interesting and true. but it feels a little bit different. Could have been messaged better. Probably not in a random comment. Yeah.

You know, basically saying like, we're going to take control of your machine. Yeah. And to carry out something without any type of, you know, press too. Like, there's no legal requirement to go to the press.

I know. Exactly. Exactly. Exactly. That's particularly crazy. I totally get like, look, there's a law that says that if we see you doing wire fraud, we have to report it to the SEC. That makes perfect sense. Yeah, that's not, that's not what this is, right?

We're going to call TMZ. It's like, why. So, and they also just define it as something, you know, you know, egregious, evil, you know, which is, you know, morally, you know, immoral, right? Yeah, which is different than laws, right? It's different than laws. I think that like, also be clear, this is not, I had a pretty strong reaction to it. actually, I don't get that angry about things, but I actually felt really angry. I don't think it was just an offhand comment, actually. If you read there, they kind of have a 128 page model card where they explain all the model behavior. They talk about it pretty similarly, right? Where they kind of say, they show that behavior, that kind of how it can happen.

And that's not the concerning part because, you know, anybody that works with these models, the fact that could do something erroneously like that under certain conditions like, okay, like these models do crazy things. But it's the way that they, they talk about it, right? They say something like, you know, I'm going to misquote it, but it's something along the lines of like, no, and this is, you know, probably appropriate behavior, but it could happen in the wrong situations, right? Yeah. I think it like really struck me as like police state shit. Like I was like, oh, okay, like, it's like even the thread, right, it's kind of like, well, if your dress isn't too short, you'll be fine. But,

Maybe like, you know, maybe don't talk, maybe don't, you know, write a story about threatening certain. Like, it's just not how our country works. Yeah. And, yeah, I find it really, really, really deeply concerning. And I think, I think that like AI safety as a whole, and this makes me sad, actually, because I think AI safety could be really good. Totally.

I think it could be needed. But I think that it's kind of like safety from who or safety from what. Well, yeah. And the whole idea that there's zero human intervention. It's like, it's like, we're not going to, we're not going to check this and be like, oh, it's a 12 year old who's just like exploring space on the LLM. And it's like, yeah, it's just cat walks across the keyboard. The whole potential, it's like, okay, amazing if you can identify bad actors. And then, you know, work within the legal framework that we have of the existing legal framework that doesn't necessarily need net new law.

You can just work within what the government has already decided is, you know, laws are also things that it's not always like good or bad. Like laws are just laws. There is one exception. I know a lot of people use these models as like personal trainers. And I think it's deeply immoral to skip leg day. And so if you found, if a model found out that someone was skipping leg day, they should call the press and contact the regulators. And the regulators. Yeah, definitely.

The FDA. Yeah, yeah. And the SWAT team. Yeah. They should break down your door. Yeah, this almost implies that like the model would SWAT you. Yeah. It does.

Yeah. That's what it implies. And it's and it's going to lock you out of relevant systems. No, it's going to hack your computer. I can't even. I can't even. That's literally said it's going to hack your computer. That's so wild.

Well, the tweet's been deleted. So it's, it's hopefully they backtrack on that. Hopefully it's not. It's actually he deleted it and then rewrote something about how people were taking it out of context that was just doubling down. Like he was like that's that's I think the really concerning thing is that I you know I've written bad tweets before we've all written bad tweets but when I see someone you know accused people taking out of context and doubled down really concerning.

So yeah so to credit to Sam he says I deleted the earlier tweet on whistleblowing as it was being taken out of context. This isn't a new clod feature and it's not possible in normal usage. It shows up in testing environments where we give it unusual. free access to tools and very unusual instructions. Okay. But I think the issue is that probably within minutes of like releasing the model, you have thousands of people that gave it root access to their computer through like cursor through

their, you know, cursor cloud code, et cetera. So I think that like maybe one one thing we've learned from this is that like there was a time where it was the idea of hooking up a model to the internet was scary. And that was like two years ago. Yeah. I mean, not even that. There was a time when there was a knowledge cutoff. Remember this? Yeah, yeah, yeah. We don't even want you to know about the last three months. And now it's like it knows it goes to every web page. It can definitely do get requests. There's definitely malicious things that could do with get requests. But it's pretty, it's pretty responsive.

And you know what? Most of the time it just gets you the answer. You're looking for it. Yeah, this guy, Justin Halford, who have DM with a bit, says, can you imagine getting shot by the authorities in your own home because you're philosophy, homework, contained a touchy topic or context that was misinterpreted as a request. We really need to avoid such paranoid contexts altogether. This ain't it. Yeah. It's a good tip. Very rough. Anyway, hopefully they sorted out. Hopefully there's more discussion here. Anything else you want to close out with? It's been great having you. No, it's been great. Fantastic. Let's do it again soon. Yeah, let's talk so. You're overdo. This is great. We'll talk to you soon.

Cheers. I want to close out with some other interesting drama in the prediction market. Of course, were sponsored by Polymarket, but this has been going back and forth on X, which is also where we distribute the show. So the official X handle, just at X, says recent speculation about XAI's involvement in the prediction market space has been circulating while we're enthusiastic about the potential of this industry and engaged in various discussions. No formal partnerships have been confirmed to date. Stay tuned. So they're saying, you know, we might be in talks. We might even be in advanced talks, but we haven't announced anything yet. So if you're reading it on the internet, it's not confirmed yet.

And so Mario Noffel broke this down. Kulshi walks back, X-A-I deal claim. So Kul-She tweeted that X-A-I is doing a deal with Kalshi. But Bloomberg retracted the story after the, after, I guess, X rescinds a statement. And so it turns out XAI says there's no deal. Kalshi rescinded their own announcement hours later. No contract, no collab, no confirmation. Pretty insane that both Bloomberg and X came out and were like, this is not real.

I don't even know how this happened. And how did it happen in the first place? How did this happen in the first place? Because like the last thing I would want to do is say I have a deal with Elon Musk when I don't. Like he doesn't seem like the type of person to just be like, he doesn't seem like a person to be like, oh yeah. like, you know, we did talk and like, yeah, they're getting, they're putting the cart before the horse a little bit. It's like, no, he, he, he, he, he, he, he, he, he, he, he, he don't let a lot of

companies that sell to SpaceX put the SpaceX logo on their website, you know, like, every company. They have to say, yeah, we work with a big space company. Yeah, yeah, and it's like, okay, we know exactly what you're talking about. But yeah, I mean, Elon's, like, like, if you're going to do a joint press release of Elon, like, you better have his buy-in. So this is a very weird, weird thing to even have happen. And I guess Bloomberg, deleted the entire post, I guess, the entire article. But anyway, I mean, it's a knockout, dragout fight to get integrations in the, in the prediction market game.

Yeah, it's just not great for Kalshi, given they already came under fire for, like, trying to get A-B to just like spread. It's basically just like meme. Dog pile on chain, which is odd. Just sort of like competitive, you know, competitive markets, but there's no need. for foul play. Yeah. Yeah, it does seem a little bit,

a little bit too aggressive, crossing the line. But anyway, hopefully they can sort of the deal and everyone can kind of learn what the real strategy and prediction markets is with XAI. It would be interesting to see where it goes.

It makes sense that it's integrating. We are the integration point for Polymarket into X because you see our live stream and you see the ticker right there. And of course, Polymarket and Kalshi both post-predicts and screenshots all the time. It's interesting to think about

GROC being able to access polymarkets. for basically getting a read on future events. Right now it's oriented around fact-checking. But it also can give insight into potential headlines of the future. Will this acquisition happen? What's the likelihood? And I would imagine that as long as both companies have a robots.

tx-d that's permissive, they would show up in chat GPT. They would show up in GROC. I imagine that right now if I ask GROC pull up the, pull up the polymarket on the U.S. recession chances in 2025. It could just go do that. I would expect that that would be the behavior just because it has the ability to browse the web. But obviously, a deeper integration would be cool

for whichever company can win it. So good luck to Shane as he goes on a tear and tries to build polymarket into a generational company. Let's go to Delian and close out here. Delian, I like this because we're working on an article. It'll drop, maybe tomorrow we'll be covering it here about TBPN. in the article and the fact check,

they hit us with, is it true? Fact check, is it true that you refer to your team as the guys, all caps, or capital T, capital G? And I was like,

honestly, like, yes, but we also call them the boys. We also call them the gentlemen. We also call them the production crew, the crew, whatever.

We don't really, we haven't really formalized the TBPN production team as the guys, but they are guys. But Delian put it into a great post he says if the boys isn't a clearly identified group of six to 12 tight-knit men in your life you're just not going to make it that far in life and uh yeah let's give it up for the boys every dude needs a group chat of guys give it a funny name uh and more importantly go and text

your your the boys text the boys right now tell them where you're going to see mission impossible the final reckoning in theaters get the tickets uh send the Venmo requests, get everyone into the theater to go see Tom Cruise, do what he does best. This is a call to action. This is a call to action. No matter what city you're in, hit the boys and say, we're going. Just off your tucks. Pick a date.

Black tie. Just make it happen. And you know, you dress up too. You got six to 12 tight-knit men in your life. Maybe six show up because people are busy, but you refund the rest of the tickets and you're good to go. Yep. So do it. This is the playbook.

This is the playbook. Get the boys together for. Mission Impossible. Fun show. Fun show. Leave us five stars in Apple Podcasts and Spotify. That is directly from the board. We'll see you tomorrow. Thank you so much for watching. Cheers.