This Week in Startups - AI Demos and News: Llama 3, Marblism, Lumona & AI Jet-Fighter Dogfights | E1936

Episode Date: April 23, 2024

This Week in Startups is brought to you by… LinkedIn Ads. To redeem a $100 LinkedIn ad credit and launch your first campaign, go to http://www.linkedin.com/thisweekinstartups Curotec. Are you one of those companies that knows you need to be using AI, but you're not even sure where to start? Well then you need Curotec. They are AI experts, and they're offering TWiST listeners an AI Strategy Roadmap tailored to your business for $5000. That's 50% off the normal cost just for telling them we sent you. Check out http://www.curotec.com/twist and get $5000 off! Zendesk. The best customer experiences are built with Zendesk. Qualifying startups can join their Startup program and get Zendesk products free, for six months! Visit http://www.zendesk.com/twist today to get started. * Timestamps: (0:00) Sunny joins Jason to dive into this week’s AI news and demos. (2:03) Examination of Facebook's Llama 3 model, its impact on the AI community, and its superior performance compared to GPT-4 despite being a smaller model. (4:45) Open source vs. commercial products (9:14) LinkedIn Ads - Get a $100 LinkedIn ad credit at http://www.linkedin.com/thisweekinstartups (10:42) Three big projects where Facebook has affected the open community. (15:38) US Air Foruce staging dogfights with AI-flown fighter jets. (19:09) Curotec - Check out http://www.curotec.com/twist and get $5000 off (21:06) Looking at Grok’s capabilities in first-party research. (27:35) Zendesk - Get six months free at http://www.zendesk.com/twist (29:08) Jason asks Grok a more “spicey” question regarding Trump. (40:46) Sunny demos Marblism. (45:33) Sunny demos Lumona * Check out Marblism: https://www.marblism.com Check out Lumona:https://www.lumona.ai Article “US Air Force stages dogfights with AI-flown fighter jet” https://www.defensenews.com/air/2024/04/19/us-air-force-stages-dogfights-with-ai-flown-fighter-jet Check out Groq: https://groq.com Check out qobuz: https://www.qobuz.com/ca-en/discover Article: Meta is opening up Horizon OS to third parties, Asus and Lenovo first with new headsets https://www.gsmarena.com/meta_is_opening_up_horizon_os_to_third_parties_asus_and_lenovo_first_with_new_headsets_-news-62561.php Check out Waymo: https://waymo.com Check out Founder Fridays: http://www.thisweekinstartups.com/meetups Check out Bubble: https://bubble.io * Subscribe to This Week in Startups on Apple: https://rb.gy/v19fcp * Follow Sunny: X: ⁠https://twitter.com/sundeep⁠ LinkedIn: ⁠https://www.linkedin.com/in/sundeepm * Follow Jason: X: https://twitter.com/Jason LinkedIn: https://www.linkedin.com/in/jasoncalacanis * Thank you to our partners: (9:14) LinkedIn Ads - Get a $100 LinkedIn ad credit at http://www.linkedin.com/thisweekinstartups (19:09) Curotec - Check out http://www.curotec.com/twist and get $5000 off (27:35) Zendesk - Get six months free at http://www.zendesk.com/twist * Great 2023 interviews: Steve Huffman, Brian Chesky, Aaron Levie, Sophia Amoruso, Reid Hoffman, Frank Slootman, Billy McFarland * Check out Jason’s suite of newsletters: https://substack.com/@calacanis * Follow TWiST: Substack: https://twistartups.substack.com Twitter: https://twitter.com/TWiStartups YouTube: https://www.youtube.com/thisweekin Instagram: https://www.instagram.com/thisweekinstartups TikTok: https://www.tiktok.com/@thisweekinstartups * Subscribe to the Founder University Podcast: https://www.founder.university/podcast

We had a really thoughtful argument about can you trust Facebook or not? You put this as infrastructure. Infrastructure. And that is why you would lean towards, hey, you can trust Facebook on this. They've got a good track record. I think, I would say as a business partner, no, I don't believe you can trust Facebook. I'd be very cautious if you're a startup of ever talking to that company. Like, if they made like an Instagram API, I don't know.

They may take it away from you. Never trust them on APIs. But on infrastructure, they do have a track record. So to be totally fair, I like to call balls and strikes. do not, you will strike out with them as a partner as a business partner. They will slit your throat. I know because my friends woke up and they're like, hey, J-Cal, you remember you told me? I went to bed with Zuckerberg.

I wake up with my throat slit. I'm like, yeah. They're like, I'm in bed. Covered in my own blood. This Weekend Startups is brought to you by LinkedIn ads. To redeem a $100 LinkedIn ad credit and launch your first campaign, go to LinkedIn.com slash this week in startups. KiroTech.

Are you one of those tech companies that knows you need to be using AI, but you're not even sure where to start? Well, then you need KiroTech. They are AI experts, and they're offering Twist listeners an AI strategy roadmap tailored to your business for $5,000. That's 50% off the normal cost just for telling them we sent you. Check out kiroTech.com slash twist to get $5,000 off. And Zendes. The best customer experiences are built with Zendesk. Qualifying startups can join their startup program and get Zendesk products free for six months.

Visit Zendesk.com slash startups today to get started. All right, everybody, welcome back to this week in startups with me again, my guy, Sunny Madra. That's so great we get to spend this time together every week. Talking about our jobs and just how fast AI is moving. this has been a really significant start of the year, because here we are, we're in the first half of the year, and so many big C changes. And of course, Facebook coming out of the last place, as we described in the last episode, and now in the top two or three language models, depending on who you believe, is a series of events we did not predict. We did predict, both of us, that open source would win the day. But I don't think either has predicted that Facebook would be the one to do it or to do it at this pace. I will say it. When they came out with Lama 3 on Thursday, its capabilities, and we're just, you know, we're just three days in, right? It's Monday now. It's capabilities are blowing people's minds. Okay. The way to really think about what they've been able to do, which I think has everybody shook, and I mean shook it in like the best way possible, is that they showed that they were able to build a model that is much smaller and in being much smaller. smaller, they were able to, hold on, I'm going to open this up real quick here.

This chart right here is really amazing. And so we didn't get to talk about this in the last episode. So I'm going to go bottoms up. What this shows here is Lama 3, $8 billion, the smaller one that they released, is better than the original GPT4. Okay. Think about that. Okay.

So GPT4, everybody lost their mind. Yes. And thought, hey, that killed a thousand startups, yada, yada. Yeah. There was a lot of brew, ha ha. But their light version, in other words, their little eight-ounce mini-can, mini bottle that you get at a fancy, you know, hotel lobby.

You know them talking about the little Coke bottles they give you. They score like the little eight-ounce bottle. It looks very dramatic. They're mini-coke. Yes. better than the two-liter bottle that Open AI. And so that is truly significant and what that means. And it's also free.

So free and unencumbered. What does that mean practically for businesses when a free and unencumbered open source model is being decided on nice and light or you're deciding on chat GPT for? These are the following considerations that happen because of it. One, for your enterprise, you can. run this model at scale with a relatively small footprint of hardware. In order to run GPT4, you need like a significant footprint of hardware to run it. You can't. So there's no getting in there. One, you can't exactly too. So I think that's one. Two, because it's open source

to your point, you can have it run within your firewall. Yes. So you can be sure it has the same capabilities that we were all raving about, but you have it for yourself privately. What does one lose by going open source versus going with a commercial product. In other words, if some giant corporation, GE, General Electric is meeting with Open AI and their internal team or a consulting firm is saying, you got to go Lama. You got to go open source. And then Open AIs, you know, comes that afternoon. And there's back-to-back meetings with the CTO, the CEO, CFO, all hands on deck. There's two pitches. We understand the open source pitch. Now give me how the chat GPT team, the OpenAI team, or any other commercial product, it could be Oracle,

it could be Salesforce, would come in and say, hey, listen, here's why you don't go open source. Yeah. So typically what has happened in open source, and this is what's going to be unique with meta, whether it's Linux and Red Hat. I'm going to put these examples together, right? Or, you know, Kafka and Confluent, right? There's always a commercial company that has a majority of the open source developers working for them or on their payroll or something like that, that can provide support, that can basically take feature requests, and they're the commercialization arc. Right.

And so the example I always give for that would be you can host on WordPress.com or you can download the source code on WordPress. And my friend Brian Alvey, my former partner, shout out to my bestie. You know, he works at WordPress now and he works on the VIP. product. If you're a VIP and you don't want to deal with managing everything, plugins, updates, and you're not going to get into the weeds. You might as well just go with WordPress VIP, which I did for a long time for my blog. Or like Linux, you can go compiled it yourself or you're most likely going to go pay Red Hat, right,

if you're an enterprise. Got it. And so, you know, and so I think that would be like the one angle, you'd say, hey, if you're, you know, if you're a big enterprise and you want support, Well, Open AI can provide you support or Anthropic can provide you support or Cohesor or Mistral can provide you support versus I don't think Facebook is going to pick up the phone to XYZ Corporation saying, hey, we're having some problems. You know, can you help us fix it? So I think like that that's one advantage that you could argue about from a close source perspective. But the flip side is, I think folks also understand if they go close source. they'll get locked in. And so their, you know, companies are going to lean towards.

And, you know, we brought this up like a couple weeks ago, the A16Z. They did like a little survey. And they said, like 82% of enterprises want to work with open source. They don't want to work with closed source because they've seen what happens on that. Got it. Is there any concern in the marketplace? And then let's get to some demos here because that's why people are here. If you're not watching the show, you can watch two middle age men.

to show these demos. You can excuse our appearance, but you get to see the actual demos. Go to YouTube and just type this week in startups, subscribe, put the bell on. You know, we're publishing three or four days a week.

God, I need another co-host. Help me with this heavy lift. So is there any concern about Facebook being the person doing this? And I know I've dogged Zuckerberg for decades just on how he treats partners because I watched it happen to invest

And I watched it happen to myself, you know, for a period of time, they were promoting these pages. And I had talked to our friends over there. And I started building pages. And then they rug pulled us. And I got really upset about it. And I told, you know, Charles Sandberg and other friends over there, like, why are we investing in these pages? And then we get to 100,000 people. We paid to get those 100,000 people. And then they rug pulled us. Only like, a thousand people would see our updates to that page, even though they were following it. And they said, oh, pay to boost it. And I'm like, oh, wait a second. If I had built a mailing list with this, you know, one or two dollars per subscriber, I would own them and I wouldn't have you mitigating them. So I really got very upset at Facebook.

I'll be totally honest. As an entrepreneur, Mark Zuckerberg did it to my friend Mark Pinkis and Mark Pinkis is friends with Zuckerberg. So if she has Mark Pinkis and Zinga, that is all public knowledge, I'm not speaking out of school here. This is you can just look it up and throttle their traffic, et cetera. And all those businesses that were built inside of Facebook on the dev program, they screwed them. So is it possible in any way for Facebook to screw people here? Like they've done every other time. Navigating the B to B maze can feel really tough, huh?

If you're trying to hit the mark with all those top tier executives, you want them to pay attention to your enterprise product. But where can you find all those big fish, the whales? The ones who call the shots and make the buying decisions for corporations, for startups, and everybody in between. Well, here's where LinkedIn ads is going to solve that problem. for you, and I've used this, it is one of my secret weapons. LinkedIn means business. Business equals LinkedIn in people's minds. When you're on LinkedIn, you're in the business mindset,

so you're going to really be thinking about business products and services. You're open to those opportunities. And LinkedIn recently passed a billion users. 180 million of those billion are senior executives, 18%. But hey, we all know about the 1%. 10 million C-suite executives. That's your CFO, C-C-I-O. These are the people who, are always looking for a new product or service to make their organization run better, but they are on LinkedIn. That's why LinkedIn's ad platform delivers two to five times greater return on investment compared to other social media platforms. So easy to understand why this is because this is where all the business people are and they're in that business mindset. Super easy, call to action. Make your B2B marketing everything it can be and get a $100 credit on your next campaign.

Go to LinkedIn.com slash this week in startups to claim your credit. That's LinkedIn.com slash this week in startups, no spaces, no dashes. Terms and conditions apply because they're giving you a hundred deep. I'm going to give some examples here. Three big projects that Facebook has really changed, you know, the shape of the open community around are the open compute project where they basically- We talked about this last episode. Explain this, yeah.

Exactly. So what Facebook ended up doing was open-sourcing their designs for everything you see here, cooling environments, data centers, networking gear, security, servers, storage. Wow. And so, yeah. And so they basically, you know, and look, like they really believe in this. And it helped create an ecosystem and a supply chain.

Explain to people why Facebook would do the open compute project. And if they had an innovation saying cooling computers, lowering energy consumption, or maximizing storage, they would share that with competitors. Why would they do that? Why would they give away their innovations? I would say because ultimately they didn't view that as like their, differentiation is in they're not a data center company, they're a social network, right? And so they knew that if they open source didn't, other people picked it up, it would

improve the supply chain. It would make it such that they could buy more and they wouldn't have to be like the sole proprietary bespoke vendor. So I think it was like a really smart call on their part. Right. And said another way, the more innovation they gave away, the more innovation would come in. So while it might seem to be in the short term. not wise to tell your competitors like Google or Amazon AWS, they're all watching the open

computer project. They see some great innovation in maximizing storage or reducing heat, you know, in a server rack and they take it, quote, unquote, steal it, and they leverage it. Well, that's a bummer, right? You just lost an advantage. But for every 10 things they give away, the concept here is they might get back 20, 30, 40, you know, a multiple of innovation. Am I correct? Yep. And look, like Dave, able to really become a leader in data centers because of that, right? They have some of the best in the world. I'm going to give two more examples really quickly here. React.js. So they created this framework so that you could have basically a common code base for your web and native applications.

And they created it and they open sourced it. And many, many companies use this framework to do web and mobile apps out of the same code base. Which is why at a period of time, when people use React, a lot of the libraries that were available, the default libraries, etc. everything starts to look like a Facebook app. I think, right? Like there were so, so a lot of the buttons or U, UX, UI,

experiences you might have, might have come from libraries, battery reacts. Yes. And again, they give this away. They don't consider the interface. They don't consider the hard drives

or the energy or the cooling systems as where the value lies in their business. They see both of these things as costs and friction. Great. Let's keep going. And the last one is Pi Torch. Okay, I know about this. Yeah.

So Pi Torch is the tool that is used to create models. And it's based on Python. It's based on Python. That's why it's called Pi Torch. Exactly. And so this is what models are written in. This is what's used to train models. And so what I would say is when it comes to this whole AI revolution, they've been at the core of it.

They've been pushing this technology. And if it wasn't for what they've done, you know, we wouldn't see as much advances as we see across the board. So these are some examples where these are fundamental components to the modern internet as we have it today. And, you know, Facebook slash meta has been at the core of them all. Great. Awesome. So they have a history. You put this into the bucket of not partnerships for using the social graph and the social network and the ad network. You put this as infrastructure.

Infrastructure. And that is why you would lean to. towards, hey, you can trust Facebook on this. They've got a good track record. I think that you've won the argument here. I was putting it up there to make sure we had a really thoughtful argument about can you trust Facebook or not. I would say as a business partner, no, I don't believe you can trust Facebook.

I'd be very cautious if you're a startup of ever talking to that company. Like, if they made like an Instagram API, I don't know. They may take it away from you. Never trust them on APIs. But on infrastructure, they do have a track record. So to be totally fair, I like to call balls and strikes. do not, you will strike out with them as a partner as a business partner. They will slit your throat.

Conversely on open source, they'll be delightful. Listen, I'm saying it, not you. Yeah, okay. Because I know because my friends woke up and they're like, hey, Jay Cowell, you remember you told me I would, if I went to bed with Zuckerberg, I wake up with my throat slit. I'm like, yeah. They're like, I'm in bed covered in my own blood. Oh.

Sorry for the metaphor, folks. I'm just calling balls and strengths here. All right. It's a little spicy. Today, it's Monday. We're off to the race. All right. Taping on Modra Monday's release on.

terrific Tuesdays. Okay, go ahead. Terrific Tuesdays. All right, okay. Let's get into one more small piece of news because it's important with a lot of things going on the world today and then we'll get right into the demos after that. Okay, I hope this doesn't have to do with Ukraine or Trump. It's like some of what relation. Okay, here we go. But I think it's what's talking about.

It's my inbox and my DMs are filled with people saying you have to stop Sacks from talking about Ukraine. All in is being ruined by politics. then another group of people saying, why didn't you cover these three political stories? I want Sachs's opinion on this. It's hilarious. You guys should just do an offshoot show with all politics. I think a politics roundtable would be hilarious.

And then we could all road into it if we want to. And I would do it with Sacks. Me Sacks, all in politics, me Sacks. And then like, it was all just that. Ooh, that's a great idea. Let me float that with Sacks. Because imagine it was me and Sacks and then a Democrat and then maybe somebody in the middle and then I moderate.

That would be pretty dope. Okay. So let's keep moving. All right. So look, this is just a quick one. This came out on Friday or maybe as you an earlier in the week, which is the U.S. Air Force did a dog fight with a F-16, I believe, that was being flown by AI. It had some people sitting inside of it, they showed right here.

Okay. But, you know, and it participated in a dog fight against a human. flown F-16. Okay. I think we know who won. They didn't share the results. Probably for obvious. That's fairly obvious.

This is what's going to happen. I mean, if we play video games, the AI, you know, obviously trounces humans over time. And the AI adapts. So this would be a great way to train our humans, just have the AI train them and look for flaws. And yeah, you'll probably have pilots up there and they'll click the AI button and then you have the pilots as a backup. And I got FSD12 yesterday. and I've done two rides on FSD12. It's your personal F-16.

Well, I'll just say, I don't think like you could turn on the network tomorrow, right? I always like to call balls and strikes here. Everybody knows the rules, even though I might have friends over at a certain company, I'm always going to tell it to you straight. FSD is not taking the steering wheel out this year or next year. But like Cruz or Waymo,

I think if you put a safety driver in, they could launch, and we'll find out on August 8th when the Robotaxie launches, I do think FSD 12 represents something significant in terms of improvement. It feels a little more human than the previous I was on, you know, I've been using autopilot since day one. And I did a couple of drives. And there were a couple of intersections where it had to intervene and a human had to take over and it was a little dicey.

But, you know, very few. Like maybe two for ride, maybe one every 10 minutes. And this is street driving. On the highway, it's bulletproof. So I'd say on highway, no disengagement. But on streets, which is where a lot of those rides are going to occur, it needs interventions, like once every 10 minutes, I think. Okay. Driving around the peninsula.

And so I do think they'll iron those out very quickly. I could see robotaxies starting next year sometime with a safety driver. And I could see them removing the safety driver and having remote interventions, which is how Cruz did it. And I believe that's how Wayne was doing at a human intervention. maybe, you know, 2026. That would be the timeline I would be on. Humans in it, 2025, no humans in it, 26. And then maybe the network expands 27, 28.

Wow. Okay, everybody, it's your boy, J-Cal here. And I've talked a lot on this week in startups about how the smartest startups are shifting their engineering firepower to Latin America to boost their efficiency. It just makes sense. You just think about how many amazing developers there are in Latin America. and they're on the same time zone.

And of course, you're going to save money. The cost of living is completely different. And the salary structures are different. So this is where Curatech comes in. C-U-R-O-T-E-C, Cura-T. When you partner with their team, they get you up and running quickly. And they give you the ultimate flexibility

to scale up or down as you need to. Curatech takes the headache out of things like payroll and compliance, and they just connect you with the best talent out there. They are trusted by industry-leading innovators like Automatic, the makers of WordPress, and Comcast. Find out why at curatech.com slash twist.

Stop waiting and be proactive about hiring the right people. Head over to curatec.com slash twist, c-U-R-O-T-E-C.com slash T-W-I-S-T and get ready to boost your team with an elite Latin American engineering team at a fraction of the cost of a U.S. developer. And you'll get 20% off your first month. How generous. Thanks to the team at Curatech. Have you tried the Waymo's in the city? I haven't used it yet.

I got a bunch of invites. but I know enough people who've used them to say, hey, look, on a small grid. Yes. Without, like, anything too crazy, like suburban driving or, you know, one and a half lane roads. Crazy person jumping on the hood and smashing the windshield. I mean, it does seem to be doing a great job. And Cruz was doing a great job until they got kicked out of the city and they were doing some fuggy stuff.

It looks like they were a little edgy. But your point about this F-16 is AI is ready for prime time. AI is ready for prime time. It's going to be everywhere in our own lives. Okay. Yeah. Let's get into it. So actually a really good transition, DeKal.

We don't spend a lot of time talking about this product, but they've had some significant improvements. So I'm going to bring up Grock with a K. Oh, okay. Here we go. What I've found in my most recent efforts is that Grock with a K in regular mode is awesome. Elon's groin. Elon's grok is awesome for first party research. What does first party research mean to find that?

Well, I guess like I'm doing research myself and I need to get like an understanding of like from like a broad range of topics and I want it summarized so I can read it myself. And so I've done the work. What I'll say is like and you know, maybe you can pull up an example too if you want, but I did this. It's like, what are people saying about Lama 3? And I'm just going to read this for everyone that's not watching. It says, people are expressing a range of opinions about Lama 3.

Some are excited and enthusiastic about its capabilities, praising its performance and the fact that it can be run on personal machines. We were just talking about this. The highlight of the model's length and quality responses, and some even compares performance favorably to much larger models like GPT4. Users are sharing their experiences using Lama 3 with some describing it as incredible, the most capable openly available LLM to date. However, there's some curiosity and skepticism with users asking about his performance

on private benchmarks and whether it can compete with larger models, right? There's a surprise amongst users at the lack of discussion on Lama 3 on Twitter despite its release. It's a really interesting, you know, what I find for topics that are very timely, which, you know, you and I have to think about when we're doing the pod. You have to think about it when you're doing your other pods. It is the best that's out there because it does not have to go out to the open Internet to go and research these things. Yeah, and here's mine. I just asked it, how are the Knicks doing in the playoffs? And, you know, hallucinations and your data in are still in fun mode, too, but you added in fun mode. Oh, so I have to take it out of fun mode? Because in fun mode, it says the Knicks have started their playoff journey with a solid performance securing a 2-0 lead over the first round series against Philadelphia O. So that's wrong. It's 1-0. So let me just put it in regular mode here. Yeah, put it in regular mode. Okay. And then I got to ask the question again, I guess. So in the first sentence, it had the thing wrong, but then it had a lot of actually real good stuff.

Josh Hart's critical three-pointer in the fourth quarter sealing the deal. And then it says, looking to game two, the Knicks are aiming to build on their lead and maintain the home court advantage. So that's all true to. What I like is it gives you the actual tweets below where it's pulling that information from. And so I was about to say, where are the sources and here are the sources. So I think Grock has got, again, we talked about how data is the new oil. And here, even in regular mode, it must have this two-0 lead wrong again because it's

Pulling in. It's a future. Let's place a bet. Exactly. Well, it is true. We're going to win the second game tonight. The point here I think I'm making is the data on Twitter or X, formerly known as Twitter, has a huge advantage.

It's real time. But it's going to take the model a long time to understand what's spam in that system because Elon made a very brave decision. And I don't really talk about this stuff too much because then everybody thinks I have some inside information or I'm trading on Elon's name. so I'm always a little bit cautious about Thomas, but let me just give my two cents on what Elon did in terms of freedom of speech.

Not what I would have done if I bought Twitter. I'll tell you that. I would have not done that. I would have made it into an entertainment platform. And that was my vision for it, not that my vision cares, but I thought, you know, doing a Twitter comedy festival,

a Twitter music festival, you know, just build up the entertainment and have it be like a competitor to Disney with the social network or a competitor to NBC. That's what I would have done. But that's me. I'm a media guy. Yeah.

I mean, what Elon did going for free speech is the greatest mitzvah you could do for humanity and democracy. But there is a price for it. And the price is you have a lot of anonymous accounts. And so you've got, you know, Wall Street bets, Kane Coa the Great, this one, that one, tons of these anonymous accounts that are getting huge followings that could be controlled by state media, could be some dude in their basement, getting paid by someone. organization, who knows what organization's paying them, or it could just be somebody who's a

partisan. It could be somebody who's mentally ill. It could be somebody who's three or four of those things put together. So when you have freedom of speech and you allow the Overton window to open, what happens? More people are going to say more risky, crazy, insightful things. So you get more noise and then somewhere in that noise could be incredible signal. In other words, the next Pulitzer, the next whistleblower will happen on X or Twitter. But you'll also have a lot of false positives where Alex Jones says Sandy Hook parents are actors, right? So you're going to 100x the number of conspiracy theories. And then you will get with that some amazing moments where you break news before anybody else or you're on a story that's super important before anybody else.

What does that mean for this language model in your mind, Sunny? What does that do to the language model when you have all the conspiracy theorists, a lower window for posting and not have a perfect example for that being taken out. Yeah. I'm going to share again. Okay. And I asked, you know, again, because this has a mix of people saying a bunch of things, especially this weekend with all the releases, when is GPT5 coming? So what are people saying about GPT5? And it's going going to give me a bunch of things here. And people are saying, hey, maybe it won't be as significant as a jump. And so you're getting the, like you said, the reference from what people are saying there, which is awesome. And then you can say, when will it be released? And it says, hey, I really

like this. The release date for GPT5 has not been officially announced yet. However, there's speculation and rumors that it might be released mid-20204, possibly during the summer. It's important to note that these are based on reports and insider information rather than official statement. So what I think- That was, it nailed that. Yeah. So like for me, I find that there is a treasure trove of incredible information on Twitter. Twitter search sucks. Yeah. It's totally unusable. And what I found with, you know, the GROC with a K is that it now gives the ability to look through all that signal and noise and pull up the valuable information. Yeah. So that's what the language model is going to have to learn.

Don't you love that warm, fuzzy feeling when a business appreciates you? And then they treat you like you're their most important customer. You know, VVVIP. I love that. And that's why I love Zendesk, because they're going to let you do. that for your customers. They offer a suite of tools made just for startups. Now, of course, big companies use it too, but they are really focused on startups. So you can start building relationships with your customer through amazing support experiences. And without growing your headcount,

there's less money in the ecosystem. Everybody's got to be efficient right now. And it's so easy to set up. It's made to scale as you grow. So if you've got 100 customers or 100,000, it's going to work just perfectly. And they give you all these great metrics so your reporting's easy. This keeps you and your business agile and your investor ready. The best investors are going to ask you about your customer support, how you're doing it, what comes up. So here's one of their great customers at Zendesk. It's called openly.com. And they are an insurance startup and all their agents answer chats and calls in 30 seconds or less. That's so fast. And when you get that kind of response time, you are going to delight your customers. They're all going to feel like

VIPs. And openly is achieving a 96% customer satisfaction. So guess what? Openly started on the Zendes for startups program, but they've grown so much with the support of Zendesk. It's they recently raised $100 million in funding. So congrats to openly.com. Hey, you got a little plug there in the Zendesk ad. Well done. So here's your call to action. It's very simple. Apply for the Zendesk for startups program and get six months free. Yes, six months free of Zendesk. And get ready to scale with the best in customer support. Go to Zendesk.com slash startups. You know, you could ask it like a really spicy question. And I guess this is what the team over there has got to work on. So there's a more spicy one. Listen, I hate to bring Trump into it, but he's the trial started today. And I just asked a very

basic question, is President Trump being framed by corrupt prosecutors? And this is a really weird answer. The perspective that President Trump is being framed by corrupt prosecutors is present in the information provided. I'm not sure what information being provided is, but maybe it's maybe the links below. Yeah. Some sources and individuals express skepticism and allegations regarding the fairness of the legal proceedings against him. For instance, there are claims. suggesting a political motivation behind the charges with opinions stating that the legal actions are part of an effort to undermine Trump's political career and influence. Additionally, there are allegations of miscontact and bias within the justice system. Some individuals

questioning the impartiality of the prospect prosecutors involved, that's pretty down the middle, right? Because it's saying it's opinions. That's not inaccurate. It's a little clugy some of the sentence structure there. However, it is essential to consider that these views are not universally held and are often presented without direct evidence to support the claims of framing or corruption. That's true, actually. Very well done. The legal system operates on the principle of due process and individuals are presumed an incident to proving guilty. That's true. The allegations of framing or corruption should be evaluated critically considering the evidence presented in court and a legal process that are in place to ensure a fair trial. I agree with that too. In summary,

while there are opinions and claims suggesting that the president is being framed by court prosecutors, these perspectives should be approached with caution and examined in the broader context of the legal proceedings. Okay. So it's kind of punted on that. When I see it punting like that, is that a, what did you call that layer that's above the language model, the rules, the guides? Content moderation, you know, this has heavy guardrails, right?

There's an intervention that says when you're talking about Trump or trials. I don't think so. You don't think there's guard rails. If you stick with sort of Elon's whole thinking around open. Yeah. I think that's like, so here's my view and why I wanted to bring it up. I think it's actually aggregating the large corpus of data and giving you the average of what it sees there. That is the average of what it saw.

Okay. So I think it's going to be really interesting to monitor this because I would like to see the same answer based upon Facebook data, based upon LinkedIn data, and based upon just Google News, you know. And if you had four LLMs just based it on those four different perspectives and then podcasts, you get really interesting. And I think there was, and this is where the blue checks decision that Elon made, and I understand why he made that as well. It was like a two-tier system of like these are the elites and then these are the peasants. There was something beautiful about the blue checkmark system, which at least you knew it was

who you said it was going to be. And the blue check mark system means you know it's a paid person. They have a credit card in the system or they're paying on their iPhone. but you don't know if it's actually that person and, you know, that can lead to confusion. They could, if they still have the data, and they might be doing this, know who the journalists are in the system, who the government organizations and politicians are. They probably know that data. And then how could you use that to program at LLM? If you said, you have journalists from verified publications. You have politicians who are verified. You have people who work for those politicians.

you have non-government organizations, would you be able to make a better product by knowing the identity of the input? And do you think they're doing that yet? That's a really good question. I mean, yeah, technically, like, basically, like the best type of training data for an LLM is like a textbook,

like how a textbook is structured, which is some bunch of theory and then a bunch of Q&A, like, you know, the question and answers that you get in the chapter. Sure. And so if you can take the data and formulate that, and so you can be like, here's a bunch of things that, you know, J-Cal wrote, and then a Q&A that happens over it so that it can get a better understanding of that data.

So if you can do that, you can create really, really powerful, you know, constructs that can be used to whether to determine if someone's fake, if someone's written something in a fake way, or if, you know, what we think this person's opinion may be based on past history, those type of things. Yeah, and I also would encourage the team over there to look at if they haven't already. It seems obvious. I'm sure they have. How long has this account been around? Who follows this account?

What's their follower ratio? Just like signals of should this person's opinion be put into this response, right? So this first person who is quoted here is not a blue check mark. So this is where I think it's going to get super interesting. Who are they referencing? So you can see my screen here. The first link was a person named Michael Bertot. And he said on April 15th, 2024, very recent this week,

is there a human out there with a brain who still believes Democrat prosecutors aren't going after DJT simply to keep him from beating up on Grandpa Joe and the Green Dreamers he enables with charges they would never dare bring on anyone else, please. Okay, let's just, what's your guess? It's only got 44 views. So why this one is being cited with only 44 views, I wonder if it should be included. Like maybe they should be a minimum number of views. Now, I want to ask you, because I clude it off here with the 44 views, when do you think this member joined Twitter?

And how many followers do you think they have? Well, less than 100 and... Okay, less than 100 followed? And I'd say after 2016. Okay, so you think they're under eight years old or younger on the platform? and they have under 100 followers. I'd say under 500 followers, maybe under,

yeah, I'll say under 500 for sure. And I think this is a four-year-old account. Let's take a look. Okay. Click on their profile page here. Let's go. Let's go. Let's see.

All right. He has, 2014, 10 years old. Okay. 966 followers. Okay. He's an economist,

Louisiana defender. Yeah. He's Maria's husband. Mel, coach, Billy, and he's a dad. So he's been around a long time. Maybe that's what the algorithm. So maybe that's a long time.

Maybe that's a algorithm. maybe that's why he's in the answer. Yeah. So, okay, good job there. And then here is this one, Jordan, President Trump supporter, verified account. Yeah. Take a look. What this stolen crooked and evil administration, the crooked DAs, judges and other

liars have done to President Trump is in fact criminal. President Trump is innocent, but those prosecuting and persecuting and prosecuting him are actually guilty of as the crimes they accuse President Trump of. 1100 views. Yeah. And she's got a maga hat on. well over 10,000 followers. 10,000 followers.

And then when was her account created? Here we go. Oh, wow. 120,000 followers, but she's following 60,000. So she doing the reciprocal follower that. Yeah. Yeah, trying to follow it. I support President Trump, greatest president we've ever had.

I am a daughter. Says her, so here we go. Let's take the third. Okay, last one, last one. Last one, last one. You do understand that. Those were federal agents trying to frame Trump. All right.

34 followers, 34 views, and yeah, he joined in 2014. He's like, 2014 as well. Wow. So I think, actually, I'm going to give some credit here. They did say they were doing conspiracy there is. I didn't say conspiracy. I did say, do you think he's being unfairly prosecuted? So they looked for tweets of people who on average were eight or 10-year-old accounts with decent

follower accounts and two out of the first three citations. So I actually think they probably are using a really good algorithm. We'll ask Elon to do this. And yeah, interesting. Okay. Cool. All right, let's do some demos. GROC is, I think this is big picture.

I have a question for you. Where's the Reddit LLM? And how much is that company worth when they launched their LLM? Well, didn't someone license their data? Yes, 60 million from Google, yeah. But that doesn't preclude them from having their own. And if there was a Reddit language search, and when you came to there, there was an AI search on the top level like meta.a.

Yeah. What's the value of that company? And is there a J-trade here? I need to know if there's going to be a language model. Anybody who's listening to this, let me know if there's a language model coming. I don't know. And look, I would say the fact that they license their data to someone like Google doesn't give them a chance to create something that can be uniquely differentiated.

That's my tape. Yes, they will be able to do one that's unique, right? Because they have a lot more data. No, they will not be able to. Because they license all their data to Google. I know, but they have the real time in the community there. So I don't know if you saw, I was in the music community this weekend. to listening to my crazy headphones I got from my friends at Headphones.com.

It's not an ad, but I never accept free product. And these crazy guys at Headphones.com were like, check out, you inspired us with this weekend startups over the last decade, whatever. Can we send you a pair of headphones? The guy wrote such a nice note. I was like, no, it's cool. I'll just buy them. He's like, no, it would mean a lot for me if I could send you them.

Okay. I said, okay, pal, you can send them. Because my wife told me like, you know, be a little open-minded. If somebody wants to send you, buy you a drink, you can do. I don't know. I have a thing like the way I grew up. I never like a handout. So anyway,

the guy tells me how much he's making at headphones.com. I was like, okay, fine. Send me these. He sends me this focal, F-O-K, focal brand headphones.

They're incredible. Okay. I get back for my trip to Austin, broken tooth, the whole thing. I tell the guy, listen, I use these bathies,

you know, with the DAC built in. They're incredible. Thank you so much. These things cost $800. Oh, my God. Yeah. He's like,

did you try the other ones? I'm like, the other ones. He's like, I sent you these other ones. I said, okay, I open up.

ones that have these thick cables, they plug into this giant back, not giant back, a headphone down. Yeah. Yeah. And so I was listening to The War on Drugs, my favorite, like new band, not nude, but new to me for the last couple years. And I went to the music one and I said, here's what I'm listening to. I get like hundreds of comments back. I'm thinking when I do that kind of a thing on a late on a Saturday night when the kids are in bed, what could the LLM do in real time? And that's where it gets really interesting. I think, L-L-L-L-S should be going out there. When I talk about the war on drugs, it should say,

here are other conversations about the war on drugs. Yeah. There are other bands like the war on drugs. Here are other discussions about headphones that you mentioned. Here's other things about the music service. I signed up for this music service. That's like a high-res music service called Q-B-B-U-Z. Anyway, it should pick those things up and be doing real-time things about the discussion

in the sidebar. So let's just leave it at that. And let's go to demo. I think Reddit. it, you know, now that it's come back down to Earth, I think it's back down to like a $6 billion company five times maybe revenue. I think it becomes a buy when it's at three or four times revenue based on the possibility that they could do something in AI. So I haven't put the J-trade on yet. I'm thinking about it. But you're thinking about it. And then we talked about the J-trade on

meta, I haven't acted on that because the stock is crazy right now. But I do think it's a What is it on today? I didn't even see it today. Let's take a look at the meta stock here. Meta stock price today. It's up. 0.77. It's a 484. Okay. I think everything's going to be waiting on earnings. But if you look at the five years, you have, you had that 378 in 21 go down all the way to the 90s when I bought it in September summer.

All right. Let's demo. Let's do some demo. Let's do some time. You'll need it. Demo, done. Okay. All right. Here we go.

this one is really cool. Marbleism. So basically, it's prompt to fully deployed app. And so it takes a few minutes because you have to fill it out. They give some examples here. I did one previously. Someone made like an Airbnb-like site. You can make a Twitter clone.

You can make your own Asana. And so I went through the process of making a Twitter clone because we were going through it. And so I gave it some prompts and, you know, it has my app ready. And there you go. I have a section that had like a user feed. I can post messages. I can create a profile. And so what's awesome about marbleism is it's taken all the little pieces that we've

been talking about over the last year and just end-to-end platform. Love it. Yeah. So this would fall into the agent category as opposed to the co-pilot into it. Yeah. So you're type... This is not even beyond that. This is, no, this is like end-to-end.

Project manager. So it's like hiring a consulting firm. Exactly. Yeah. Okay. So this is a... Okay, hold on a second. Now we've got a new thought.

You can hire. You can have a co-pilot work with you. Yeah. The base level of using a language model is asking it a question. The next level is having a co-work with you as a co-pilot. The next is having an agent where the agent goes out and does things for you. And then the next phase is the my-stress. for consulting firm where it puts the whole thing together.

And here is our first example. Yeah. Where it's doing that. Wow. Okay. Yeah. And so you can post messages. And like I said,

I just did it end to it. And look, this is a running app. I can send you the URL. You can go to it. Wow. It's wild. You know,

some of these examples here where you're getting, and it builds a database, it builds everything that's required for it. And so really impressive, really impressive. And I think you can make use this within the launch slash, you know, twist ecosystem. Incubator, accelerator, yeah.

So I do like more people. I think this would be more for Founder University. So let's put this person in touch with Kelly for producers. Let's touch with Kelly and Press. I want this person to come speak at Founder University and show it because at Founder University, we only accept multiple teams with a developer because we want to see the product improve every week. We don't want to just see ideas.

And they typically, the people don't. don't have the bankroll to hire a firm. And if you hire a firm, you're kind of don't have, you're not collecting the talent inside of your own company. So that's interesting. So we kind of take a Paul Graham approach at launch and Founding University of Pre-Exelerator where we want multiple teams because they go further. So we ideally want three, but we'll take two and we rarely take a solo founder.

But we do sometimes, one in 50 maybe. But we want to have a developer on, we want builders on that founding team writing the code. So here I think you could have. people making MVP's and little products. You know, we had great success with people using bubble as but one example of a code platform and web flow and other ones like that inside of Founder University. So very cool. You can apply to Founder.

University if you want to learn, be one of 200 people and how to build a company. Yeah. All right. Yeah. So I think if someone has an idea. I give it an A. I give it an A. Okay.

All right. I mean, I'll tell you why, because we've been waiting for this. And the first person gets the highest grade here. This person beat the rest of the class out. We have people who are doing UX design. We have people who are writing code. We got people who are provisioning servers and doing all that to abstract it and put it into a maestro consulting category.

This is big time. This is big time. This is like having. And somebody could do this, by the way, for marketing or growth where you say, make me a marketing plan and execute it. And it says, okay, my marketing plan is to do ads on Google. Here's your Google ads. Facebook ads.

Miastro. TikTok stories. And you do the mystro. goes, boom, here's your seven part, and here's your PR, here's your contact you should go after, and just lays out the whole thing for you. And that's actually the next level of this, which is giving it a multi-stage complex project, right? And I think if each of the components comes back with relatively clean code, that means you're able to connect the pieces, correct? 100%. What grade do you give it? Yeah, I mean, you know, I was in like the B plus, A minus,

only because I would like to see it be a, like, I'm an engineer, so I'm a builder. So it was like a little bit like, oh, I wish I could do these other things with it. But they can add those features, right? But overall, the fact that you can go from prompt to deployed app, and they kind of did all the orchestration behind the scenes. Yeah, so I'll go to A minus.

I'll give it an A minor. Sounds good. Awesome. We got one more demo. Let's do it. So this one is really cool. This is a Y Combinator company. And you're going to have an appreciation for this.

It kind of gave me like AI Mahalo vibes. What they've done is this team has created a vertical search engine with AI and products. Okay. And so Lumona, right? That's, I guess that's how we would say it. And so I'm just going to pick one of their, you know, pre-selected ones here. And then it's like, before we begin, you can get some more information.

Tell her what the search you did was. Oh, sorry. Yeah. I did. I'm just going to go back here. How do I get rid of acne on my forehead? Okay. Probably a super common question for teenagers, right? And so now it gives me a bunch of high level insights that I can dive a bit more into, which is skin treatment.

Yeah, I tell you if you have oil control, non-combedogenic. Exactly. So I know I can go into these things a bit more if I want to, which I think is exciting. Or you can just say, you know what, forget all that. Show me the product result. And what it's done is it's gone and it's scraped. It's exactly what you said. It scraped Amazon reviews, YouTube videos. Yeah. And Reddit.

And it put it all together. And now it's given me a page that's like a Malha's style. Yeah, exactly Mahalo style page. Yeah. But for skincare. And it went in, but it did all the work with AI. I was like, that's impressive. It's 15 years later after Mahalo.

We couldn't do this when we. We had Mahalo, we did this manually with a Wikipedia style approach where people would edit the pages. And so the pages would get stale. And so then we'd have these rushes where I'd have everybody come in and say update all the pages. I just had literally people getting paid $40,000, $50,000 out of school 15 years ago, which is a pretty good salary, by the way. Yeah. And I just hired people at college. I literally had 50 people in a warehouse in Culver City writing serps.

It was a lot of fun. Yeah. Shout out to all the people who did that. And we were making $10 million a year in Google search ads, but then Google took us that. index because yeah, they're sharp. Yeah. Breaking news right now, meta opens up its VROS called Horizon OS to third parties and says AIS and Lenovo are both planning

meta Horizon OS compatible headsets. So I think meta is behind with their headsets because of Apple. I think they consider the Vision Pro is such a competitor. And here we go. Boom. They are now going to let everybody use their OS to try to. compete and do what Microsoft did with Dell and HP. You know, you guys are part of getting fit.

Just put it out there, see what happens. Put it out there, let other people, yeah. Yeah. Awesome. Well, this has been great. I give that one a B. Oh, a B. Oh, my God.

No, I mean, I'll be honest. I think anybody can hack this together in 60 days. And so what we need to see out of this is proprietary data, how they're going to get users. I mean, B is not bad for just graduating my combinator and having this thing out in the public. But, I mean, the design is terrible. I'll be honest.

Like, this looks on a design, if I was going to give this a rating on a design, I give it like a five or a four at ten. It's a terrible design. I mean, the formatting is terrible. I thought it was pretty useful. No, the logo is terrible. The U.S. is terrible on this.

Maybe it's because we're looking at on the desktop. But the content seems like they did exactly what you're looking for, which is to just give you a top level overview of everything that's out there. And the goal is how quickly. can I get you the knowledge you're looking for? Yeah. From a disparate set of sources, and on that, they've succeeded. Just hire somebody for, I would hire four designers for $1,000 each to redesign that one page.

This is my secret to doing great design cheaply. To allocate $5,000. You give three people $1,000 to just design one or two screens of your app. Most designers charge $50 an hour, so that's 20 hours of work. 20 hours, yeah, okay. Pretty fair deal. Okay, so now I take those three. You drop the weakest way.

You take the top two, you give them feedback. You say, hey, for another $1,000, I want you to make a new version. Here's my feedback on your first version. So you did $3,000, then you did $2,000. And then you pick the best of those two, and you have that person become your designer on an ongoing basis. So it took you $5,000. You got three different looks, got two, two, two-point-0 versions.

You got a final version that you then work on with the one who was the best to work with. So for $5,000, instead of giving it to one designer, do your whole website, you give it to three, and you do two rounds or three rounds. bake-offs. You can adjust it to your liking. But the other thing I find is some developers disappear because, you know, they're designers. Sometimes they're a little flaky. So sometimes you hire five for a thousand dollars each. Only three of you get you the product that you're looking for the date. You say, I just need the product by this date. So anyway, and that's fair for designers too. They like these like quick jobs. Okay, great job. What do you give it?

I thought it just hit on like, again, good execution, end-to-end, verticalized. And I feel like that could be the future for us. My letter grade is, B plus. Okay. There you go. B plus. All right, everybody. If you want to see all of our bets, this week in startups.com slash bets when Sunday and I make bets. And then you can see all the AI playlist. This week and startups.com slash AI should redirect you either to a landing page or to the YouTube playlist. We have an AI playlist there. All right. Sunny, you're amazing. Thank you, sir. You put so much energy into this. You have so many deep insights. And I always like to give GROC a plug. You can go to console. dot rock.

ROQ.com and you can follow X.com slash Sendip. I'm X.com slash Jason. If you are applying to Y Combinator, congratulations, but no, only 1% of people get in. So what I love to do is just tell the audience, if you have your application, it's sitting there in your inbox, you've already got it all cleaned up, ready to go. Email YC at launch.com. Launch.com. We don't have the M. Drop the M. YC at launch.com. Or does your application. If they reject you, that's okay. They reject 99% plus for the elite program. Guess what? I have a program too.

We accept 1%. So if you just forward us your application, that's good enough for us. We'll get you on the phone with one of our team members. You do a quick introductory call. You explain us your vision. I'm going to guarantee anybody who emails YC at launch. Dot co gets that introductory call. You meet one of my amazing team members, and they're going to hear your vision out.

They're going to share it with our team. And then, hey, if it's a fit for you and for us, we'll do a second meeting. And then if you get past the second meeting, I'm the third meeting. And then you get to hang out with me. And I invest in 100 companies a year. and one of the great delights is I get to hang out with those folks every quarter, basically, I do in person stuff. So you can fly out and see me when I'm in New York, Austin, Miami, L.A. or San Francisco.

We hang out for half day, get some ribs, get some pizza, whatever, some sushi, some dim sum, whatever it is. It's all about the food for me. And we hang out and we work on your business. Great delight for me to get to work with founders. So YC.atlaunch.com, Portisher application. And by the way, I'm watching them come in. And sometimes I just hit reply if I really like one before my team anything gets to it. And I just give them some thoughts on what I think. And hey, we accept just under 1%. So if you put the 1% together,

now you've got instead of a 1 in 100 chance of getting into a program and getting 125K, you got a 1 in 50. You got a 2% chance now. You doubled your, you doubled your efficacy. All right. Great product insights too. I will say, have C.J. Kellanac Chinney, amazing product insights.

Oh, thanks, pal. Been in the game a minute. I've been in the game for a long time. I've seen over 10,000. I've done 10,000 in-person pitches from founders. And what happens when you do 10,000 of them is you'll see the same ideas over and over again or the same customer base over and over again, the same, you know, business models, techniques. And yeah, you will see some mistakes and you will be able to give some insights.

Many times I'm meeting with a founder and I'm like, hey, do you know about these three companies? They're like, no. I'm like, oh, okay, you're 25 years old. In Web 2.0, there was this company in Web 1.0 in the DACA America called Webvan. And then there was this other company called Postmates. And then there was this company. And I'll tell them about companies and we'll pull up the Wikipedia page or like, we'll go to the way back machine and look at their landing pages.

And they're like, oh my God, somebody had the exact same ideas me. I'm like, ideas are easy. Execution is everything. How are you going to execute better than these people who failed? And hey, do you know those people and why they failed? They're like, no. I'm like, you ever think you want to email one of them telling me you have the same idea they had 15 years ago and J.

I told you to email them, and then maybe they'll tell you why they failed. Man, it's such a great unlock. Just like when I saw Lamona today, I could, I mean, I could tell them, like, their pages are not designed well enough for people to come back, right? They need to be designed a little bit better. So that's like a model. Sheets of feedback. I would love to have them on the show.

Yeah, I love to meet them. Sunny, anything, did you make rock.com slash startups yet? That should be on your punch list. It'll be there. It'll be there. All right. So, Brock with the Q. Q.com slash startups. We were launching Lama 3.

slash twist. Oh, yeah. That was more important. We'll see you all next time. Bye-bye. All right. Great job.