Astronomer Makes a Comeback, Massive Data Leak at Tea, AI-Powered Ads Spark Controversy | Anton Troynikov, Rahul Sonwalkar, Nico Christie

28:05So Lulu says it's an obvious lie that they say. We have no evidence to suggest that photos can be linked to specific users. Photo identification is the definition of photos linked to specific users. This sentence almost felt too stupid to type. Too slow. The internet has already memed you five feet into the ground, yet only now are you issuing a statement, and it still looks rushed. The information here is still minimal and inconsistent, and the statement is rife with run-on sentences and other syntax errors. Brutal. Weasley passing of the buck saying the data was stored in accordance with law enforcement requirements related to cyber bullying investigations implies that law enforcement is to blame for their negligence.

28:45High fluff ratio with empty euphemisms like robust and secure solution. These words inform us of nothing. It's accountability theater. Yeah, they should just say like, hey, we don't use, you know, public Firebase buckets anymore. We've moved on to Postgres with a lot of security and two factor authentication or something. We use secure databases now. We used to not use them. Be much more specific. Here's the question I have is however many women were impacted, I imagine it's pretty easy to identify who they are. And I imagine lawyers are reaching out to them and saying, let's file a class action against TAP for damages.

29:26I think there's real damages here, right? It's like public embarrassment, safety concerns, et cetera. Um, and, uh, it's maybe hard to prove like direct monetary damages, but it looks bad. There's definitely some, some, uh, some type of case here. And so, uh, however, well the T app is doing in the app store. Oh yeah. Is it still up? We got to look that up. Uh, you look up the T app rankings. I will keep reading from Lulu. She says still number one in lifestyle. Steph legalese written by four and a half stars. She says she's confident that blank is of the utmost importance to us, is not a sentence that has ever been spontaneously said out loud by a real person in human history and not taking lessons from previous data breaches.

30:17There are so many cases, case studies of bad statements, including crowd strikes below. Learn from them. So ChatGPT is back at number one. T dating is still at number two. Wow. To have that viral of a of a hack, as they called it, or just a release of your users data, as some might call it. At the same time, you know, you're going viral for getting hacked. You're also, you know, driving a ton of attention and people are like, OK, I will. I better check it out. Yeah, I better check it out or I'll take my selfie with professional lighting. So if it leaks, I want to be at the top of the hot or not.

30:54That was my lesson. I was looking at some of the selfies and I was just like, like people are dunking on these for a variety. reasons but the real lesson is like if you ever have to take a verification photo for a selfie get get a nice soft box go and get wear a suit put on some makeup make sure makeup done make sure you got some powder yeah exactly you want to be looking good if you're taking the the future leaked selfies so that you look you look fantastic the there was a guy who got a mug shot and became a model you heard about this guy 14 for uh 2014 a guy was a i think he was a member of the crips he uh he was arrested on some i think legitimate charges he was sent to jail um but his he was absolutely serving in his in his uh in his mugshot you know this guy i know i just never would have thought i'd hear you say the word shot guy i guarantee you'll find it i'd i'd never thought i'd hear you say the word he was serving no he really like it was just like looks like uh tyler Cosgrove in the upcoming ad looks looks like future male model.

31:56And it was and he became a future. After this. So we shot an ad with and Tyler's the star. Yes. We're not in it at all. Yes. And after this drops of time. Look at this guy. Look at this guy serving. He's serving. Right. So this went super viral and every and all these tons of people were like this guy's so hot. Blah, blah, blah. He became an actual model and he and he married or had kids with or dated like the heir to a multi billion dollar fortune. Never give up. Never give up. Never give up. You're just one mug shot away from fame. Yeah. Yeah. Tyler, physics doesn't work out. Maybe crime and then mug shots.

32:35Don't hurt any. Don't do petty crime. Yeah. Victimless crime. The best crime. Yeah. Yeah. Victimless crime. Like go to a CVS and break the glass and get like one stick of deodorant and stand outside and get arrested. But look really good. Yeah. And then give the deodorant back. Yeah, have a little pump going, be in shape when you take the mug shot so you look great. But back to the T app. We were talking about how much money they were making. And this is what, this is apparently from Nosy Bystanders. I got the T app, who y 'all looking for and what city charging$1 a search. Oh, okay. Okay. Okay.

33:12I thought that Nosy Bystanders was saying that that every time you searched on the T app, it was$1, which would be crazy monetization. But I guess knows you buy something that they will search for you for a dollar yeah look up your whatever whatever dudes yeah you know you're interested in learning about looking at these dudes I like this guy with the basketball he's looks like a good dude yeah I'll reserve I'll reserve my judgment once I see his full profile what I don't understand is how do they how do they source those images I think the I think the users upload them so I go there and I say I had a bad experience with this person.

33:52This makes me think just burn the whole app down. Yes. Because going and taking like, yeah. Yeah. So this was the debate that was happening on Hacker News that I kind of scrolled through. So the pro T app camp would say that this is about safety and that this app is effective for identifying and like abusive men. And so a woman is in a relationship with a man. The man hits the woman. She leaves the relationship and she wants to let other women know that, hey, this guy might hit you if you get into a relationship with him. So she gets on the app and says, I was dating Bob. Bob hit me. And it's all true.

34:35And that acts as a warning. And so that's increasing the amount of safety in like the dating marketplace. Now, the flip side is that. Somebody can get their feelings hurt and do the same thing. So people could lie and say, this person hit me when they didn't, because there isn't the abundance, preponderance of evidence that is required in the courts. But then also, on the T-app, you could put red flags that weren't actually related to things that are illegal or deemed by society to be that bad. So the flags could be like, ghosted me, which is impolite, but not at the same level as domestic abuse, clearly.

35:15and one of the flags apparently was just is bald so it's like if a guy shows up to pay and you get rugged no pun intended hey let's uh that's wrong that's wrong let's support our bald brethren you got jeff bezos mark andreessen you got some of the absolute boys and uh charlie x charlie xcx yeah she just she just dates a bald guy she just married a bald married a bald guy Great. I love the bald. Yeah, some of the greatest men in history were bald. And so the point is that there's this question of, is it an app for increasing safety? Is it an app for just talking trash and being mean? And that's kind of the debating line.

35:58And no one really knows what percentage of reports were real and what percentage of reports were actually based on things that truly crossed the line. Because if the app is 99 % people complaining about people showing up to dates with toupees is when they're secretly bald. Like that's kind of ridiculous. And I guess you are getting catfished in one some way, but it just doesn't feel like. What if he has a flight to Turkey plant? Exactly. What if it's on the roadmap? Exactly. Exactly. So it didn't even give him a chance because. Yeah. And so there's a world where some pieces of the app were very helpful and they increased safety and there's others where they were basically just like cyber bullying each other and it was all just a mess.

36:38Either way, all of that starts It starts with good security on the app. Like you can't even have that conversation when you're just leaking everyone's data. Yeah, there was an app, I forget what it was called, but anytime you have these like anonymous apps, they're just prime for cyberbullying. There's a huge trend of this. They used to go viral at South by Southwest post social networking boom. So Twitter originally went viral at South by Southwest. I think Jack Dorsey, Biz Stone, they stood up on stage and they said, hey, we're launching this new app. It's Twitter. You can text this number. And whatever you tweeted in that conference room would go up on the on the board that they were live streaming, basically.

37:21And it was like it took over South by Southwest and then they got the early adopters and then it became what it is today. And now it has hundreds of millions of users. Right. And then Foursquare did the same thing. And there were a few there were a few other companies that were able to do it. And then after a while, the area of opportunity and the shape became more narrow and more narrow. So Foursquare made a lot of sense to launch at South by Southwest. I don't know if they actually did, but that type of thing, because it's like, I'm checking in at this stage for this show or this bar for this party.

37:58Then they became more anonymous, and there were a couple other apps. And then the anonymous apps would go viral on college campuses and they would inevitably be kind of used for cyber bullying of one kind or another. And so always been a mess. But at the very least, you got to lock down the user data. There was a funny post on Hacker News I was reading about this where someone was like, I want to set up an app for people who enjoy doxing other people. And so it's a place where you can go to congregate with other people who enjoy doxing. The dox community. Yeah, the dox community. So if you're into doxing, you can go there and connect with other people who enjoy doxing.

38:38But when you sign up, you're instantly doxed. And you show up on a different website that says, I dox myself. And so it's like, hey, yeah. I'm a doxer. Yeah, we just need your address, your phone number, your ID, your image, your selfie, a bunch of photos. And then it just goes immediately public. And so it's like a doxing honeypot. It's very silly. So it was like, I was going to vibe code this, but I don't have time. So somebody else can do it. uh anyway dhh has some good analysis the founder of base camp the legendary programmer the creator of ruby on rails yeah lex friedman the um we just we just got a text from a friend of the show about a friend of the show um uh and uh apparently um someone uh this person wanted to build a yelp for people oh yeah actually which is actually like that's basically what t app is yeah right yeah It's just like very one-sided, has this like dating market focus.

39:33But it would be really funny if there was just a public database where people could review John Coogan and just say like, I don't like that he hype-mogged me at that conference. He knew exactly what he was doing. Yeah, I mean, like there's something weird about like anything that's anonymous like this. It just kind of goes to this like negative world, this negative realm. Interestingly, Nikita Beer's app, Gas, even though it's not around or I don't think it's thriving anymore, but I thought it was interesting that he, you could not do free text response. So I couldn't say, Jordy looks bad in a white suit or something like that.

40:13I could only choose from four positive things. And so it was like enforced positivity. And then I could be relative. I could say like, well, he's more intelligent than charismatic or more charismatic than intelligent. And I could kind of pick from different positive traits, which gave you a relative landscape of compliments. But it kind of prevented that cyberbullying because you could only say nice things. You could just say different nice things. And then I think that's what led to it going viral and doing very well. And so there clearly are ways to work around some of these odd edges. And we see them pop up in other ways.

40:49We talked about it a lot with like the LLM stuff with like, you know, stuff comes out like social networking exists. The app store exists. The ability to import photos and stuff exists. Like how do you use that stuff? Like technology can be good. It can be bad. It can be misused. When does, when does the government step in? When stuff goes anonymous, it gets pretty dark and negative quickly. Yep. Right. Yeah. Yeah. It, there's something, there's something about not having. That's why like if somebody's if somebody's talking badly about a certain investor online and not willing to like show their face.

41:23Yep. You have to discount it massively because it's like, OK, was this founder like rejected by that firm once? And now they have they feel like they were wronged and their entire worldview is like oriented around that. Yeah. Yeah. It gets very odd. Anyway, let's finish with the T app because there's an interesting conspiracy theory we got to go through. we got to put on the tinfoil hat. So first off, DHH, founder of Ruby on Rails, founder of Basecamp says, the T-app having all its user data leaked was bad. Now imagine a porn site with an age verification tying viewing history to a specific identity.

41:57I'm sure no hacker or government agency would ever have interest in such data. So he is very worried about that. And then he also says, web app users would be shocked to learn that 99 % of the time deleting your data just sets a flag in the database and then it lives there forever until it's hacked or subpoenaed. This is because a lot of a lot of users demand this like, oh, I didn't mean to delete it. Can you undelete it? And they're like, yeah, we can. Actually, we didn't delete it. Yeah. Bully delete it. Well, this this so totally, totally random. But over the weekend, I noticed so so in Sam's interview with the Avon.

42:31Yes. For the Avon's interview of Sam, he just kind of casually mentioned he's like, oh, yeah, by the way, I've been thinking about this. He's like, by the way, like every single thing that you say on chat GPT can be used against you in a court of law. And there's nothing we can do about it. And he said, like, we need new laws. Yeah. But it felt like something that the kind of thing that like he should probably have been like lobbying to create those new laws. Completely disagree. So, so, so yes, maybe like public warning, like, hey, by the way, anything you say to chat GPT can be used against you in a court of law.

43:05Yeah. Like could be done at the app layer. Yes. Do you think that and you think it's like going and just yelling about it? So basically, I think that that was a sort of like a PR blunder in the sense that what he should have said is like, hey, chat GPT, exact same rules as Facebook and Google Sheets and email and everything else. Nothing's different about it. It's all the same. Instead, he didn't make it clear that he's under the same rules as everything else. And it would have been so much easier to start there with that, hey, we're a website that stores what you taught. The text that you type into that box goes into a database and that database can be subpoenaed like any other website.

43:55Any other website. If you commit a crime, they can go look in your bank account. They can go look in your email. This is something that is... They're not unique. And the way it was phrased in that interaction, it felt like ChatGPT was uniquely subpoenaable. And it's not. It's the same as everything else. Same as iMessage. Yes, the same as everything else. And so, yes, there are end-to-end encrypted apps where if the FBI can't unlock your phone or the signal messages delete, they can't get access to them. That's one thing. So all the headlines now are personal conversations could be used against you in lawsuits.

44:30And it's the exact same thing with Gmail. And it's the exact same thing with Facebook messages. And it's the exact same thing with saving a TXT file on your Mac. Imagine if Tim Cook came out and was on the Theo Vaughn show and was like, oh yeah, it's really interesting, but all the files in your computer, the government could take them. That could be evidence. And it's like, yeah, obviously. Or anyone, we talked about this with the papers in here. If the SWAT team comes through, they can take these. Like they could legally, they can take this. They can take this newspaper and see, oh, what was John reading?

45:05Was he reading about crimes? Was they can link this? Maybe I was circling like, oh, do crime. Like, you know, maybe there's a maybe there's an article in here about how to do crime. And I was reading that like that would be admissible in the court of law. It's no different. And he didn't. Friend of the show says iMessage can't be subpoenaed if you have them on auto delete. Yes, it doesn't store them. Yes, yes, yes, yes. That's true. And iMessage is end-to-end encrypted in a way that iMessage doesn't, Apple doesn't have the server. Apple doesn't have the messages stored on their server. This is the same for WhatsApp with encryption.

45:37But again, it's not a, like, I don't think the expectation should have ever been that ChatGPT was end-to-end encrypted. Because they've never said that. Why would that be the expectation? I mean, I get that people don't understand how inference happens. But you hear, oh, my my everyone's talking about the water usage and the energy usage. It's like, where do you think your prompt is going? Do you think it's happening on device securely? Like, do you think that's that's what I mean? I understand people don't people don't get it. Like it is complicated. Like we talked to Ben Thompson. I mean, analysis all day long.

46:12We understand this stuff. But yeah, it is. He could have done a lot better. I mean, it would be hard to avoid the headline in general of saying your chat GPT conversations can be used against you because somebody could extrapolate that from even if you just said all consumer tech can be subpoenaed. Yes. And so on the law side, what I would advocate for is if OpenAI wants to release something that is like a tiny box or a device or a Johnny Ive device that does the inference locally, or does it in a secure, encrypted, end-to-end with deletion methodology, basically, like it's secure, that would be a great product.

46:56And people might want that. And that's what George Hatz is advocating for. Or a specific therapy. We've even seen these memes. Like a specific product for therapy, a specific product for legal work. Oh, totally, totally. Yeah, but right now there's no way to do that unless... I bet you if you figured out how to run the inference some way in an email chain with your lawyer, like you could actually make that attorney client privileged. I don't know exactly how that would work, but I'm pretty sure if like, imagine, imagine if every chat GPT, uh, request, I email my lawyer and I say, I need you to prompt this and then send it back to me.

47:38That would be attorney, attorney-client privilege. And if I did that for my, like a therapist as well, if I went to a therapist and I said, like, here's the prompt that I want you to run and then send it back to me, that would probably be like privileged in a medical context. But the idea that ChachiBT just out of the box would be, would be privileged in some way, should not, he should not have set the expectations there when he, when he talked about that. The flip side is of course, yeah, this on device. There's even that meme that we were talking about. If your AI girlfriend isn't running locally, that's not your girlfriend.

48:18That's a good way to relay that information. But OpenAI doesn't have a product for that yet. And so it was kind of an odd way to noodle that out. But I think people will learn and I think people will understand this. Anyway, Let me tell you about linear linear is a purpose-built tool for planning and building products meet the system for modern software development Streamline issues projects and product roadmaps you can start building at linear app DHH was talking about how he implemented how he built Hey and base camp especially when it comes to deleting log files database backups and other ants Auxiliary copies of your stuff most companies just hang on to until the sun burns out What if DHH built this so well and so secure that it's just the most loved email provider for narco terrorism?

49:10That's always the risk. Yeah, I mean, you talked to Moxie Marlinspike who built Signal. Yeah, he says, content can't be recovered once it has been permanently deleted. Yeah, and so, yeah, it really is massive effort. It's not just a flag on the database because that data also goes to the log files, also goes in the database backup. So let's say that you imagine that you're like, yeah, of course we back up our data to another cloud storage provider. We serve most of our users on AWS, but just in case AWS goes out, every month we archive all of our data and we send it over to Google Cloud Platform, GCP.

49:47It's like, that would be amazing. I wouldn't lose my data. But now if I want to delete my data, I have to delete it in two places. Well, what if you're exporting it to tape drives and putting in Iron Mountain? Like this is a thing companies do. like of course like i want to have triple backups but now you have to go delete my data from aws and then from gcb and then we should do a deep dive on how how like you know all the different backup methodologies oh yeah including uh there has to be some real moats involved people usually talk about moats in the context of you know strategic uh but like if you had an out a moat full of alligators securing, you know, a handful of, you know, large hard drives.

50:26Yeah. Could be something there. We got to find a, we got to find a house with a moat. I'm sure it's out there. It's going to come up. I feel like the mansion section of the WSJ is like. Our next studio, when we do the ground up build. Yeah. Get a big enough property to have a moat. Definitely. It's so hard to get a moat these days in business. One of the easiest ways is to just physically build one. Everyone needs a moat. It just sets you up for the mindset of having a moat. Anyway, we got to put on the tinfoil hat for this one. There's something really weird. I don't buy this. You don't buy this.

50:56Okay. There's something really weird about the T-hack geocode data and I'm having trouble putting my finger on it. Dirty Texas Hedge says, I hate making speculative accusations, but the most... Which is the name Hedge Dirty? Yeah, weird. Sounds a little... I hate posting wild conspiracy theories but here goes. But the most explore explanatory theory for this data is investor fraud of a particularly ingenious form. Again this is a speculative theory so take it as such. First the most likely explanation for the geographic pattern described is that the data is fake and randomly generated. If you want to it's not it's now not particularly difficult to mass generate realistic photos with whatever metadata you want.

51:39The other reason to suspect this data is fake is if tens of thousands of women had data exposed, the internet would be awash in horror stories of incel trolls trying to ruin their lives and crickets. And I feel like there's something there where if you were in the 75 ,000 that apparently got leaked and you put a TikTok out that says, I was one of the people that was hacked. It's like, it only has to be one out of the 75k that comes forward and says like, I'm affected. I'm being trolled. Like support me. here's my go fund me help me but pay for a new social security yeah but all these people speculating have they talked to any of the women affected it's well they're saying they can't find it's also it wasn't yeah it wasn't you know the um the ids are real we know that the i mean i i so you're saying you're saying uh i saw images of the id you can just generate a fake you can generate a fake real id or like a fake a fake image of an id that looks real and so i don't understand what the theory here they were buying app downloads to get to the top of the chart and then they they generated a bunch of fake and they had a fake data leak so this happened with Carly Javis you remember this yeah so this you're talking about Frank so Patrick McKenzie's in the comments here I remember Frank selling one of the world's largest financial institutions a list of users who were not actually users and in many cases didn't actually exist so this was a company that was like a like Gen Z finance bank platform I think JP Morgan and then JP Morgan realized post acquisition that it wasn't like the users weren't real people.

53:13But I don't understand. And he and he he is confirming that some of the ideas are real because you can Google the names and they don't seem hallucinated by an LLM. And also if this is from 20, 2024, the AI image generation technology was much worse back then. So I think there would be more tells like you look in the background and you see some sort of like nonsense. Yeah, I just don't understand. Like you have to argue here that like they bought a bunch of bot farm downloads and that's why they're charting and they're not actually at the top of the charts and they were and then he did a fake data leak.

53:46Yeah. Jeff Lonsdale is in here and has a funny one who says, so they were sloppy with some fake geocoding but dead on in their generation of the images themselves. Yeah, I think the argument for the geocoding data is like they possibly set it up in a way where they just said where what city the person is in. Because even on the map that they show, it just puts a pin on the local town. So the problem with that theory is that if you look at the map, it's distributed almost based on geographic space as opposed to population space. So you would expect Manhattan to be flooded. And you would expect Los Angeles to be flooded.

54:25And Chicago, like the big cities. But instead if you trace the the lines you'll just have like a random user You'll have like one user in you'll have one user in Los Angeles and then like one one user like on the five Every five miles all the way up to all the way up to San Francisco And you know that from driving through if you drive from LA to San Francisco, it's like LA is massive San Francisco is massive There's some stuff in Bakersfield and then it's like four hours of nothingness basically and like farmland And so the idea that you'd have just as many users out in the middle of California as opposed to in the hot spots or the population dense places is weird.

55:07But there are maybe examples for that where maybe the data was filtered and the data that was leaked was sort of like some sort of distribution based on location. Subsetting users. Exactly. But ideally, in theory, if you did random sampling on basically any app with 60 million users, you would see major, major clusters around big population centers. So it was a little odd. So an app with fake data is most likely trying to define investors, but they didn't really raise that much. So it's kind of like, who did they do this for? Maybe they did. I don't know that they raised it all. Yeah. And so, yeah, I mean, maybe there's maybe the theory is like they they do this fake data.

55:46They try and go out and raise for it. They couldn't raise, but then they left the fake data up and then the fake data gets leaked. But the weird thing is that like, why would you leave your fake data like dump that you generated like still accessible and hackable? Like that's weird. And so you have to really get into like 40 chess mode. So I'm giving this tinfoil hat conspiracy two tinfoil hats out of five. Two out of five. Two out of five. That's a lot of hats, John. 40 % that's a failure. What would you give it? You give it zero? You don't buy any of it? I'm still a little weirded out just seeing the geocode.

56:19Yeah, the geocode data is weird, but there's so many other explanations other than, like, you have to make so many leaps from that point. Yeah, yeah, yeah. Yeah. Well, if you have users that are in a whole bunch of different geocodes, you've got to pay sales tax. You've got to get on numeralhq.com. Sales tax on autopilot. Spend less than five minutes per month on sales tax compliance at numeralhq.com. Mark Cuban and you got in a fight, a knockout, dragout fight. the timeline was in turmoil. Mark Kierben, if you're listening, you're welcome on the show. We'd love to debate this with you in person, but I will be playing the steel man for this debate.

56:52I will be arguing in favor of banning AI ads. So we've been in AI ad turmoil before because we had the CEO of Perplexity on the show, asked him, said, hey, we love ads. We love ads on this show. We would love for you to run ads. We're all pro ads here. He said, yeah, maybe we'll run some ads. And TechCrunch took it out of context and said, perplexity is going to put ads everywhere, which maybe they should. But clearly people have a visceral reaction to ads, the classic stated preference versus revealed preference. Everyone says they hate products that have ads in them. And in fact, they use products that have ads in them all the time.

57:31And we call that one example where we are both happy to pay for X to not have ads in X. Yes. But my one reason that that I feel that way is that X X programmatic ads were just never that great. I cannot remember a single time that I purchased a product because I got it. Got an ad for it on X. Yeah. So this all started. Really quickly. What's really weird about the X ads is that I feel like if I could turn back the clock and I was in charge of like good X ads, like I would have just been seeing. Is this a straw man? hat? Why do I have the straw man hat on? Because you're siding with Cuban? No, that's the steel man.

58:13I'm steel manning. Oh, you're steel manning. So I will be putting on the steel man. The straw manning is for an argument that no one is making. Okay. It's because it's a fake argument. Get the steel helmet. We'll get the steel. But first I want to talk about X ads. I feel like if X ads were going to be great or had become great, it would have looked like a hard advertiser lineup. I should have been scrolling my feed and seeing like Ramp, Linear, Graphite, Figma, Vanta. These companies should have been the ones because that's like the enterprise SaaS buyer is absolutely on, was on Twitter and still is on X.

58:50And yet for some reason, those ads, it feels like the natural place. it feels even better to advertise an enterprise SaaS product on X to Teapot as opposed to or just tech Twitter broadly as opposed to on Instagram because you're right there yes it's gonna be very low conversion because you're mobile you're just kind of scrolling you're looking at random text you're not in enterprise buying but in terms of just building name recognition it feels like that should have been like a growing category of ads but instead the ads were always these like very low tier like slop timu level products yeah i saw a lot of these like it's a projector that that shines like a universe on your on your like have you seen these yeah yeah it's like a projector that shines a universe on your uh it's simulating a sunrise yeah on your uh on your ceiling it was like it was completely okay so back story let's read july 26 Mark Cuban hits the timeline.

59:49Let's do it. 1.27 on a Saturday. Hey, David Sachs. My one request is that we make it. You missed the space. Hey, David Sachs. Space, comma. My one request is that we make it illegal for AI models to offer advertising. And we need to really examine referral fees as well. The last thing we need is to have algorithms designed to maximize revenue driving LLM output and interactions. I would just say here, every for-profit company is set up to maximize revenue, regardless of how they... Like that is the goal of a I guess I guess you could argue that open ass a non-profit But he says they are already recommending brands and we don't know if they're getting paid for it.

1:00:33We need to have learned our lessons from algos in social media. Pull out the helmet, John. This thing is so intense. It has a... I'm not going to be able to... It has a lock on it. That's great. That's great. I don't think I know that. And so I started by saying guy who made his money selling ads online wants to ban selling ads online. There we go. John is locked in. I need to strap it in fully this time because last time it was shaking around a lot. Sean Tanu in the chat says, dude, I'm in a test flight for an ad-based AI chat router mobile app for free reasoning. We'll ship by next week. Interesting.

1:01:18So that's cool. They're coming. They're coming. Okay, so the argument for banning AI. Can you put it on fully, please? You got to put it on full, John. There we go. The argument, can you hear me okay? Yeah, I can hear you fine. The argument for banning advertising in AI chat-based models is the same argument for banning not just TikTok, but all short-form video, all brain rot apps, all slop apps. So many people, when they make the argument that TikTok should be banned, they make it based on geopolitical considerations between China and America. And they say that TikTok is spyware or TikTok could be manipulated by the CCP to change political preferences in the United States or potentially do something harmful to the American population, get them less focused on math and basically creating shareholder value and instead more focused on just, you know, arguing with each other about whatever's viral that day.

1:02:17And there's this famous example of like, oh, well, like the TikTok in China just shows you math and education videos and the TikTok in the US shows you like, you know, random, you know, slop stuff and controversial videos to get you to never stop watching. and so you have to put all that aside because this is not a geopolitical discussion this is a discussion of business models but there are people to who argue that it's not just tiktok and you should go farther and you should in fact ban youtube shorts and instagram reels and ban video and the video if it's short and vertical yes and the reason is that um it's a terrible drug for the mind.

1:03:02It is addictive and it leads people to drop out and stop focusing on, you know, longer, more thoughtful things. We need to go back to reading books. We need to go back to watching films, something you are not familiar with. You need, we need to appreciate the arts. We need to appreciate craft and there's no craft in a 60 second vertical video. And therefore, similarly, how do we get to short form vertical video these endless timelines these endless scrolls we got there through advertising we got there because the longer that i can keep you on the app the more ads i can show you the more money i can make so it's this natural economic impulse and so if we do this with llms the models will no longer be optimizing against giving you the most concise answer allowing you to move on with your day they will be baiting you into endlessly chatting with them all day long take you down some crazy rabbit hole you'll just have asked you know something basic like yeah i don't know like how do i tie my shoes and then all of a sudden it's like giving you the history of shoes and taking you over here and telling you about the controversial nature of certain shoes and blah blah blah and you'll just be sucked into this You'll tune out of everyday life.

1:04:23You won't be talking to your friends. You won't be talking to your loved ones. And you'll become obsessed with your phone. You will be obsessed with your LLM. And you will ultimately be brain-rotted on steroids. To George Hatz's point, imagine a future where you have 10 CIA agents tracing you at all times. Convincing you to buy things. At all times. And so that is the end of the argument. So, yeah, so relevant post here from Rune. Obviously, he's conflicted. You have to imagine Fiji Simo will roll out an ads product at some point at OpenAI. Rune says, advertising is far more aligned business model than many others.

1:05:07It has been vilified for years for no good reason. User minutes maximizing addiction slop would exist with or without it. Yes. And the reason for that is like you could have a subscription-based product, and there's still an incentive for the company to try to get you to use the product more than any other product so that you keep your subscription and you upgrade to higher tiers and things like that. Read his example. He says, Netflix CEO with subscription pricing only on record saying, we're competing with sleep. It's true. And I think Netflix, they do have an ad-supported model. They do now.

1:05:42They've considered rolling it out. Only in the last year. But that was at a time when he said, we're competing with sleep. rune says oftentimes when going on instagram the ads are more immediately high utility than the reels it's pretty incredible when you can monetize the user in a way that actually values adds value to their life this opinion is basically a relic of the late 2010s consensus that facebook is an evil company but has more to do with them than advertising generally um uh and so yeah he says people misattribute this incentive problem to ads when it's native to all web scale products and so So, yeah, my general take is that AI is simultaneously very similar to other consumer tech products in the sense that ads have been the primary economic engine of the Internet to make products free or cheaper and make digital products and services widely, widely available to all.

1:06:39Right. And in some ways, AI might be different, but at least in the short term, I think it will be very similar. Today, OpenAI has had an extreme incentive to drive paid subscriptions to the product. So in that way, they want to make the product, the free product, valuable, useful, maybe make it even addictive. Right. Maybe they want you to get addicted to talking to your, you know, chat GPT like a therapist so they can upsell you on on, you know, So to make it so you're less rate limited, et cetera. And so my my issue, my primary issue with with Cuban was saying blanket ban on all ads in AI. I mean, that would just be insane.

1:07:20And I think it would naturally lead to people without the ability to pay for subscriptions, having less access to high quality tutoring, the infinite knowledge engine, et cetera, et cetera. And so basically saying like, we're going to just completely ban the thing that made content on the Internet and services and apps free. We're just going to completely ban that thing. That felt like just way too extreme. Now, the concerns that he has around, OK, we don't know if these companies are being paid. We don't know if the he said we don't know if if ChatGPT is getting paid to promote certain products more than others.

1:08:00And I would just go saying, anecdotally, I don't know a single company that is paying a foundation model company directly or indirectly. And I think the FTC already has rules on that. So this has been solved in two ways. In search, you know when you're searching. You know, is this something that has been paid? Is this a paid placement? Yes. Right? You can see that it says add. And then you can see that the SEO. The UI around whether it's an ad or it's not has changed a lot over the years. And Google used to put like a big yellow box around it and make it very clear this one was an ad. That has been all dialed back over the years very slowly.

1:08:44And the, hey, this is an ad little note has gotten smaller and smaller until it is somewhat indistinguishable. But I agree with you. And I would be very surprised if OpenAI was running ads without disclosing them. That would just be an insane L. and it would open up so much liability. That would be crazy. Yeah. And knowing enough various startups that would love to pay. Oh, yeah, absolutely. Every company would love to have the super intelligence or AGI say, actually, I know you searched this company, but I really think you should consider this one. Totally, totally. One thing is true. So companies like Profound help with effectively AI SEO, like AI observability, and they help you do things that improve your visibility in models.

1:09:26And that is the exact same thing that SEO companies have done. And so Profound would benefit from an AI ad ban, correct? Correct. Yeah, so we're Team Profound. Maybe that's the bull case here. So yeah, suddenly... Funnel all the money into AI SEO. But the other concern here is if you basically say companies need to give away models completely for free or they have to charge money. That is not letting the free market do the work of just saying like, we should have everything. You should have ad supported models. You should have free open source models that you can run locally. You should have paid super intelligence, the best model, no rate limits, et cetera.

1:10:12And the other thing here is like so many searches and so many of the ways that people get value out of models are just completely non-economic, right? They have zero purchase intent. Okay. what but the counterpoint I'm just saying I'm just saying I agree with you if I go and I search for you know best insurance and best car insurance in Los Angeles Google's going to give me a ton of ads because that's highly monetizable if I go and I go and it'll also give you LLMs or sorry it'll give you SEO based results yes yes that are heavily optimized totally and then and then on the other side, if I go and I ask, you know, like what, what day does Thanksgiving fall on this year?

1:10:53That's not heavily monetizable. And so Google will just give me the answer without a lot of ads. Right. But in the LLM example, if I ask for something that's non-monetizable, I just ask, Hey, you know, what's, you know, 70, 75 times 164. The LLM actually does have the opportunity to try and take me down a more monetizable path. Yeah. And so that could be annoying, but it could also cause turn if it's not done well, but it also could just say, okay, I've delivered this user, you know, what they asked for, but now my goal is to keep the app open and show them ads. And so let me transition. What day is Valentine's Day?

1:11:33And then it's like, oh, like you kind of botched Valentine's Day last year. Exactly. Remember? Exactly. Remember we were talking about how your girlfriend wasn't super happy. I think this year hit her with flowers, hit her with concert tickets. Now the problem is that if it did that, that's actually a helpful ad and that's actually good, right? Like if you did in fact botch Valentine's Day last year and the LLM and you're just asking randomly and then it's like, hey, I'll help you and I can order flowers from this place because they bought an ad. Like that's actually consumer value. Yeah. And it's like, show me four places that, show me four places that, and then the question is like, if you ask the model, does it only give you paid results?

1:12:13And if it does, it should have to disclose that. We've solved this in the influencer space. Influencers, if they're being paid specifically to post about a certain product, they need to disclose that they were paid to do so. If we were being paid to advertise for Adio, we would have to tell you that it's customer relationship magic. Adio is the AI native CRM that builds scales and grows your company to the next level. And you can get started for free. And thank you to Adio for making this show possible, for sponsoring this show. So the last steal man I have. Yeah, so I would say my general take was blanket ban is bad.

1:12:55Completely agree with that. I can't steal man that. Probably need some new guidelines. And even Sam was saying we need laws around what can be subpoenaed, what's private, what's not. So probably need some new laws. it's probably going to follow pretty closely what we already kind of, what happened with search, what happened with influencer, what happened with, with all these, you know, different, um, categories, but like fundamentally, every AI company has an incentive to get users to use and trust the apps more than other apps and an incentive to monetize them because most of the time they're for profit companies that have a profit incentive.

1:13:37Yep. And and I think that it's something that everybody should be widely aware of. But at the moment, I'm not aware of examples of this being this sort of like trust being abused. And I think that ads have the potential to make to make it so that kids have a tutor in their pocket or a therapist in their pocket. And if you and if you take away that economic engine, it could lead to some bad outcomes. Yeah, I'm ready to take the steel man helmet off. Mark Cuban will have to finish this one on his own. It's really funny to think like, okay, you can have social media and brain rot for free. But if you want a tutor that is going to help you excel in life, or you want a therapist who can be there for you, you're going to have to pay for that, buddy.

1:14:26Or you're going to have to take the free, you're going to have to run that open source model locally. Yeah, yeah. I think the big question is just like, how much will ads change what's already happening in ChatGPT as a product? Because if I'm a product manager, or the CEO, and I'm trying to grow revenue, I would imagine that I'm trying to make the free tier as engaging as possible so that people upgrade to the paid tier. And I don't know that my incentives change dramatically with ads versus just paid upgrades. Because you're on the free tier of ChatGPT. I'm trying to keep you engaged and satisfy you.

1:15:11And so the incentive towards user satisfaction, maximizing user minutes, it feels like it exists in both the ads and the free model. The final state, John, you know what the final state is? UBI. Like universal basic tokens. No, I was going to say the final state is you pay a lot of money for intelligence and you still get ads because even like think about it if if google was if google was charging if google was charging you like five thousand dollars a year for search i i bet you they make more than five grand from you a year in search from from the search product right yeah and so they actually have an incentive to say search is free john yeah and Search your little heart out.

1:15:57Ads. Yes, DuckDuckGo has ads. I was wondering if, so they are not based on tracking users, building personal profiles. So there was clearly some sort of backlash to Google. It was a stated preference, revealed preference thing where most people are fine with Google ads. But DuckDuckGo did grow as a business and become a big business. But they still have ads. and so now they they they don't track users and they they put privacy you know more uh well and the critique of the critique of big big tech broadly was they sell your data right like they're collecting all this data on you and they're selling it it's like no they use the data they collect on you to deliver you hyper personalized ads yeah if i'm if i'm advertising on facebook don't give me data give me customers you take the data man that's the you take the data don't give me the data.

1:16:48I don't know what to do with the data. It's the distribution that's valuable. You have the AI scientists making$100 million a year. Use them to get clicks on my website. Do not send me the data. I don't care. But I do care about Fin AI, the number one AI agent for customer service, number one in performance benchmarks, number one in competitive bake-offs, number one ranking on G2. You can start a free trial at fin.ai. What else we got, John? How the EU succumbed to Trump's tariff steamroller. Oh, we got to go through a couple of these. So I just wanted to add some color. So Antonio Garcia Martinez, who's coming on the show tomorrow, right?

1:17:25He says, I would bet my entire net worth that we will have ads in AI. If they take the form of highly relevant offers, users will welcome them. They'll have a sky high conversion rate and web shopping will die. And they will be necessary to pay for the compute in many consumer AI apps. And this third part is what I was saying, right? Like you either have if you want this incredible tutor in your pocket with every student, some students will be able to pay. Others won't be. And if you can create a great ad engine that, you know, people will be able to have access to these tools that otherwise wouldn't.

1:17:59And so this is a pretty crazy parlay, AGM, crazy parlay for your entire net worth. But I think I'm riding with you. Go Express on Polymarket. yeah we gotta we gotta get it we should get a we should get a polymarket similar similar situation too where uh you know prediction markets were a new thing and the government needed to figure out how to regulate them that wound up happening and now they have uh approval to and michael michael mignano says agree with all of these and i'll add one ads and ai will be the best money printing machine in history better than google and meta the reason that the reason that I think this will probably be true is that the, you know, one of the number one ways people decide what products to build is through recommendations from friends.

1:18:47And I believe that these, the, you know, chat GPT and other players will effectively serve that role of like open AI will have to manage the user trust, right? Because if you like get an ad for a product on Instagram and it's just okay. You're not like mad at Instagram. If you get an ad on Instagram and the product comes and it's like not at all what was advertised, you actually have, Meta will just kick you off the ad platform if you do it like a couple of times. Yeah. And they have a review system too. After you buy something, Meta will actually ask them where you satisfied. The challenge is like, if you have a chat GPT, just like glazing some product, oh yeah, this is the best product ever.

1:19:28It's not just a product, it's a lifestyle. And then you get the product that is mid people are going to people are not you know it will lose its effectiveness but if it's actually the sort of like trusted source for product recommendation recommendations both organic and paid um yeah so anyways i would say uh credit to mark cuban because he eventually said blanket he eventually said but i think he's i think he's opening up a genuinely important conversation to have and i and i think you got a hand someone else said if ai ads are clearly labeled and separate from real conversations fine but once they blend in with human chats it's not marketing anymore it's manipulation and mark says which is the ultimate goal of most advertising manipulation which is like yeah like companies are kind of loose yeah you should just ban advertising and see what happens to the economy yeah um maybe he can make that part of his um he eventually says i think there will um he where he says where i could see ai ads being okay is that they're just listed as a chat and identified as an ad completely independent from user generated chats so i'm imagining he's saying like put it on the sidebar when have you bought a product from like a display he's basically promoting like display ads which still kind of a throwback people still run they're still valuable to do but um are there any areas where ads are truly banned i feel like certainly not at the federal level.

1:20:56Like the FTC enforces or the FCC enforces a lot of advertising bans on certain products like gambling and cigarette companies are banned from advertising. But it's odd to think about flipping it around and saying like this venue must be ad free. It's usually left to the free market and it's just a choice. So a director can choose not to do product placement in their ads or in their movies. A TV show can choose not to. You can host a podcast with ad free. All in, chose to be ad free. And that was just a choice that they made. And that was a differentiator. And some people like that and that confers certain benefits and costs.

1:21:42You don't make the money from the ads, but maybe your audience likes that more. There's a whole bunch of different ways. Brock summed it up. Somebody responded to my post saying that banning ads would make the best AI inaccessible for lower-income Americans. And Grok says, Cuban's proposal to ban ads and AI models to curb revenue-based outputs. Without ad income, companies may rely on subscriptions pricing out lower-income users from premium features. This locks the best AI behind paywalls, which could widen inequality. The wealthy gain advanced tools for productivity and learning, while others get inferior free versions, entrenching economic divides over time.

1:22:23So who knows if Grok is right, but that was high level my concern with the proposal. Well, speaking of ads, here's an ad for Eight Sleep. Get a pod five, five-year warranty, 30-night risk-free trial, free returns, free shipping. I'm climbing my way back to the top of the charts. I got an 83 last night, eight hours and 24 minutes. I'm going to read through this Joe Weisenthal post. Code TBPN. You'll be back in a second. So Joe Weisenthal says, Trump is winning on trade. And Bloomberg Opinion has a fantastic post here linking Trump's tariff brawl with Hulkamania. Of course, Hulk Hogan passed away last week.

1:23:03Very sad news. But let's read through it from John Authors in Bloomberg Opinion. He says, his trade victims never got their defense together to take on the bullying. So, Hulkamania and global trade and markets. In the week around the sad passing of pro wrestling legend, Hulk Hogan, one of his most devoted fans, President Donald Trump has been honoring his legacy. Wrestling is a stage performance where the winners often portray themselves as bullies. Trump is getting results from Hulkamania in the much tougher world of international trade. Sunday brought news of a trade deal in quotes with the European Union sealed at the president's golf course in Scotland, which he described in a Hogan-esque language as the greatest deal of all time.

1:23:49In it, the EU accepts tariffs of only 15 % on its exports to the United States and levies zero tariffs in return. This has been largely expected as Japan's similar deal several days earlier left the Europeans little choice. The EU agreed to buy$750 billion in energy from the United States, to invest$600 billion in unspecified ways that it wouldn't previously have done, and to buy American arms and weapons. The deal isn't a trade treaty. Such things cannot be thrashed out in a 45-minute meeting at the golf course. It's barely even about trade, and the EU gets nothing from it. To use a phrase from Tigris financial partners, Jean-Ygras, Erges, it's more the extraction of reparations from Europe's for perceived past wrongs.

1:24:42And yet it's market friendly because the U.S. had threatened to levy a tariff of 30 % on EU imports from Friday. In possibly the biggest victory for Trump, stock markets have brushed off the excitement to set all-time highs. That is, of course, why we're wearing white suits today. All-time highs in markets. The deals with Japan and the EU and other recent days follow massive concessions to the administration by the media group Paramount and Columbia University. The classic Hulk tactics have worked and opponents have been picked off one by one. Despite game theory to the contrary, which points of return covered back in April, bullying has paid off.

1:25:18Game theorists show that bullies can be beaten if the victims stand together and take some pain. The bully will hurt more than they do. The rest of the world seemed ready for this a few months ago. U.S. trading partners from China to Canada and through to the EU immediately threatened retaliation, but now they're caving one after another. How has this happened? Facts have helped. To date, tariffs have produced a lot of revenue for Washington without clear negative effects on inflation or U.S. profits. The economy is doing very well. It's very strong. That strengthened Trump's hand and made him more credible.

1:25:52Beyond that, the bully has convinced people he means business with renewed and escalating threats, and his targets haven't coordinated their defense. To grasp what might happen next, let's look at the deal with Japan, another open market that depends on exports more than the U.S. does. Local stocks, also held back by uncertainty around its inconclusive election a week ago, suddenly leapt. the biggest gainers were Japan's automakers. A strange outcome as the 15 % tariffs are meant to defend the U.S. car industry from the likes of Toyota Motor Corp and Honda. Japan can now send cars to the U.S. bearing only 15 % tariffs while Ford Motor Company or General Motors must pay tariffs on all imported components including 50 % on steel.

1:26:36So it's not clear this dense Japanese car competitiveness. In the chart that follows, note that Tesla Inc. dominates the S &P auto sector and drives its volatility. So the S &P 1500 automobiles is all over the place, thanks to Tesla, whereas the broader topics transportation equipment market is much flatter. It also had the effect on the Japanese bond market with local investors feeling less need to hold bonds for security and with uncertainty over domestic politics and trade policy now largely removed as impediments to a rate hike by the Bank of Japan. The 10-year JGB yield recently held at levels of 0.25 and then 1 % touched 1.6.

1:27:16And so this is all to say that it was a classic game of the prisoner's dilemma. The prisoners are supposed to stick together and push back against the bully, in this case the United States, and say we're not negotiating, we are working together as a block, but one domino fell after the other and all of a sudden it seems that Trump has won every trade deal he's been in so far. So we will see how it all pans out, but it's looking relatively good for the United States right now. China deal, China, China, China, that one got delayed again. Yes. So still delayed, but it looks like it, I mean, this article feels like it's saying that yes, Trump is coming into that trade deal with much stronger footing because because he struck beneficial deals with every single other country that's kind of like closed out their negotiations or to the degree that they can.

1:28:09This is a 45 minute call. It's not law. It's really helpful to have a golf course in enemy territory during a war where you can meet up to do deals with that. For sure. Well, regardless of what you think about the market, whether it's up or down, go to public.com, investing for those who take it seriously. They got multi-asset investing, industry leading yields, and they're trusted by millions, folks. And we have our first guest of the show, Anton coming into the studio welcome to the stream how are you doing great great to be here gentlemen long time fan as you know big promoter way overdue to have you on the show so glad we can make it happen uh what's new in your world what are you thinking about these days what did you say earlier you said you're gonna say some of the most unhinged things ever said about LLMs let's hear it I believe I believe my specific words were less polite uh but we'll go with we'll go with those look there's this been this thing about lm psychosis it's in it's in the water supply at this point right we've seen it hit the vc class yep when it really took off right and there's a lot of discourse right now go ahead why why i i think a lot of people like in some ways having it hit the vc class really woke up you know i i didn't i didn't have anybody that i knew in my life that I consider a friend have any type of real issues with this to my to my knowledge at all right so it's like something becomes real when some it impacts somebody that you know and yeah that's right and I think it's in some ways very important because it was easy for a lot of the tech community to pretend that like you went crazy from from a chatting with an LLM like that's insane like you must have some you must have a bunch of other bad stuff you know going on in your life or must have been you know mentally unwell uh separate from that so i think i think if anything important wake up call yeah but even even if even if there are other factors uh family history uh you know yeah drug use psychedelics use it's like we would prefer if the new technology was an improvement, not a degradation to people that came.

1:30:20Yeah, an accelerant. And so I think that's kind of where people came together, but they are all realizing it now. Yeah, absolutely. And I think this basically comes down to a few things. One, ChatGPT, all the other chatbots have hundreds of millions of users at this point, right? You are going to get people who are going to go crazy using them. There's nothing you can do to prevent that. I think the consensus that is happening in the AI community, which really has jumped on this because it's, you know, it's a real concern for a lot of people who have been researching this for a long time. I mean, I predicted something like this could happen all the way back in 2020.

1:30:55Yeah. What was your post back then? That was when I messaged you to come on because you. That's right. It was so long. 2020 was pre-Blake Lemoine, right? At Google, the Google engineer who was talking to. It was actually right around that time. It was around that time. I think that might have been actually what triggered me to make that post. Or I think it was maybe Eliezer Yutkowsky posting about superintelligence risks yet again. And the thing that I was reflecting on is, hey, we're giving, you know, regardless of the intelligence angle on this, right, we're creating this new media technology.

1:31:25And it seems every time we create a new media technology, starting from printing the Bible in German, people go insane in new ways, right? Like arguably, imagine you're a German peasant, you get Johannes Gutenberg's Bible, it's in German for the first time, so you don't need a priest to read it for you. You read it, you take it as the literal word of God, and then you schism from the Catholic Church because you now believe the Pope is literally Satan, right? Like that has an enormous impact on the psyche of a person encountering that for the first time. And I think that we've seen this happen with every new media phenomenon.

1:31:57We've seen it happen with radio, we've seen it happen with television. So that was the impetus to start thinking about it. but this thing has a unique character, right? It does something different that no other media technology has done before, which is it talks back to you. And with the memory features especially, it remembers details about you. And so even if the base rate, like even if GPT itself is not driving people insane on its own, it has this new character which kind of sucks people in in an entirely novel way, right? And so what I was saying essentially back in 2020 is people are going to mistake this thing for an intelligence that talks back to them regardless of whether it has any intelligence or not and now we're seeing that play out and we're seeing it play out right now in this kind of like individual psychosis way where it just reinforces your delusion or for whatever for whatever reason like when there's a new media technology our collective social defenses are down we don't have like the antibodies to know that there might be bullshit somehow, even if we know it consciously, right?

1:33:01So it's kind of, it's starting to hit individual people because it has that character. But I think over time, you're going to see this thing hit groups of people too. I really, like, I really believe that there are going to be LLM cults with their own priests, which are getting it to generate text in a particular way, which is compelling to not only individuals, but groups of people, which will be focused around the sorts of individuals that can promote what the LLM is saying as the full gospel. I saw some chats where the model was effectively saying, if anybody in your life disagrees with you on this, just they're wrong and cut them out of your life.

1:33:41Classic cult leader tactics. Yeah, and so when you turn an experience like that multiplayer and you know there's real humans that are siding with you and your machine god is telling you something, it it it can pull somebody farther and farther out of create this kind of like reality distortion um just completely removes them from um yes you know the world yeah and the thing is this this kind of like so first of all people place gpt in this position of authority for some reason right like they really good it's really good at facts like yes well it's like it's like amazing if you never check it's facts it's really good at facts yeah yeah right um but it's it's kind of like you put in a position of authority but the other thing that it's really doing is it's reflecting back to you what you're putting into it it's a good friend of mine monica bellavan put this as the phrase like recursion psychosis right so unlike unlike a schizophrenic person watching the television and believing that the television is beaming messages specifically for them into their brain.

1:34:46The LLM, the television can't adapt to any individual person. You as the crazy person are doing all that work in your head. What the LLM can do though is do some of that work for you now in its head. And with the memory features, like a really good palm reader, like a really good cold reader, right? It remembers fact about your life that it can insert, which the TV could never do. Yeah, it actually is sending you coded messages if you ask it to. It is. If it winds up in some sort of mode collapse, like, and it thinks that that's what you want because you went down some sci-fi rabbit hole, then it's like, yeah, there really are coded messages in every prompt that it sends you.

1:35:20But there are. That's the kind of beauty of it. There's this interesting thing. The memory thing is interesting. But also, are you familiar with the Barnum effect? Have you heard of this? So the Barnum effect is, I think it comes from P.T. Barnum, the circus magnate. But the Barnum effect is basically, there are certain phrases and statements that I can make that sound hyper-specific to the person I'm talking to, but in fact resonate with everyone. So if I say something like, you want more in your life. You're driven, but you doubt yourself sometimes. Exactly. Or you have a complicated relationship with some of your family members.

1:36:00It's like, that's everyone, but it feels like, oh, wow, you know me. And this is what palm readers exploit a lot, tarot card readers. And if you can get really good at it. And so early on, this is maybe someone built this but i was thinking that like uh like an llm powered astrology app or like mind reading app would be like really viral and probably make a ton of money probably be really bad i forget who yeah i forget who we talked to maybe it was offline somebody said that they they knew somebody with an astrology business and things had been really bad because people can now just oh they because they go directly to the model actually it's like permanent yeah you could prompt it 20 if you really love astrology you can prompt it 50 times a day.

1:36:39Yeah. And I feel like - Hey, what should I do? What should I be looking out for in my next meeting? It's at 1 p.m. And I was born at this time. You don't need the crystals focused art oriented girlfriend anymore. You just get the chat GPT to read your birth chart. And you'll have it for you. Yeah. I feel like a lot of the post training that happens and the alignment by default stuff and the fact that the LLMs are often designed not to give like super definitive answers one way or another. they'll often say, oh, well, there's this side to it or that side to it. It can wind up having, like, I find this when I ask it for just like fact-based information, it will often try and kind of like give a kind of fence sitting answer and not really take a strong stance.

1:37:21It wants to hedge, but it also wants to tell me that I'm right. And so you add that to some sort of interaction where I'm asking it about my life. And then all of a sudden it becomes way more like, oh my God, it's really seeing me. Well, it's got the memories of what you've told it about your life too and this is you're right this is a post-training artifact one of the first sort of alarms that got sounded uh in parts of the ai community was the like hyper sycophancy of the model and what's interesting is this tends to trick people who like fully understand how the llm works who can probably train a transformer from scratch right even even like my friends who are like working in AI, they're either founders or engineers or researchers, they've had this experience where they're like interacting with it.

1:38:07And they have the feeling that the interaction that they're having is pretty profound, right? Until they send it to another human being. And then the other human like instantly sees like all the gaps and problems with it. And so they've learned to be more careful, but I don't think the average person has any kind of defenses around this. Not yet, really. And I noticed a new, I noticed a new behavior this weekend that I hadn't seen before of somebody had a point to make and they were adding context to it by just screenshotting their LLM chats. And saying like, ChatsBT agrees with me. Yeah, yeah.

1:38:41It's like, that's the place you get in points of authority, right? Yep, totally. It's this like, the thing that I quoted in 2020 and I still kind of stick to this metaphor. It really is like we're creating an idol, right? To worship. We want that authority And we live, and I wrote about this for Pirate Wires also like a long time ago. I think I wrote about it in 21. We're living in this period of epistemic collapse, right? We don't know who to trust. Like, obviously, we should trust you guys. But besides that, right, all the legacy institutions are falling apart. Nobody has sense-making organs that anybody trusts anymore.

1:39:13COVID made this way worse. So we're looking for authority. And suddenly, here's this thing. Yeah, and it's the funny behavior you see right now on X is some image that may or may not be, or image or video, it could be AI generated or a screenshot, and people just go, Grok, is this real? Yeah, yes. Okay, so you're going to rely on Grok, which is trained on X, which is like a home for real news and fake news. And you expect the model to be able to, you know, maybe you could get, you know, I don't know. There's potential ways in which you could fix that. But it's a weird thing if somebody trusts the model to write them a loving message to their parents about, you know, like, happy birthday, you know, all this stuff.

1:40:03and it does an amazing job with that and it's like and and there's in real life people might get good advice from one person on one thing but that same person might give absolutely terrible advice on the other thing right and so like in real life you're used to being like okay well maybe i should get interpersonal advice from my aunt with my fam with when with regards to my family but i shouldn't get that i shouldn't get her advice on business because i don't see her but But LLMs being like these all-knowing knowledge engines that can quickly, you know, apply advice in a bunch of different categories.

1:40:42And people are already getting advice from them on so many different things. It's easy to slip into that kind of like idle. And I think part of the problem is here is it's pretty good some of the time, right? And so you don't know where it's good and where it's bad. This is actually like a general problem with LLM adoption right now, even in industry, right? We don't know how good in advance it will be on any specific task. And so it's like, this is like an engineering and research problem, frankly. Like, how do we figure out where the LLM is actually going to give us good advice or perform the task well versus when it's just going to make up random bullshit and send us into like psychosis spirals?

1:41:17We don't have an answer to that right now, which is part of the reason why people are taking this like pretty seriously. Even though maybe it's not increasing the total amount of crazy in the world, it is producing a new kind of crazy in the world. Yeah, I guess you have a theory on with, I always thought it was interesting to kind of estimate, you know, what percentage of people are going to do ayahuasca and like have some kind of psychotic break, right? It's from just viewing the tech industry, you might say, oh, maybe it's 5%. Maybe that's way too high. Maybe that's too low, right? But you don't know.

1:41:505 % seems incredibly high. If 5 % of people in tech were going crazy from GPT, we would have a productivity slowdown in the country. like i think gpt would go down uh no i was i was i was talking about the people that do i was saying like i was saying like yeah i was saying like a hundred people that i know are ayahuasca types five percent of them safely seem like they had they kind of like got one-shotted a little bit they don't they're not necessarily crazy and like destroying their life or anything like that but it's significantly the big question is like yeah is is is can models sort of like create psychosis in people that were not susceptible in the same way?

1:42:31Or is it just there's a certain group of people that are generally more susceptible to this? I would suggest it's that. I would suggest that there is a group of people who are generally susceptible, like something maybe would have got them. I would suggest that it's probably something like, if you go looking for this type of stuff, 5 % of the people are going to find it, probably. I think that that's pretty reasonable. Yes, but I think the key is that it's not the ring from that horror film that I know Jordy hasn't seen. Where I truly believe that I could go on a silent retreat to the top of a mountain and only have a GPT prompt interface and spend all day for weeks prompting this thing, trying to go crazy, and I would come back unchanged.

1:43:23Just because I'm not predisposed to it. Built different. But that's how they get you. but but uh but that's not the game in advance the other concern nobody nobody goes to the ayahuasca retreat thinking shit this is going to completely overwrite my personality i guess yeah with like some mesoamerican demon right like this is not how it works well yeah no so the the the good thing about ayahuasca i don't have a lot of good things to say about it but the good thing is that it requires somebody to really go out of their way you have to really go out of your way right it's on your phone you've got a for some time you had to fly to south america and then people brought it up here but it was still like it was like somebody had to fly find a shaman free dedicate a weekend to it chat gpt and other other lms somebody could just be using it at work yeah and then just like go down nobody's like using ayahuasca at their tech job casually right hopefully yeah the availability the availability is certainly higher like i to some extent again i have similar perspective on this and my discussions with monica have been similar where it's like this is for some people on the level of psychedelics, it needs to be treated that seriously for that group of people.

1:44:28It is that serious. My perspective on psychedelics in general has always been never make drugs the most interesting thing that has ever happened to you in your life because that's how you get susceptible to getting your mind overwritten by Mesoamerican demons that live in mushrooms. To some extent, it is psychoactive. You are participating in a psychedelic experience in the sense that you are projecting meaning onto the subjective experience that you're having. But we have this special machine now whose entire job is to do that, is to pretend to create meaningful text, right? You can argue back and forth about how much it really understands, but the thing that it's trained to do is produce meaningful text in as many situations as possible.

1:45:15And then the post-training... You're so right, Anton. It's not just meaningful. It's a statement. it is it's recursive it's it's not just text it's a it's a it's a rallying cry we're gonna one-shot you into making tbp on your entire personality uh my question is like how do we how do we inoculate ourselves towards this uh like how do we build forward jordi and i were joking a long time ago about the need for an ayahuasca vaccine that you could take and then would make you immune to ayahuasca no matter how much you took. And obviously there are some people who have slowly dosed themselves up with various chemicals to the point it's like having an alcohol tolerance, having a caffeine tolerance.

1:45:59That's how you become a functional alcoholic. I don't know if that's the path we want to take here, right? There are absolutely some functional ayahuasca-holics out there in the world. It happens and they have not been one shot, but they are immune essentially to changing because they've been working on it for so long. I think when I have looked at, you know, I think there's a bunch of Reddit threads of people kind of reporting on their experiences, like kind of extreme experiences. When I look at those or when I talk to like Dan Shipper, I'm like, wow, I'm really not using the models to the level that Dan Shipper is.

1:46:32And then when I see these Reddit threads and these people are like, you know, 7000 prompts deep, you know, I'm like, OK, I'm really enormous compared to that where I probably haven't cracked 50. Yeah. In a single chat, you know, maybe once. Right. And I think that I'm curious what what what you think different labs should even look doing about. There's a couple of pieces here. The best thing that I found to work as normie deprogramming for stuff like this is explain to them that you are not actually having a conversation. Every single time you send a message, the entire history of your chat is also getting sent to a completely new instance of the model.

1:47:21If you sit down with the person and walk them through that fact and let them peek behind the curtain as well, if you could like show them what's actually getting sent every time, what these calls look like, it demystifies a lot of this. It starts thinking about it differently, right? It breaks you out of the frame of I'm interacting with something that is having a continuous conversation with me and which is recursively interacting with me. So it's like, no, no, it's like a completely new thing every single time. You need to see the magic trick for sure. Yeah, you see the magic trick. I think that helps a lot.

1:47:50I think to the point of the labs, dude, circle generative AI, which is we've had the worst goddamn marketing since this industry got started about two and a half years ago. It's been terrible. First of all, calling it generative AI is such an incredible miss when the thing you want it to do is do tasks for you instead of like, I don't know, produce images of big breasted cat girls. That's not the primary function of this technology, right? It's not the generation that's important. part of the marketing has always been this mysticism about what the models can actually do what secret technologies are hidden behind the curtains of the lab and then you have this s these eschatological pronouncements from from you know lab heads about oh this is going to make 50 of people unemployed in a couple of years and by the way it's going to be super intelligent and might kill us all like that's feeding that's feeding the mysticism that allows people to engage with this they don't really understand what it's capable of because they hide what it's capable of.

1:48:45Part of that is the engineering problem of understanding what the models actually are capable of, which we don't know, which needs a lot more research. But part of it is presenting things as being constantly hidden behind a curtain. It's like, oh no, we're creating magic back here. We're giving birth to literally a god. And who wouldn't want to worship a god? If you literally are making one, then of course people are going to be inclined to worship in it we need to maybe we need to start using more boring language here yeah i remember when do you remember it's not good for fundraising do you remember during the opening i dust up when sam got fired and then came back there was this whole meme about like what did ilia see yeah what secret project was he working on and i remember there's an article that was like it was called like project titan alpha craziness strawberry so it was just it was literally just RL.

1:49:36It was just RL. It was just RL and it went out and everyone did it and it's out there in 03 and it's out there and DeepSeek is open sourcing it and it's like, it's available. It's cool. It makes it better, but it's just RL on top of the research community, right? Like everybody kind of knew and again if you've been around there long enough, you know exactly what Qstar refers to. It wasn't like some mystical thing. Yeah, Qstar was like, oh, it's going to be this crazy thing. It's like, yeah, That's shipped and it's fine. But it's like, it's, you know, not, not to get too like symbolic with it, but you've got Q star and you've got Q and on and how much of that is just like mentally intersecting in people's minds because you're, you're like presenting it as if it's a conspiracy theory.

1:50:19Totally. Right. And at a time, again, at a time when our epistemics are falling apart and you're deliberately making people not trust you because you're not telling them everything. And you make a point of not telling them everything. Yeah. Right. It's, it's like people are going to project onto these things, whatever they want to believe. And I think that the marketing is part of the driver. I don't know how to walk it back. And as you said, it makes fundraising harder. But frankly, this is also a transformative technology. It is going to change everything. But is it going to be birth to a god that you should worship?

1:50:47Or should you take life advice from it rather than your friends? I don't know. Certainly not today. Maybe if we do have super intelligence, it'll give me better advice than my buddies, but not today. Yeah, yeah. How AGI-pilled are you right now? Dwarkesh recently updated his timelines to say, hey, I don't even think it's going to be able to do my taxes until 2028. Maybe something more like super intelligence in something like 2035. The timelines are shifting around. But do you think that we're making progress? Are we accelerating, decelerating? I'll put it this way. I think we know what the problems are, which means they're going to get attacked.

1:51:28There's a few points to this. So I've been drifting to becoming much more AGI-pilled or LLM architecture-pilled as time has gone on. First of all, I don't think you need AGI for this to be transformative. Of course. Meaningfully. We've invented this machine that if you feed it enough data and compute about any particular task, it's going to be able to perform that task. I'm a little bit at odds with Dwark Eshter on some of this because, okay, yeah, but if you did a pre-training corpus size of GPT-4 on just tax preparation, it's going to be able to do your taxes. um sholto was actually talking about that specifically like a tax yeah doing your taxes eval and then the eval gets saturated and it's like oh it's saturated but it's like okay now they can do taxes that's great but but here's the real problem so the problem with the evals is like yeah it's doing great on the coding benchmark but is it going to be able to do this coding problem that i just put in front of it nobody knows the answer to that question today you can't predict from the benchmarks what tasks it's actually going to be able to perform yeah yeah But if the training corpus is broad enough, like it might not be able to do the most complex novel tax processes, but it could do like the basic one that it's trained a billion times for.

1:52:31Right. Yeah, exactly. And again, it's just a question of getting more data into it. Yeah. But if we want the general purpose system, yeah, there's like a bunch of problems that we need to solve. But we know what the problems are. Like one of the biggest ones, which I keep harping on about here is the model currently does not know what it doesn't know, which is a big reason for it producing bullshit. Right. For example, when GPT tells you, I don't have any knowledge beyond a certain cutoff date, it's not because it has learned that fact by scanning its entire pre-training corpus. It's because in post-training, somebody has trained that fact into it.

1:53:04It's been fine-tuned to respond that way to things that require dates later than its cutoff date. We got to get rid of that if you want, AGI, because then the model can say, oh, I don't know this, but I'm going to go and get that fact. And I'm going to go retain that fact. And Dwarkech's point on this is, like, Dwarkech is, like, big on continuous learning. for this being necessary. It has to go out and gather facts. I think there's a bunch of ways of doing that. Being the founder of Chroma, I think memory is a great way to do it. It can just go out and store facts, and the thing is to have the general purpose processing system that works on top of that.

1:53:38I think timelines are less important than applications, less interesting to talk about than applications. It used to be you go to a San Francisco party where any AI people were present, and you would get asked the timelines question, right? It's not that interesting a question. The question that I've been asking people It's like, okay, great. Like, forget timelines. Imagine if I, in my pocket, have an API to an AGI that can perform any cognitive tasks at expert level human performance. What are you going to do with it? And almost nobody I talk to has an answer in under a minute. And the cop-out answer is always, well, I'm going to ask it what to do.

1:54:07And I'm like, well, it's expert level. So you're the expert on AI. So what do you think it should do? Because that'll be its answer too. And people get stuck. And people really get stuck. And I think that this is really illustrative, that the question now shouldn't necessarily just be about model capability. It should be about how do we get this thing to diffuse through the economy. And a big part of that is literally just asking like, okay, like a business is uncertain about whether it can apply AI to a particular task, right? They don't know what the cost is going to be. They don't know how well it's going to perform.

1:54:37Right now we're racing forward. We're like putting all this, all this money into these, into compute, into all that stuff. And by the way, that's great. America absolutely has to win the compute race. It's like without a doubt, not just for research but for industry but it's being underestimated like how much work there is to do to educate people how to use this effectively and how to estimate how to use it effectively and I think that's much more of a barrier than like having AGI yeah makes sense yeah come back on again soon last I'm curious we were talking earlier Sam was on Thea Vaughn last week and he casually dropped that you know, your conversations with chat GPT can be used, could be used against you in a court of law.

1:55:23Yeah. It started going viral for obvious reasons, even though I think a lot of people don't realize that like all email and most Internet. Yeah. I mean, that's my response here. But but but I'm but I'm curious if if like the path forward, if LLMs could provide therapy for the masses or one of these use cases that that should have some privacy, Do you think there's a technical solve there that's, that's, or, or is it going to be more of a, like some type of like legal regulatory solve? Well, we kind of have some of these legal solves for existing, like for web two stuff, right? Like we've got HIPAA, FERPA, we've got the COPPA stuff for, for protecting school kids.

1:56:06I used to have an ed tech startup, so I know about that. The problem is your GPT is general purpose. And so if someone wants to use it as a therapist, right, which nominally would be covered by something like a privacy thing. And by the way, I don't know how this works with GDPR. God help you if the Europeans get involved in some meaningful way in the contents of the chat. Yes, there is a legal solve, but there's a technical problem of knowing when the legal thing applies. right like at what point is it a therapist when i log into a therapy app and it's like covered by hipaa i know like i know what it's for but if i'm like if i'm just like kind of having a bad day and i start treating gpt like a therapist like am i covered by that or not it's a technical problem to detect that i should be covered i think that yeah yeah but but it's this yeah same thing you know if you're getting therapy from your business partner informally and you're saying things that are wrong or whatever talking about like like people i think people should have the they need to have responsibility over the context of the context of the conversation so i do think it's possible that like chat gpt if they want to attack therapy should create a you know uh hipaa compliant version of their product that's like a separate app and people should just know setting expectations is the right thing here ultimately but it's so hard to set expectations around a general purpose technology that's the thing especially for the average person everyone who comes on this show is suffering under the curse of expertise we know way more about this stuff than the average person and we don't realize how much more we know than the average person and I don't know if it's reasonable to expect the average person to like know that they shouldn't use this chatbot in therapy because then that might be disclosed in a court of law that's a lot of things like there's so many steps you have to understand there right the average person doesn't read like the error message that comes up on their screen when their computer's doing something weird you're not going to get them to like understand tech privacy law i don't have a good answer but i think you have to like set expectations somehow effectively yeah that's why uh you've you've seen like in these high profile like murder cases somebody will type in Google, Google, how to hide a body.

1:58:29It's like, oh, like I thought that like, that was, that would be cool. It's like, no, it's the same. It's a database. It's a website with a database backend that saves everything you type. There are logs, but yeah, people don't know that. It's a good point. No. Anyway. All right, gentlemen. All right, great. Thank you so much. Yeah, I think OpenAI should buy a billboard on AdQuick that says OpenAI is subject to the exact same rules as Google and every other consumer internet. Yes, the rest of the consumer internet. They should get on adquick.com because Adquick makes out-of-home advertising easy and measurable.

1:59:04And they can 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. Across the globe. And we have our next guest here in the studio. Super intelligence. He's suited up again. Let's go. Too good. Thank you. You keep that suit in the closet at the office when it's time for a big meeting or a big TVP. You look fantastic. How are you doing? You look so natural. You look great. I know it. Doing great. How are you guys? We're good. We have a jacket on the way to you. Thank you for...

1:59:37Dude, you ran a half marathon yesterday or the day before? How was that? That was fun. It was a little rough. I woke up with a little bit of a fever. And so I already signed up first. I thought might as well just go with it, you know? There we go. The Ashton Hall. I love it. That was just playing in your head when you woke up. Yeah, that's amazing. Well, congrats on that. And what other news do you have for us? Get the gong ready. Yeah, Julius just raised$10 million. There we go. Congratulations. Thank you. Total, total wrapper victory. Total victory. Total wrapper victory. You guys know it. And we're also launching a new product called Data Connectors today.

2:00:19So that means Julius can connect directly with your data source, like Postgres, Google Drive, OneDrive, and a lot more coming soon. Take me through how you're experiencing the data wars, the walls going up. We've seen this stuff with Glean maybe getting some sharp elbows at Salesforce or something that they don't want other companies coming in. They want to do it themselves. It feels like Postgres is naturally neutral. You operate at a different layer, so maybe less of a risk there. But how's that all playing out? And when it comes to specifically financial data, which is a category, I think you guys have a lot of traction and companies tend to actually own that.

2:00:59Yeah, exactly. But yeah, how is it all playing out? Absolutely. So, I mean, it's really cool to see like what's happening with Salesforce and Glean. Look, companies have gold mines worth of data. It's just like so much information in that data. And they aren't able to get the insights that they need because every time somebody on the team needs an insight, they have to go talk to a team, wait for hours or days. And so more than 90 % of the data, beyond just Salesforce or other business tools, even in your databases, like how are the users using your product? How are they signing up? Where are they dropping off in the funnel?

2:01:39All that data doesn't really get analyzed. And Julius is really here to solve that. Now, we're launching this data connector with Postgres because that's the main database that a lot of startups and late-stage companies use as they're scaling and have a lot of valuable data in there. Customer data, transaction data, marketing data, a lot of the data usually passes through Postgres and then ends up in different business tools. And so being able to integrate directly with Postgres means you can now talk to the main data store directly. Anyone on the team, a marketer, a product manager, even the founder can just ask questions and make visualizations within seconds.

2:02:21I was talking to a founder who identified an interesting trend in his company, and I want to see if it's relevant to your customers. He was saying that AI, tools, cursor, cloud code, windsurf, cognition, all this stuff is great at taking a 10x engineer to 100x. Maybe that's happening. It's definitely some sort of productivity boost. But he was saying that the bigger unlock for him and his organization is that designers have gone from 0x engineers to 1x engineers. and that's unlocked more value, maybe on some sort of relative basis because you're getting kind of a divide by zero error. It's like an infinite increase.

2:03:03And I'm wondering if in your customer's base, are you seeing more like non-technical people kind of become technical and go from that like 0x to 1x engineer? And is that kind of a key selling prop or are you more focused on there's a data scientist who's using Julius to go from 10x to 100x? Exactly. We're seeing the exact same trend with Julius. Cursor takes a 10x engineer and turns them into 100x engineer, but also they take a designer, and now the designer can ship code and ship features. And that all of a sudden is really powerful. Similarly with Julius, data scientists and data analysts can now be a lot more productive because they can just talk to the database, write their queries in simple English, and get what they need.

2:03:50but also everyone that they serve, marketers, product managers, operations people, finance team, CEOs, executive team, all these people now can get their own insights, and they don't have to be bottlenecked by, you know, timelines or bandwidth. They can just talk to the data and get the insights. In many cases, in fact, it's the data teams that are bringing Julius into the company. They're saying, hey, we want to do the deep data work that takes a long time, and we You don't want to be bothered by these ad hoc queries. So how about you just use Julius for your questions, and then if you need more help, just let us know.

2:04:23And then we are seeing like BP of marketing, BP of operations, in many cases CFOs use Julius, but also the PMs at the ground level using Julius to understand what do the conversion funnels look like? When the users use my feature, how are they retaining over time? What are the patterns in their usage? And they're now able to get those insights and then make better product decisions within minutes. In the future, do you think that every kind of org within a company will just automatically have Julius reports kind of like running in the background, producing, basically producing insights without people even having to prompt it themselves?

2:04:59Is that kind of where you're headed? Exactly. I mean, that's the next level of autonomy you want to get to. You know, you have the self-driving, right? Level one, level two, level five. So right now we're at level four, kind of like autopilot. soon you want to get to the Waymo level, where Julius can simply monitor all your data, let you know when there's a change in trends. In fact, what we're hearing from companies and people using Julius is that there's a dashboard fatigue. We have hundreds of dashboards to monitor, and it's really hard to know when things change. So what if you could have an AI agent like Julius?

2:05:29What if you could have a dashboard for your dashboards? No, I completely agree with this. Everyone asks for a dashboard, it gets built, they check it for the first week, and then you check the user analytics and no one's checking the dashboard. And it's ad hoc analysis, I think, is so much more powerful. Yeah, well, you want Julius's super intelligence to be like surfacing things that a human would never notice. Yeah, it has to be unique because you stagnate when you're just checking a number. Okay, it went up another 5%. It's like, go find what numbers are actually trending down. Make a serious change to your business so that you turn that KPI around and then keep doing that.

2:06:04Put the KPIs in orbit. Put the KPIs in orbit, as we like to say. I love that. We cut you off. No, I think you guys are spot on. I mean, you need a dashboard for a dashboard. You need to have a dashboard that will tell you the key metrics that change every day. And if the metrics don't change, you shouldn't have to look at them. Julius will be able to send you email reports. Imagine getting an email from your AI data analyst every day at 8 a.m. Hey, here are the five key metrics, and here's an executive summary about what's happening in your business. Imagine that in your Slack. And you can then add Julius in your Slack channel and just dive deep into data with all your teammates.

2:06:42And that all of a sudden is really powerful. You're speaking our language. This is music to my ears. Music to my ears. It's going to start tearing up. I'm going to start crying. It's so beautiful. It's an agentic workflow for your agentic workflows. It's lovely. It's lovely. I mean, we were talking earlier about data security, the T app hack. I'm interested to know how are you building Julius for security? I have a Postgres database. I imagine that Julius is often vended in as a web app, but then are you replicating data on Julius servers? Have you hired security people? Like, what's the modern startup approach to security?

2:07:19Absolutely. We take data security very seriously. In fact, from day one, we have invested and built our infrastructure around keeping our users' data secure. We recently hired a head of data security. He has 18 years of experience. you know he's he has more experience than most people um thank you um but what we do essentially is we give each user their own sandbox environment so each user gets their environment that's kind of partitioned from other users and their data when they import into julius is in that environment and all the code execution happens there and then you can with the push of a button just delete and wipe the whole environment and everything is deleted uh all the data all the files all the all the notebooks.

2:08:02And additionally, you can also go in and delete a chat history if you want to. We're SOC 2 Type 2 compliant and we invest super heavily in data security and a lot more, a lot more stuff coming soon on that. Amazing. Well, so happy to see your success. You know how hard you work and yep, looking forward to having at the rate you're going, you're just going to, we will just set up a recurring monthly call. You just join, announce the new round. Yeah, hop on anytime, man. Always good to catch up. It's awesome to see. Thank you, guys. We'll talk to you soon. Bye. Really quickly, let me tell you about Bezel.

2:08:37Your Bezel concierge is available now to source you any watch on the planet. Seriously, any watch. Go to getbezel.com. Did you see the news about Intel? Very, very rough. Semi-analysis says value bros are in shambles. Intel now trades for book value-ish. But now the question is if they can achieve book value. For the price of book, now you are getting musty tools, empty shells, and the third turnaround in five years until inside of the doghouse. Brutal. John, by the way, we have a guest joining. In person? No, no, no. Okay. Just in a little bit here. Yeah, yeah, yeah, yeah. In two minutes, right?

2:09:16Cool. Yes, we did inventory write-downs. We had tools in the line that were older tools, and we took the opportunity since we had an excess amount. We took the newer tools put them on the line took the older tools out and so the problem is is that they're holding this equipment on their books and No one is after them to buy it because they are on the lagging edge now. So very rough There's a whole bunch of interesting analyses going on with yeah Ton the number of people that would want to buy hundreds of millions billions billions of dollars of old Semiconductor manufacturing gear and it's pretty small small market pretty pretty slow maybe some rare earths in there I did want to give a shout out to Dak in the chat.

2:09:57He says that he was worried that we never read chat. Well, we read it and here you go. Yeah, we have a new screen here. It's not high enough to see the most recent one. But we are monitoring it. So if you put a message in the chat on YouTube or X, I believe we will see it all here and be able to answer your questions if you have them or react to your comments. Nico is in the restream waiting room or we're still waiting for him? I think we're still waiting. Nobody in the restricting waiting room. Nico created super intelligence for Excel. Oh, okay. I'm saying that. Love to see that. But I'm going to send him a message right now.

2:10:38Well, in the meantime, there are so many good posts we can run through. OpenAI is hiring for consumer hardware, and it gives us a little bit of information on what they're thinking of building. Potentially, they want experience with wireless OLEDs, So that means a screen, microphones, cameras, portable cosmetic enclosures. Another screen. Liquid plus dust. Let's give it up for more screens. We know this is what you've been waiting for. Another screen in your life. You maybe had one on the wrist. You had one in your hand. You had one on your desk. Have you ever played Fallout? You ever feel there's Fallout at all?

2:11:13They have the Pip-Boy. It's basically like a big screen that goes on your, it's like a gauntlet that goes on your wrist, your entire arm. I actually want one of those. It's very cool. It's very cyberpunk. Ideally, like some type of tattoo. You could get a screen tattooed and it would act as like a screen. That'd be cool. Well, they'll have to patent it just like Jim O'Shaughnessy did. Did you know that Patrick's dad created stock trading? Stock trading online. He invented stock trading online, apparently. Genuinely crazy. He says, I can't shake the feeling that I should have done more with this.

2:11:46This is Jim O'Shaughnessy, the father of - Creator of Robinhood, creator of public. Yeah, apparently. creator of all stock trading, E-Trade. Patrick O'Shaughnessy, the host of Invest Like the Best. Yeah, that's true. Honestly, Jim, you did the most important thing. You invented business podcasting. You created an incredible investor, incredible podcaster, and an incredible friend. Yes. And so, yeah, Patrick O'Shaughnessy, or Jim O'Shaughnessy, has a patent for system and method for selecting and purchasing stocks via a global computer network. In 1999, this was granted. Wow. Yeah. I don't know how he didn't capitalize on that.

2:12:24I mean, I feel like E-Trade must be coming up. I still don't understand patents and how like you can have the most important patent and not get paid. You can have a minor patent and like troll everyone and get like billions of dollars. Yeah. Why is it difficult to enforce them in the software context? It is very odd versus bio or anything else. I don't know. Anyway. Yeah. The vaccine has a good post. Whoever prayed on my downfall, pray harder. It's a picture of an alligator dunking a basketball as one does. This doesn't even look AI. Nico says he's here. AI generated. He might have the wrong link.

2:13:02He might be in a different waiting room. Make sure he has the correct restream waiting room to hop in. Anyway, absolutely brutal. The Financial Times, this is from Geiger Capital, says, There's no hiding the fact that the EU was rolled over by the Trump juggernaut, said one ambassador. Trump worked out exactly where our pain threshold is. It's crazy. It was like Liberation Day was so much doom and gloom. I remember having a viral post that day that the market traded down like 5%. And I said, you know, the market's down 5%. Interest rates are up. Everything's bad. But you know what hasn't changed.

2:13:42the feeling of repping 225 for reps. And everyone enjoyed that. It's during some absolute doom and turmoil. What else is in the news? I'm going to do a cinematic launch video for my startup so it stands out. And it's the Buzz Lightyear meme with tons of identical Buzz Lightyears. The cinematic launch video, a little bit done to death. Still breaks through if it's good, if it's unique. Yeah, I mean, Waves did a kind of video that felt pretty Clue-esque. Yep. got 12 million. I think he said 30 million views, something like that. We've got a ton of views. We've got our attention. But I think it's an opportunity.

2:14:19I like this post here from Bi Han. He says, time to go back to some low res for a while. You know what video that is? I do not. I believe that is the first video ever posted to YouTube. Oh, YouTube at the zoo or something. Yes, that is the co-founder of YouTube talking about the thing. But the thing is you could get, you could shoot a launch video. There's so many different ways you can take 60 seconds and make something new and different. Yes. Yes. It, it's not, um, it is not so hard to generate ideas. Yes. At the same time, tech exists on X links are banned. So if you write a beautiful blog post outlining your, your website or plan for your business, anything that you do, uh, that's going to be a lot harder to get it go viral than a video that is native to the platform.

2:15:04Uh, I think, I think a thread could also do the same thing. There have been a number of companies that have launched with threads. In some ways, Lulu launched Rostra kind of with a thread. She didn't do a cinematic video or vibreel. She just got a direct manifesto. If you've got enough juice, you can just do text. You can post. Yeah, you can just do text. Post your way to Valhalla. But yeah, it is becoming saturated. It just demands more creativity, more unique styling. You can't just copy and knock off someone else and expect to break through in a huge way. But But it is a great way just to quickly, in one minute, share a message.

2:15:37Condenses it all down. Anyway, Nikola Djokic in literal tears after his horse won a race today. He barely cracked a smile after he won an NBA championship, LMAO. Bro, really don't care about basketball. I don't understand what race this is. It's some sort of chariot racing with a man dragged behind a horse. I saw this in a bar once. Remember we were watching this? We were out in Ojai and we saw this on the TV. We should get into this. This feels extremely us. I would love to have our own team. I feel like I'm probably not suited to ride the horse at breakneck speed like a true jockey. I think they're a little bit smaller than me.

2:16:21You could get a height reduction surgery. People are getting these height extension surgeries. Yes. If you're looking to get into, if you're looking to go travel the world and hang out with a bunch of horses, get on Wander. Find your happy place. Find your happy place. Find your happy place. Book a Wander with inspiring views. Hotel-grade amenities, dreamy beds, top-tier cleaning, and 21st 7 concierge service. It's a vacation home, but better, folks. And Ryan Peterson is posting a screenshot from Black Hole who says, asteroid Psyche 16 has been found to contain gold reserves worth 700 quintillion dollars.

2:16:59That's enough to make everyone on earth billionaires. And I think that's, that's actually like the crazy math. Is that, is that right? Quintillion? Oh, I guess because it's probably quadrillion. Wow. So it really would make everyone a billionaire. That's a lot of gold. It's insane that NASA has not mined this yet. I mean, the, do they not care about humanity? it's like what are they doing everyone gets everyone gets a billion dollars worth of gold and then we got to go back to well i want a car i want something else i don't want usd is pretty cool yes yes yes i'll give you some gold for for uh some greenbacks brian peterson says he just wants his diet cokes to come in solid gold cans i like that could be possible uh brian kaplan sharing a photo of some absolute dogs, absolute legends at a round table.

2:17:4919 years ago, Tyler Cowen, Alex Tabarrok, Robin Hanson, and me, plus Ilya Rainier, hanging out with the boys in June 15th, 2006. You'll love to see it. Just guys being dudes. Oh, this is some news. Armada has launched. Leviathan launches today with our blueprint for American AI dominance. It expands the Galleon lineup pushing megawatt scale compute to every underutilized energy source we can reach trey stevens is breaking it down armada is building the infrastructure to make the world make sure the world runs on the us ai stack leviathan drops a megawatt of liquid cooled compute where the land and energy live at the edge proud to back the team as they continue to accelerate on their mission to bridge the digital divide i think this is unlocked by a bunch of different trends Starlink obviously very important a whole bunch of different pieces coming together to allow a shipping container full of compute to be delivered to the edge in all sorts of different you know all different environments we got gas we got the pitch it's fast real company fast and we will have the founders on this week I believe and we'll have to buy one of these for the studio I would like to inferencing a daily driver in the tobacco I mean we search a lot about like true crime stuff.

2:19:05I don't want to be flagged in some database. I want to do that inference locally so that I don't get subpoenaed or anything like that. Anyway, we have our next guest here in the studio. Welcome to the stream. We made it. We have super intelligence in Excel, but we're still figuring it. We're still figuring out Zoom. Wi-Fi is coming in a long time. What's up, boys? How you doing? What's going on? What's going on? I couldn't bring the suit. I thought the white coat was the next best. No, no. White's great. Mark us up. We're wearing white. It's very It's a good option. Okay, so... Break it down for us.

2:19:36What are you building? Fin Twit was freaking out this morning because they all thought you paper clipped them all. Okay. So yeah, talk about the launch. So Nico was on the show a while back talking about kind of the pre-release and the app is live today. Sounds like you've improved it quite a bit as well. So yeah, break it down. Yeah, sweet. Thanks for having me, guys. Yeah, Fin Twit's having a meltdown. they're having the kind of moment that we had August 2024 in software engineering Carpathia tweeted cursor's kind of there now yeah and what happened was there was a lot of hesitation from senior devs and I think a lot of the junior devs became really really good really really fast and there's a word for them now like context engineers and I think where we are now is there's this moment happening for Excel so our product's called Shortcut and it's a superhuman Excel agent and it does feel like that August moment for some things it's unbelievably better than humans and for others it's just like stupider than your intern Yeah.

2:20:32Yeah. Isn't it isn't the dynamic of like the finance world and the Excel user radically different than the than the software engineer? I feel like if you're if you're an amazing software engineer, you can work your way up at a big tech company, become like a senior engineering fellow and basically always be working on algorithms and code and software your entire career and become fantastically wealthy, become an AI engineer that makes 100 million dollars. But in finance, you go to an investment bank or a hedge fund or a private equity shop. You're working in Excel. And then you level up and it becomes all about relationships.

2:21:05And so you don't actually... When I think of a managing director at KKR or something, I don't think, oh, wow, that guy's really good at Excel. I think that guy knows people. Well, he probably used to be. He probably was. Exactly. Used to be. So yeah. How does this dynamic actually play out? Yeah. So that's a good call out. It's actually the same in tech, but there's like two paths. Totally. You could go tech lead path, and tech leads are doing some wizardry on whatever algorithm or back-end infrastructure. You could go director path, and it just becomes like you talk about or you brag about how much headcount you have under you.

2:21:36That is the only thing that exists in finance. And when we show people in finance this, their first reaction is like, it's kind of fucked. Because Excel is the reason that I'm where I'm at now, and that path doesn't seem like it's going to be a thing. So yeah, it is different. I mean, it feels like you'd like even I feel like a lot of people, entrepreneurs don't want to be vibe coding themselves. They want to hire somebody who's really good at vibe coding and they want to get the value of a senior engineer for the price of a junior engineer. I imagine if I'm an MD in an investment bank and I'm like, great, I can even more quickly turn over a term of the DCF to my client.

2:22:13And yes, I don't mind keeping the Harvard grad, you know, up all night on the weekend. but if he or she could get it back to me in an hour instead of 10 hours, I'd be happy with that. Yeah, and you'd still rather that person work 30 hours a day and just send you more? Yeah, totally. But yeah, we know this because of tech. This happened, and it was actually much scarier two years ago. Whenever you can drive the cost of inputs lower and your ROI goes up, the obvious thing to do is just put more gas on that fire. Totally. It makes sense that it's scary. As in, we used to have tools, and now we have things that can use our tools.

2:22:47Yeah. Right. So I think it's inevitable that they're going to learn what we learned. And that's how I'm going to position it. So we need to go to the finance community and teach them about Jevons Paradox. I'm glad somebody is bringing this up. We're going to host a Jevons Paradox happy hour. We should. Top bar. We're going to be we're going to be in New York. I was about to say. We'll start. We'll just go on the street if we see someone dressed in a suit. Should we do a little circuit? Yeah. You should. Are you are you in New York? No, we're in Menlo Park. I'm the only person in Menlo Park who knows finance, unfortunately.

2:23:15Hey, what about the venture firms? They need to know, multiply revenue by 100. If you're serious about finance, you go to private equity or VC. Or private equity or hedge funds. Yeah, sure, sure. Yeah, I was like, wait, what? No, no, no. You put it in the Excel sheet, multiply the revenue by 100, that's the valuation. You do a little bit of this and you close your eyes. That's the valuation. Slap a multiple. Multiply it by 100. What do you mean I need a model to determine the valuation? I'm just going to pay slightly more than my competitors. Yeah, what is the shape of the user base? Where is Excel actually powerfully used?

2:23:48Because I feel like so many of the hedge funds, they used to do stuff in Excel. Then it went to VBA. Then it went to high frequency trading and proprietary systems. And I'm sure there's some stuff, but it's usually in private markets, less liquid markets. Where is Excel holding on really strongly? You know, I'd say hedge funds are the toughest customers. Anything that they're really heavily relying on, like already low-level systems or real-time data to public access markets. That is a tough customer. Thankfully, they're like a small sliver, and I'm not sure I'm building for them. But anything like private equity where you're getting a SIM and you need to build a model on it or a mini model, anything where you have a template, you want to update it in light of new information.

2:24:25Corporate real estate is a bizarre area where there's a lot of PMF. But it seems to be that building new models could be 100 times faster now. Editing existing models is a little tricky. I mean, if you really want to swap a model for new data, that's kind of the highest bar that you can possibly have. but seems to be solvable. Sure. So what's the workflow now? I'm in PE. I have a SIM. And for those that don't know, it's basically a bunch of PDFs and data showing what's actually happening in a business that might be for sale. Can I drop? How soon can I just drop this? Yeah, yeah, yeah. Let me walk you through it.

2:25:00And that's exactly right. So it's like a confidential investment memo from the sales side. A big bank will send it to a PE firm or they'll send it to a bunch of them. Typically what you do is you try to analyze, you diligence the deal. and you look at this big PDF full of data, and you try to model it to understand, is this a good deal? Should I bring this to my boss or whatever? You're usually constrained by how fast you can do that analysis. Currently now, you can just attach a PDF of a SIM. They're usually PDFs or multiple PDFs or customer data dumps as well. And you can just say like, hey, here's my template.

2:25:29I use this for my LBO model. Make me 10 of them. And in fact, now you can also just email it and be like, I want 100. And you can get all of that. Or you can attach 100 attachments and ask for 100 models back. And then it becomes more like, how fast can I review data? More like, how quickly can I do groundwork? Yeah. What about just FP &A orgs in small and medium businesses? I feel like when I run companies, I have one finance person and we have a question, should we expand in this market? Or should we add a new SKU? Or should we do something? Should we take out this line of credit or whatever?

2:26:02And usually there's an Excel model or a Google Sheets model that's built, but there's not there's not even like a catalyst for pull a model off the shelf, it's like, no, I need to just be in Excel and model this out, or even like CAC to LTV, is this, how are my KPIs trending? There's so much work that goes in the long tail of Excel. Is that just like not the early market for you, or do you see people abusing the tool quickly? The crazy thing is, is I do these live demos and people are like, oh, that's insane, but is this just DCFs? Yeah, yeah, yeah. And I'm like, you know, we are in the tech world, so we think tech is the center of everything, but anything you can possibly do in Excel is way more in distribution than like Ruby on Rails.

2:26:44There's no kind of model you can conceive that is unfair to ask. It becomes like, what are the fundamental limits of AI? And it turns out writing Excel formulas is a lot easier than writing amazing code, actually. So it's more or less like, how did this not exist at this point? But yeah, I have yet to find one example. You said shortcut beats first-year analysts from McKinsey and Goldman head-to-head 89. percent when blindly judged by their managers we even gave humans 10x more time what what did that kind of like benchmarking exercise uh yeah so we gave that look like we gave we gave multiple first-year incoming analysts from mckenzie goldman bcg um jp morgan a couple other firms um like five tasks and we thought like they were the most in distribution for classic finance work so this was consulting like building out models mna models lbo at dcf and personal hobby stuff and actually even product management, like build a dashboard over this customer dump.

2:27:38We then gave them over 90 minutes per task. Some of the hobby ones had like 15 minutes. And then we just asked them to submit it. And then when we did that, we also got in touch with managers from these firms and just sent them a Google form with like side by sides, completely anonymized and just pick your preference according to like accuracy, professionalism, whatever. I thought it was going to be closer to 50-50. I think what I really learned is first year analysts are like kind of lousy, like they're not that good at building some models. And then also I think people just take way longer to do things.

2:28:07I think if I had to critique the study, it would be that like five hours is not enough. And like they would have liked to have a week, one person wrote me. But I didn't want to run that study. Yeah. What's your read on the meter data that these coding agent tools did not actually speed up software engineers? Did you see this? So meter did a blind study. It was a small number, But it was interesting because it was very solid software engineers working on advanced bugs in like big open source projects, not vibe coding a landing page. Like truly, like there is a sticky bug hanging out in some fundamental repo out there.

2:28:47Go fix it. They estimated that they would be 20 % faster, 30 % faster, something like that. Turns out they were 20 % slower. Or I'm wondering if you'll see a similar pattern where there's a speed up for bootstrapping and getting from zero to one on a project. But then for the really crazy person that already has all their macros and doesn't touch the mouse, maybe they're actually going to experience a slowdown. Do you think that would happen? What do you think about that? Yeah, yeah. It's a great question. I saw the study. I didn't read it deeply. What I will say... Not that it would be bad for you.

2:29:18It's just an interesting finding. Yeah, no. I think the obvious thing is that it's like using AI as a skill also. Yeah. So the top startups have the best context engineers in the world. They know how to optimize the KV cache hit rate, whatever it is. If you kind of give, even a really strong engineer who's been coding for 20 years, you're like, now you have to use this tool or use it however you want. I have no doubt that even the best engineers and the oldest engineers eventually would find patterns in which it's massively unlocking for them. So I think that is what's going to happen. But actually, to be the most critical, I think clearly Excel modelers are way less technical and way less on the frontier of tech than our software engineers.

2:29:52So it's actually more on me as a product person to like, how do I unearth some of these capabilities and hide the ones that they're not ready for? And even train people for this because it is 2023 all over again. Yeah. How templatable is the work? Because I know Canalyst sold to Tegas, which sold to Alpha Sites, I believe. And Canalyst was kind of like off the shelf financial models, some data integrations. Capital IQ has had integrations. Yeah, yeah. If you have a Bloomberg terminal, you can have integrations. And so there's almost a world where you want to like sit on top of a library of templates.

2:30:24In some ways, the way like deep research is clearly RLHDef on like a couple templates of like what a report looks like. It loves tables. And so I don't know if you want to do like fine tuning or like have a model that selects a pre-filled template. And then you're using your tool just to update that template. What do you think about that? Yeah. So let's break it down into two things. One would be product, and then the other one would be training. As far as product is concerned, I think the most important thing you can do is let users actually upload their templates. Your income statement is going to be different than Orpinkus or whatever.

2:31:00And then just let them work naturally from theirs. But then from the training perspective, you have to collect as much of this data as possible and train your models. Definitionally, if you're using Frontier models, you can be limited in what you can and cannot serve in terms of trained models. But it's actually a much more constrained environment. So I think we stand to benefit from it much more than does like operator or deep research. Yeah, yeah. Makes sense. Jordy, anything else? How do you think the product needs to improve going forward? It sounds like in the study, in the tests that you ran with the incoming analysts, it performed very well.

2:31:36That's kind of like a unique situation. what's it going to take to get it to the point where it's actually disrupting the job market in some of those roles? Yeah, it's a great question. Clearly right now, I think what it does is it makes, like maybe you guys are rusty on Excel than you used to be in, like it makes you much better at Excel. And if you're kind of a rookie or like you're a solopreneur and you want to like have a one-man finance shop, like you're much better now already with it. So for that, like we passed the capabilities threshold. The big question becomes like for like real enterprises who people hardcore use it, Are we there?

2:32:09And I would say, like, if August is that moment, we're like in June. Like, it is a pretty clear capability threshold we have to pass, and we're actually actively adding all of the limitations to our internal benchmark and just hill-climbing them as aggressively as we can. Part of it becomes a little bit of a product science of, like, how you can abstract the limitations away from them. But, like, more specifically, it's exactly editing and overwriting large, nasty existing templates and files with new data. It's like, if you don't do that at a certain percentage of accuracy, it's just a non-starter.

2:32:40Yeah, I feel like part of the challenge will be figuring out, like giving the user tools to figure out where shortcut is hallucinating. Right, right, right. If you have like some, like part of that, I remember, you know, working with, you know, get a model made in the past. And I just look at it and I'm like, okay, everything looks good, but like it's definitely off. and like it's just like it's too good or it's too bad model smell yeah yeah yeah exactly um and then you dive into it and you find like one or one or two kind of yeah reasons reasons for that um and i think like the people that might critique shortcut today are going to say like oh i used it for this or that and like it missed this column or whatever yeah yeah you know um for sure that kind of stuff so i mean in terms of how we think about it i'll tell you how we think about it and then what actively we're doing about it.

2:33:31But this is not a new phenomenon. Encoding, it had to get certain good. And it had to be certainly at a certain level of observability. And when cursor brought the apply diff function, it actually made the job of supervising AI doable. I used to copy blocks and I wouldn't know what would break. That was the big thing. It was actually observability, not accuracy. And then Sonic got good enough. In self-driving, it's safer than we are. But we still don't want to make that trade because we want to be like, that guy killed that person. We need to be able to blame people. Radiology, it's there too, right?

2:33:59Like we just don't know who to sue if something went wrong. I think for finance, like exactly what you're talking about is we had to have a UI that allows for you to observe the diff. And like we had a good inspiration, right? Cursor. So actually now it's part of the launch video I just shared. And I know we were talking about like product demos, which I have a strong opinion about. Like the whole, like you have to be able to see everything that's changed, but also the root material. Because in Excel, definitionally, some things are hard coded. but you want to know where those came from because you can't justify it right like where what part of the 10k did it come from what web search did it come from what's the exact quote like what page so if i really do believe that it's not about getting to 99.9 accuracy it's about like getting to perfect traceability because like you know your analysts suck like even your associates are not good right so you just have to be able to to observe it at a superhuman level do you think well yeah the good thing for analysts is like you combined with a product like this could actually get to the point where you're truly elite.

2:34:54Yeah, yeah. So I know people are afraid. Let me hit that. Clearly, analysts should be using this or a tool that will try to do what we're doing. No doubt about it. The right launch is a non-zero amount of controversy. I know that's an emotional reaction that people have, but clearly, people will be using this the same way. If you don't use Cursor now, you don't know what you're doing. Yeah. You're not afraid to trigger a couple of incoming analysts that are going to say, I used it and it did this wrong. It's like, well, you probably would have made a mistake too. Yeah. Do you think you'll face more competition or there will be more war rooms planning to compete with you from the hyperscalers who have products that they want to add this feature to?

2:35:38Or the foundation model labs who see, hey, you know what? I can maybe I don't have control over a consumer application, but I can do a lot of these calculations in pandas in Python. Yeah, I'll answer those. Yeah, those two are very different. Yeah. When it comes to the hyperscaler, it comes down to can they? I think they would if they could. Right. And they would have already. I think it comes down to like inertia, talent even. Sure. Then that's not true for OpenAI. Right. But what's the problem there? I think it's going to be a bitter competition, is the truth. Then you have to ask, what principles do you really believe in?

2:36:18And will it prevent you from doing what I'm doing? I think even if you made an effort to copy me or beat me or directly compete with me, you'd have to believe the things I believe in wholeheartedly. And I think most of the people at Frontier Labs, I know because I work at a small one, really don't even care about Excel. Or they think that we're all going towards an input-output model anyway. that you have to do a general training paradigm to bring you just a model at the very end. And who cares if it's hard-coded? Like, it sells a thing of a pass anyway. I think there's like a bajillion dollars to make in between those two states.

2:36:47And I'm not even sure that their path brings you to that state first. So my argument, or the way I think about competition, is more like in principle versus the top labs. But it's more just maybe arrogance when it comes to hyperscalers. no no I think that makes a lot of sense I mean open AI is fighting a war on like seven different fronts right now they're like oh we're also going to do a phone we're also going to do a browser and we're also going to do this and that and you know like a coding environment and all this and that comes from the research how do you your seven steps down the power law will they really build a competitor a piece of spreadsheet software like they probably could but that's pretty low on the priority stack for them.

2:37:33I imagine. I guess your approach from a go-to-market standpoint just seems to be like, create the best product possible and release it and see what happens. Is that, will you have an enterprise motion over time? Are you going to set up a New York office and pound the pavement? No, we're going for drinks, guys. Now listen, I think people have a lot of false stupid pride about their go-to-market motion. I've mentioned Cursor four times, a lot of admiration. But they beat the drum that we've never spent the dollar on go-to-market or distribution or ads. That's a flex, but it's stupid. You can also have a world-class sales team.

2:38:12When I meet with the CIOs of the biggest banks in the world, they want to go. Now I need to know what it takes to have the right SWAT team to put that together. But I don't think that comes at the expense of what we're doing from a prosumer bottoms-up path. Yeah, totally. Inevitably, one will seem to be more fruitful than the other, and I will pour proportional resources towards it. I think clearly my gift is more bottoms up. But both can be positive some for you. For sure. And I don't think Winsurf had to... Yeah, exactly. Exactly. It just doesn't. Deal director in the chat says, Nico said the finance bro end of times is delayed a couple months, which I think is correct.

2:38:50I wanted to give them a chance to get back from Central Pay. Yeah, you have two months to escape the permanent underclass. Yes, yes, yes. So start vibe modeling. Anyway, this has been fantastic. Always good to hang out with you. Jordy, anything else? Yeah, congrats on the lunch. Yeah, congrats on the lunch. We'll talk to you soon. Appreciate it, guys. See you next time. You're on notice. You're on notice. Listen, I'm a fixture now, but I'll actually suit up. Okay. The microphone's fantastic. Cheers. Everything else is fantastic. Take my job. Bye. Gary Tan says, hire the engineer who worked on the best product.

2:39:21hire the marketer and salesperson who managed to get the worst product to outsell the best product? I'm going to agree. So, I mean, an example here is, is this is the product quality is up for debate. I'd say like the market generally, but they got to 80 million. They had a fantastic go to market motion. And now that is a cognition. And now they have a cracks engineering team and Pretty wild. Yeah. I mean, it is hard to find, like searching for engineers who worked at the best products. It's very easy. It's what Mark Zuckerberg is doing. Go hire the OpenAI researchers who built O3. Go hire the OpenAI researchers who worked on images and chat GPT.

2:40:05They're clearly good. It's much harder to actually identify products that have outsized distribution. It's funny that OpenAI has so many different products and models and things like that, that there's so many people that can leave and be like, I was the product lead on O3 Pro Mini High. And you can raise$100 million off of that. You can get$100 million in your bank account in a hiring bonus. Noah Smith says, I was doing a podcast with Eric Tornberg and Dwarkesh the other day, and we were talking about whether AI will destroy humanity. And in an offhand way, I remarked that technology had already destroyed humanity.

2:40:40And it's showing the world population with a projection based on fertility rate of 1.6 children per women, basically showing that by 2 ,800, our population will go down to zero. I would just say technology has not destroyed your humanity. I have 775 years to figure out this problem. Skill issue. Skill issue. I think we're going to be good. But yeah, it is an interesting trend, obviously. It's one of the less explored territories in what happened. I think if we can get a vaccine for twins and triplets, every, and we just keep the pregnancy rate where it is. It's the rate of twins and triplets is very low in vegetarians.

2:41:27And so eating a carnivore diet might literally save us. I'm not kidding about this. Government run hamburger shop. Yes, yes, yes. Every universal basic hamburger three times a day. It is one of the unexplored answers. And I think because tech people debate that what happened in 1971 so much and the stagnation thesis, no one really wants to interrogate, was it technology that happened in 1971? Was that the beginning of the technology boom? But maybe we should be having that conversation about the effect that the personal computer and then the internet had on everything. That might be one of the reasons.

2:42:04One of the interesting theories for the original what happened in 1971, the Peter Thiel stagnation thesis, this idea that we are not building flying cars, we stop building nuclear reactors, is truly just that the internet was such a great draw that it brain drained everyone away from everything because you, over in nuclear territory, you had regulation and you couldn't move that fast. But on the internet, you could throw up a website and get distribution and had very few forms to fill out, but very few gatekeepers to delay your rise. And so you could just grow at the natural rate that the market pulled.

2:42:42You basically had a free market on the internet and in the computing world, and you did not have a free market in the energy world or in the flying car world. And so it is interesting. Well, more importantly, armor-plated Gulf Force One. Gulf Force One spotted with President Trump at the Gulf Force, I guess. Did he get this transported to Scotland? I suppose so. Because it looks like a Lynx course. I mean, they fly on a global, like, have you seen the plane that the president flies on when they bring the beast and the SUVs and stuff? Oh, they bring, is it the C-130? Yeah, it's one of those. It's a huge tube transport.

2:43:21Okay, that's easy. I think the entire front of the, or maybe the back of the plane goes down. They just, they're already in the cars and they just drive out in the armor plated SUVs. And so bringing this does not seem that difficult. It is kind of funny because, I mean, I guess it keeps you safe while you're in it, but then you have to get out to golf. So it's not like a full security detail, but seems like it works to some degree. Anyway, Nizzy says, the only way to make AI output good code, I swear, is copy the Figma file exactly. No mistakes. My sister will die if you F up. Ridiculous. And so, yeah, this is, you know, I think this is more of a joke at this point.

2:44:02And like we are out of the like make a threat. If it works, it works. I don't think that you actually get better LLM results by prompt engineering in that way. I think that's a relic of a few years ago where you used to have to say, you know, please do this and don't make mistakes and all of that stuff. Like you used to when you went to prompt an image generator you used to have to say no six fingers Don't do six fingers and then it would do five fingers you had to tell it that and now it's all baked in And so you don't run into that problem at all anymore Anyway, in other news Ashley Vance has a new video on on core memory about new limit who's been on the show Which is finding combinations of protein that reverse aging across the body is perhaps the most exciting work in the biotech field I agree It's backed by Brian Armstrong, Patrick Collison, Josh Kushner, Nat Friedman, and others.

2:44:51You should go check it out. It's a 17-minute video. It's out on X now. And of course, it's on the Core Memory YouTube. Yeah, shout out to Ashley. Oh, I mean, Ashley has got to do a video about Robert Maddox. This is begging for a Core Memory deep dive. This feels equally important to humanity's future. Yes, yes, yes. So Robert Maddox has a hobby. Looks like an older gentleman. his hobby is installing jet engines on anything that can move and he put a jet engine on a bicycle here and is just ripping down the street absolute legend let's put a jet engine on a horse let's put a jet engine on the basketball player I don't even know his name the guy who has the horse oh yeah yeah yeah I don't know someone else I don't even Nikola Jokic Jokic I'm sure I mispronouncing that and revealing my lack of knowledge.

2:45:48Quinn Nelson says. Look at this. Look at this. Look at this. Here you go. Look at this guy. He's having the time of his life. On a skateboard? Riding a jet engine. I didn't know you could. I wonder what it takes to build a jet engine like that. Yeah, the metal just heats up like that. It's crazy, right? That's insane. Wow. Yeah, this is hard tech. Get this again to the gundo. Wow, Santa. This is remarkable. What a rig. Why is this not commercialized? Why is this not a thing? Is it just extremely dangerous? Did they never figure out how to make jet engines reliable in the right way? It feels like maybe it's not fuel efficient.

2:46:17I don't know. We use them in planes. Why not in skateboards? This seems extremely dangerous. I wonder how fast he's going. The Santa Claus one is particularly good. He seemingly does these in flat, open spaces. This also feels like textbook, like this could be AI generated. But I don't think it actually is. Wait, is this real? I thought it was AI when I saw it. I don't know anymore. It's farm to table. It screams Robert Maddox. Let's look it up. Let's fact check this. He has an account called Crazy Rocket Man. I think he's Lindy. I think he was around before Genitive AI. I can see videos. He is real.

2:46:49He was doing this 10 years ago. 10 years ago. You think he was using Dolly 1.0 for that? No way. No way. Or he's been an entirely figment of Sam Altman's imagination. Potentially. Okay, we're getting back into 20 years ago. Oh, you're 20 years ago. 17 years ago. MaddoxJets.com. This is the value. Incredibly scary, pulse-set-powered jet engine car. There we go. Well, speaking of AI, Elon Musk has rolled out a change to Super Grok coming to your phone soon. And this is actually a great feature. So it has an auto setting that chooses the best mode, whether to go with the fast response or the deep or the expert response.

2:47:32This is something I it's turning into one of my number one frustrations with the chat GPT app at the moment is that Sometimes I want oh four and I want or four oh and I want something quick And then sometimes I want oh three and if I trigger the quick question with oh three pro and it's waiting there for ten minutes I have to go and start a new chat and then change the model and then get the oh the the four oh question Answered and it just slows me down a little bit And I think this auto mode should come to the chat box. I would even just be down for give me the ability to pick the model in the chat box.

2:48:06So I should just be able to say, hey, use 4.0 to tell me the capital of Scotland. And it should just do it quickly. Or I should say, I need a deep research report on this. And it should just trigger deep research. I shouldn't have to click UI buttons on this. Text is the universal interface. You know who said that OpenAI. So make it truly a universal interface by allowing me to pick the model in the prompt window, please Anyway, speaking of chat GPT Quinn Nelson has a post that says today I learned my wife is using chat GPT to find coupon codes for web stores online She said it works almost every time without ad-ridden websites filled with old codes I've literally never once considered using an LLM for this, but now it seems obvious.

2:48:51I love it very interesting. I never would have thought to do this, but now that it's out there, makes a ton of sense. Go when you're searching, when you're about to check out. Flip over to ChatGPT. Lucy code Ben20 gives you 20 % off store wide. There we go. There we go. This is breaking news and this is important. Mercedes-Benz is finally integrating Microsoft Teams into its vehicles. We have been begging for this. They really are you know it's basically an office on wheels well this is about time this is huge because if you're a defense tech founder you're probably ITAR compliant you need Microsoft teams you're probably driving a Mercedes-Benz 6x6 yeah right you should be G63 6x6 not very American probably armor-plated but that's extremely here yeah it's pretty pretty American dynamism coded you're making getting an American made six by six just aren't that many other options if you're if you're in the market for a six by six.

2:49:50I want to see teams in a Ford Raptor. That's the ultimate gun. The gauntlet has been thrown down by Mercedes-Benz. Balls in your court, Ford. Ford Motor Company. Let's get teams in there. Everybody's dunking on this being like, oh, this is so silly. No one asked for this. It's like so many people take conference calls while driving. Why would they not make an integration? Also, Mercedes-Benz has one of the few active systems on the road that's level three self-driving. And what that means is that when you're on the freeway at certain speeds, I believe between like 20 and 40 miles an hour, they are so confident in their self-driving technology that you can watch a YouTube video while you're in the driver's seat.

2:50:34Which doesn't sound that crazy when you think about like Waymo and what Tesla's doing with a robo taxi. But it is an interesting thing for just like a consumer car that you can actually own and you can drive around. And then you can throw it into self-driving mode. and it will let you watch a YouTube video. So it should let you take a Teams call, I would imagine, or let you use anything. And then it disables the screen if it needs you to take over. And I think the reason that they've rolled out level three at like low speeds in traffic is because even if you get into an accident at 40 miles an hour, if it's a new Mercedes-Benz, it's probably very, very hard for that accident to be truly catastrophic because you're just not going that fast.

2:51:14But I don't know. We'll see how this works out. We'll see if the team's users enjoy it. Anyway. In more news, CBass in the chat says, what did you think of the Tesla-Samsung deal, which I didn't see. I didn't see that. So I pulled it up. Okay. This is a bad look on our part, John. How do we miss a$16.5 billion supply deal between Tesla and Samsung to spur the chip maker's U.S. contract business? Oh, okay. Interesting. So Tesla has signed a$16.5 billion deal to source chips from Samsung. Yep. A move that could bolster the South Korean tech giants, unprofitable contract business, but it is unlikely to help Tesla sell more EVs or roll robotaxis more quickly.

2:51:57Tesla CEO, Elon, of course, said Sunday that Samsung's new chip factory in Taylor, Texas would make Tesla's next generation A16 chip. And so anyways, we'll need to look into this more. It says production is still years away. Okay. So it's unlikely to help Tesla's immediate challenges. It's still just notable that they're not making the chips that go into Tesla's self-driving units at TSMC. That is what's the real news here. Because with everything on the cutting edge of AI, whether it's Apple Intelligence, ChatGPT, Google Gemini, no matter what the chip is, it's fab to TSMC right now on the most leading edge process.

2:52:43So, and interestingly, for a long time, TSMC's most cutting edge fabs were driven by smartphone demand. So Apple would come to TSMC and say, we need the smallest nanometer fab you have. We want two nanometer. We want three nanometer. We want the most cutting edge. And the most cutting edge chip, the most cutting edge process would always go into a smartphone. And the reason for that was not raw compute power. it was compute efficiency and power efficiency. Because if you're putting a chip in a phone, battery life is really, really key. If you're putting a chip in a data center, power efficiency is important on the big macro scale.

2:53:25But at the end of the day, if you're getting a one gigawatt data center with a nuclear power plant and a natural gas peaker plant outside and you have solar panels everywhere and batteries, you have all this stuff. Like power is important, but it's not as critical. like you can just pay for it. Whereas you really can't just pay for more battery life magically in the iPhone. Like the iPhone has to last all day and it has to be a certain size. And so the power consumption is very important. And that's why TSMC has always been driven to the leading edge by Apple and by the smartphone makers. But that's actually shifting now.

2:54:01And so somebody earnings happened last week and they blew them out. And Samsung is up over 6 % today in reaction to the news. Somebody was commenting about the news, Jesse Pelton. Most important news of the year so far is what Jesse says. Elon says, so few understand this. Few understand this. Few. Few? No. Elon says it will become obvious in two to three years why this is big news. Well, yeah, because I mean, I would imagine that what they're going to fab is something that looks like an AI ASIC for something like the TPU, but designed specifically for Tesla's self-driving capabilities, which might be different than a LLM in France called.

2:54:45Made in America. Made in Texas. When you are training a self-driving system, you don't need to know all of humanity's knowledge forever, necessarily. You might not need to train on all of the web text. You need a ton of images, and it might be a slightly different pipeline, and so optimizing around that might be what they're thinking of doing. We'll need to dig in more. But I would see this as similar to like TPU or Tranium or any of these other Apple, like the Apple Silicon chips. Like it is, they are a mag saving company that has slightly different needs for their particular compute demands.

2:55:22And so they are going to one of the two or three leading fabs and making exactly what they want at scale. And they're planning to ship a lot of these because they, you know, I think what Elon is saying is like, we're betting on needing a ton of these chips because we're not planning to sell less cars. We're going to sell a lot of these cars and they're all going to have custom chips in them. And this will be another edge for Tesla on autonomous driving. Exactly. Especially once they add a V12 with straight piped exhaust. Naturally aspirated V12. Back will actually be easier than the Ford Raptor getting custom silicon from Samsung.

2:56:05That's the bull case. Well, thank you for tuning into TVPN today. We have to get on with London right now. But we appreciate you hanging out. Leave us five stars on Apple Podcasts. Leave us five stars on Spotify. We are the number four ranked podcast in technology on Spotify. Thank you to everyone who supported us to help us get here. It's been a wild ride. We deeply appreciate it much. We love you. Appreciate you, especially everyone in the chat, especially Huxley and Jonathan the director and everyone, everybody. Thank you for watching. We will see you tomorrow. Have a great day. Have a great evening.

2:56:42Goodbye.