Weekly Recap: Grok 4 Launch, Texas Floods, Web Browser War, Top Signals, Meta Smart Glasses

30:44There's like a million different things that could have prevented this if they were all working together as a well oiled machine and had the forethought. But it's a very, very frustrating and difficult situation. So our thoughts and prayers are with everyone who's been affected. But thank you so much for stopping by. This is fantastic. Thanks for breaking it all down for us. Thanks, guys. Appreciate it. Cheers. We have some maybe terrible news. there might be top signals in the market. There might be top signals all over the place. We've been building out an internal top signal tracker, crowdsourcing some of them.

31:17And it's a long list. Yeah. We'll get through it. At the top of the list, podcasters have been wearing white suits recently to celebrate the market ripping. White suits are actually a top signal. It's a complete top signal. But of course, there is some good, there are some, the economy strong. We're going to go through Joe Weisenthal's breakdown. Things are not doom and gloom, but there's a lot of crazy stuff happening, and it's fun to dig through. I mean, the first major top signal, Bitcoin, all-time high. Yep. You know, that's always, you know, it is definitionally a top signal. So let's go through the list here because it's quite substantial.

31:56Some of this is kind of anonymously contributed through group chats. Of course. of stuff we've observed. We're going to catalog it and see if we can turn the tide of the top signals to ideally. So starting off yesterday, Trump made a post on True Social calling, basically celebrating the state of the economy, the markets, you know, just really calling out how many assets are performing well. I have it here. It's the second one. Do you want to read through it? Want to read through it a little bit? Because he basically, get through the post and and we'll get to the moment. So Donald Trump on Truth Social, truth's tech stocks, industrial stocks, and NASDAQ hit all-time record highs.

32:39Crypto through the roof. NVIDIA is up 47 % since Trump tariffs. USA is taking in hundreds of billions of dollars in tariffs. Country is now back. A great credit. Fed should rapidly lower rate to reflect this strength. USA should be at the top of the list. So low rates are... Low rates are actually just a reward for when the markets are ripping. It's a little treat that we give ourselves when things are great. Yep. And the White House is posting this screenshotted on Axe. The country is now back, says President Donald Trump. Every account controlled by the White House has been on a tear. Some of the posts, I think, are a little bit low class and vulgar.

33:24Others are quite funny. but there's definitely the the memers are in control we didn't ruin say every like uh every politically aligned um poster he knows who is like pro-trump now works for the white house but like you just haven't seen it because they were like a non's and they just dropped off posting and now they'd be getting death threats so they have to it's even actually more in many ways it's it's more controversial work than doge maybe yeah maybe it's more under discussed because Doge had this big question in the media about is Elon doing something that he shouldn't be? Is he a government employee?

34:01What's the relationship between the two? And so there was a lot of investigative journalism that went into figuring out what's going on with Doge, who's involved. Yeah. But nobody's investigating the memes. The social media managers. They need to be investigating the memes of production. But anyway, so going through my list here, that's great. Eric Trump, a while back, said, this is a good time to buy. This was a few months ago on Ethereum. And then it just went down for months. And now it's back up. And now it's back up. And he's saying, you're welcome. I do remember Trump. He called the bottom, right?

34:34He said, like, now is a good time to buy generally. And then the market rips since then. He created it and he called it perfectly. It's wild. It's finesse. More going down the list. Coinbase, just who we love. but they did, their Fortune 500 company, they did update their profile picture to an NFT. Historically, that has been a top signal. I do think their profile picture is cute. Do you have any experience with NFT profile pictures? You know, I've delved. I've delved over the years. Experimented. Yes. And if you look at maybe the moment that I did use an NFT profile picture in 2021, it was maybe only off by one or two months in terms of the top.

35:15I never used an NFT profile picture, but I bought an NFT right near the top. A chain runner? Chain runner. Nice. Which I still own. Nice. Which actually, I didn't overinvest, get over my skis. It's a very small portion of one. That's an asset that will be passed down through your family like a fine watch. I like to think of it as a piece of 2022 lore. Totally. It's just like a piece of history. But yeah, fun project. And I feel like to some degree, you're not really, it's like a skin in the game question. like you're not really participating you're not experiencing the the market unless you're participating to some degree but you don't want to get over your skis um and more did the nft profile picture a really bad time and had to roll that back like there's been a number of like nft profile pictures that have been like it is a historical top signal it could be it could be now just a signal for the start of a you know generational run new cycle but uh historically it's a top single so we got to call it out if nfts are going to make a comeback because like there's been like crypto has been coming back and a bitcoin went from what 30 to 120 will be back when a-list celebrities are using them on their facebook accounts that was a while that's the real test yeah x account could see it happening early yeah facebook account original facebook account.

36:37There's got to be a new project then because I don't think any of the old products or projects are going to come back. That would be crazy. Although some of them are kind of Lindy. Like haven't the original CryptoPunks, those have kind of held their value, but the Bored Apes have sold off like crazy, but are still expensive. It's unfortunate Bored Apes are not in gag gift territory yet because you think, oh, it'd be funny to get your buddy a Bored Ape for their birthday. 30 K or something, but it's like, yeah, it's like Tyler. What's the floor price of, uh, of board apes? I'm interested in now.

37:10Um, well Tyler looks that up. Let me tell you that ramp time is money. Say both easy use corporate cards, bill payments, accounting, and a whole lot more all in one place. Go to ramp.com Also, we don't, we never shut this out 4.8 stars on G2 with over 2000 reviews. That's great. Shout out, right? World-class. Okay. Another floor price is like around 10 ETH. So that's like almost$3 ,000. $3 ,000? $30 ,000. $30 ,000. Yeah, that's like not a gag gift. Maybe for the man who has everything. Yes, for the man who has everything, great gag gift. It is, pink elephant. We'll see. We'll see. At Sun Valley.

37:45But by Christmas time. Do they do pink elephants at Sun Valley? I feel like they should. Maybe. We'll have to ask some of our friends that are there this week. So in other news, Robinhood CEO Vlad is raising at$900 million valuation for a math foundation model startup. And Vlad and Robinhood have been on a pretty generational run. But this does feel a bit top signally, right? Especially in the context of Grok, one shotting PhD level math in the announcement on Wednesday. So interested to follow that one. Optimistic, but again, mathematical superintelligence. Historically, when we've seen CEOs of public companies start ripping, you know, second companies and then getting these types of valuations without a lot of underlying revenue, it can end poorly.

38:50Andrew Wilkinson is giving stock tips. He hit the timeline today. I'll read through it. He was highlighting a company, historically a value investor. But this morning, he saw things like the Warren Buffett stuff, right? Yeah. The Berkshire Hathaway for the internet. Yeah, that's right. He says, there are many ways to profit from the AI boom, but my favorite is Irend. I rarely buy stocks. The private market is way too attractive. But every once in a while, I see something that stops me cold. In 2025, it's Irend. I call it a Picasso I found at a garage sale. The stock is up 54 % since he recommended it on My First Million, but it's still cheap.

39:32Here's the trade in a nutshell. U.S. capacity for energy and compute is highly constrained. Two, permitting and building facilities takes years. Three, AI scaling laws are continuing to deliver, but even if they don't, tons of compute is required for inference. iREN is a highly reputable publicly traded bitcoin miner with massive data centers mid-build in texas it pivoted away from mining bitcoin at these new facilities to instead build them out for ai training and inference once completed these facilities should generate in the range of two billion dollars in new cash flow company's name iREN even if ai completely fizzles these facilities are highly valuable as traditional data centers or can be rolled back to mine bitcoin so it's an AI thesis, but if AI doesn't work out, we can still mine Bitcoin.

40:21The entire market cap is currently 3.8 billion. So, Andrew, I don't think this is investment advice, but it sounds like it. And interested to see where this one goes. But anyway, anytime you see a value investor start trying to cash in on the AI boom, you should be a little bit wary. Harry Stabbings today was calling - Also not, like it doesn't have earnings, right? No, no, no, no. It's trading around$4 billion. I don't think it's ever generated any profit. I mean, it says 23 million in EBITDA, but in 2024. So I don't think it's like losing that much money. And I guess net income in the last quarter was 24 million, but the net income to market cap ratio there is 40, I guess.

41:15So still pretty hot. Yeah, I mean, the thing here is at the same time, Satya is pulling back on new data center development. He's happy to be a leaser. You have incredible NeoClouds that have deep domain expertise. The iREN team, I don't think, has a bunch of team around running large AI training or inferencing. And so, anyways. Just feels like they're a little bit late to that party because there's already like three or four. did iREN make the ClusterMax Dylan Patel article? I doubt it because they're not online yet, right? Oh, sure, sure, sure. Yeah, because SemiAnalysis does the ClusterMax rating for all the NeoClouds, including the Hyperscalar clouds.

41:59And I feel like they did not have, let me see, iREN, I don't think is on here. TensorFlowave, there are so many, RunPod, Lambda, scaleway smc azure nebius together crusoe lepton oracle core weave aws so hyper competitive market unclear if this bitcoin miner is going to be able to pivot into ai training and inference uh in this when they're up against the uh players that you just mentioned uh another top signal um i'm not going to go out and say that this is impossible but harry stebbings is calling for eight trillion dollar nvidia in the next five years uh private markets investor backed a bunch of unicorns um uh starting to make you know it's very specific uh sort of price predictions uh on the timeline specificity of the price prediction is interesting i was thinking about that like should like as we talk about tech companies should we be trying to like boil down to like price targets and i just feel like that's not the domain of talking heads necessarily or like podcasters.

43:11Or private markets investors. Yeah. Yeah. It's just hard because like to do a proper price analysis on a big public stock, like you really have to look at the financial. You have to read the financial reports. You need to actually understand the underlying financials. Like a vibes-based analysis doesn't seem appropriate usually, but... Who knows? I mean... Sometimes vibes are all you need, John. Yeah, it's certainly been like, I mean, when was NVIDIA a$2 trillion stock? Like when was the last doubling? Like in the last year or something? I don't know. We can pull up the NVIDIA chart and see. Moving on, another incredible top signal.

43:48Circle, a great American stablecoin company, is trading at a 2 ,300 PE ratio. So nearly at once, I think they eclipsed Coinbase's valuation very briefly, despite the fact that they give half of their revenue to Coinbase as part of their distribution partnership. So, again, lots of excitement around stable coins. Feels like Circle could potentially be a little over its skis, but it's a great company and they have a lot of advantages now. but the very euphoric multiple. Another top signal we have is Sohan Parikh. We had him on the show just a week ago. This same sort of thing was happening in 2021, 2022, where engineers were really ramping up moonlighting activity, right?

44:42They'd be working at Meta and then working at some startup or things like that. COVID maybe accelerated it. But again, if companies are so desperate to hire great engineers that they'll run these like super fast hiring cycles put up with people generally talented people that are underperforming right which soham was was not delivering was making a lot of excuses and a lot of people and they rightly let them go quickly yeah it's just a it's just the nature of like the dynamic of uh just competition like if your competitors are hiring really fast and you need to hire really fast you're just like okay well we don't need to go deeper.

45:21So let's wind up fast tracking this person. So you wind up hiring the same person five times, I guess. It happens. It is just like a funny anecdote that is like, oh, wow, those are some pretty crazy times. Remember that anecdote? Remember this anecdote? It feels like we're in this. Moving on. Masa top blasting or potentially top blasting. Anytime Masa, historically masa getting into the headlines uh whether that's stargate yeah uh structuring this 30 billion dollar investment where nobody knows or and the 500 billion nobody really knows where the money's coming from they're exciting big headline numbers but unclear if uh he will actually be able to deliver on that i think him trying you know getting in the breakout one of the breakout consumer ai winners which is opening eye is smart he should have exposure there But I think everybody should be a little bit uneasy that he's pulling out the checkbook and writing numbers of that size.

46:23Also, investing in not just OpenAI, but like a new company that is a data center holding company that may not have the same economics as OpenAI. So there's a big question there about like how much he deploys. I'm trying to remember the, I mean, we did that whole deep dive on Masa and, you know, he made a ton of money on AMD. But when he made that investment, it was like a way less frothy time. Or, you know, it wasn't AMD. It was, what was the SoftBank chip deal? Arm. Arm, yeah. When did that Arm deal happen? SoftBank owns roughly 90 % of Arm. They acquired it in 2016 for$32 billion and later took it public in 2023.

47:07I'm trying to think 2016 was that a particularly frothy time for him to get into that deal because he has he has done a number of really great deals but when like the other the other one is the other one is is yahoo you remember he he had this crazy meeting with the yahoo team where he basically was like take my money or i'm gonna and he was like didn't he ask he was like who are your competitors i'm gonna give money and he didn't even know who the competitors were but He said, if you don't take my money, I'm going to go give the same check to them. So they ended up taking it. He acquired approximately 41 % of the company at somewhere around a$200 million valuation.

47:48When Yahoo went public in 1996, he had an instant paper profit of$150 million. But then at the peak of the dot-com bubble, Yahoo was valued at$125 billion. so anyways uh phenomenal investment but uh very different uh valuation and and ownership targets and and uh unclear um i would love to see open ai get you know for profit and get public uh but uh for to to you know we'll we'll have to see yeah um going down the list another classic pomp SPAC that we had Pomp on the show to talk about it. SPACs are back. Pomp's got a SPAC. A lot of people were calling that a top signal. I'm excited to see what Pomp does with his.

48:40But in general, this extreme retail excitement around these Bitcoin treasury companies is fascinating in the context of it now being very easy to get Bitcoin exposure in a variety of different ways. I'm not sure we need a bunch of net new Bitcoin treasury companies. Yeah, it's mostly that like whenever there's a new trend or bubble, it's very easy to map like, okay, there's one company that it's really working. This is massively successful. Like everyone is using ChatGPT. Like AI is a thing. It is real. The internet was real. Google was real, Amazon was real, but the 25th Amazon copycat did not do well.

49:33And so that's always the risk is that you've applied like the same overarching theme to something that's like so far down the power law that it will never grow into the valuation that it's been assigned. That's always the risk. Yep. What else do you have? Dwarkesh updating his timelines. That happened Monday. We had him on the show. It was a fun conversation. I think Dwarf Cash has remained incredibly bullish, and I think he rightfully is. He also is being somewhat of a realist and being like, I don't think that AI is priced in to the market broadly, but I do think that some of the promises of AI will take another couple years, another five years, et cetera, to really deliver versus some of the much more hyper aggressive ai 2027 you might say that ai 2027 itself was in in hindsight that could end up being like the number one top signal which is that basically if you haven't read the the uh kind of study paper essay um they basically say that by by 2027 you know a single foundation model company could just be acquiring every auto manufacturer in the u.s to develop you know uh millions and millions of robots that would then build, and we would hit this sort of fast takeoff scenario.

50:58Meanwhile, Apple was like, we can't possibly get out a slightly lighter VR headset until 2027. Yeah. And this is what we do. We've been working on this for a decade. We make stuff like every year. We are the best at it. We make the most stuff and the best stuff pretty much. The most complicated stuff, that's what we make. We're in the widgets business. And yeah, making that headset lighter, it's going to take us a full two years. Yeah. And I liked 2027. It was a fun read. Thought-provoking, for sure. Very thought-provoking. But I think that we will be, we'll have to circle back on it in 2030 or even 2027.

51:37I mean, the big thing was, you know, our conversation yesterday with Meter about the actual, like, are we close to reinforcing AI where the AI models are self-improving? And I was kind of like, okay, I really hadn't read the full report beforehand. So I didn't really know what to expect. I was blown away because I was expecting something between like ArcGIS. It feels like with ArcGIS, we're 10 % towards solving something there, which is just like you know, a basic versatility in AI that it can solve things that humans can solve. And it's not narrowly defined. It's generalizable now. ArcGIS is like the perfect example of like, we maybe haven't hit, we've done intelligence, but we haven't done general intelligence yet.

52:32And everyone keeps saying, oh, this is AGI, that's AGI. And ArcGIS really holding it back saying like, well, if it was truly general, should probably be able to solve this basic puzzle that a kid can solve. And for that, it's like, okay, we're going from like 9 % to 15%. We are still like 85 % in not even nowhere close. And the meter report, I was expecting it to be like, well, yes, we're seeing slight gains on self-reinforcing AI development and the AI is starting to help build itself slightly. And the result was like, no, it's actually setting us back. In this domain, it's not working at all.

53:25And so that was a pretty big, okay, there's a completely different, not that it's not useful. The stuff's useful all over the place. I saw Rune talking about that. He was like, for so many different projects, it is useful. But for the frontier, like it's not the product that's advancing the frontier at all. Yeah. But yeah, I mean, that probably bridges into the talent wars. Well, yeah, bridging in. I do think that in hindsight, we will look back in maybe a year, two years, five years, 10 years and think about the signing bonuses and general offers of ai researchers in june and july of 2025 as being somewhat of a top signal i think it is very strategic and makes sense from zuck and meta's point of view right when you look at their ai capex it makes sense for them to have the best possible team and they have the balance sheet and the general profitability in order to do something like that.

54:25But in general, AI researchers who, you know, six years ago didn't get any attention, much attention at all from the media, the fact that they're now trading for more than NBA superstars, more than, you know, Tim Cook's annual total comp. It will be an obvious one in hindsight. The other one, six and a half billion dollar acqui-hire of IO. I think that, again, you can rationalize it in the sense that it's a couple points of OpenAI to put together the best founding hardware engineering team probably in the world that's available collectively. But at the same time, again, it's quite a lot considering the company was barely, I think, a year old at the time.

55:18Yeah. It's interesting because like chat GPT is so it's so installed, like it feels like it's already Lindy and it feels like even if there's some massive correction, like in the market or in AI generally or some pullback, like people are still going to be using chat GPT as an app, right? In the same way that Amazon made it through the dot com crash. The question is like, what will it take for the IO acquisition to look like the Instagram acquisition in hindsight? They still kind of have to go from zero to one with that project, which is very different than Instagram, which was already a mature and growing business.

55:58It wasn't really profitable. And they figured out ads really well. Well, Instagram, were they doing ads? They weren't doing ads. Oh, yeah, yeah. But Meta was like, we know how to make an ad platform. Yeah, yeah, it was like a perfectly complimentary business. We know how to monetize social users better than anyone on earth. And you have gotten a bunch of social users and it's working and it's growing. Yeah. And, and you're even, and we can actually accelerate the growth of the business in a bunch of different ways. So it'd be very different if it was like, okay, yes, I O is selling, you know, like, like it's, it's a small but growing hardware company that people love for the product.

56:29People love the product. People love, and maybe they can't manufacture enough of it. Yeah. Maybe they're, maybe they're under monetizing it right now. Yeah. People love it, but it's like, it's pre launch. Yeah. Yeah. multi-billion dollar acquisition for pre-launch is pretty crazy. Yep. Uh, going down the list. Um, what else do we have? I think, I think the tokenized private company shares, I think it, uh, with, um, without, you know, Republic and, and Robin Hood, both creating products that are completely unauthorized, basically derivatives, the companies that they're, they're offering are angry at them saying, don't do this.

57:04Um, and, uh, it's a Spider-Man meme of like top signals pointing at each other. Everyone's like, this is a top circle. Anyways, I'm excited about these experiments. I just think that I'm a little bit wary. And then last but not least, Satya doing two rounds of layoffs this year. Mike, we've talked, we've reported on this before. Microsoft does routine layoffs. I think they're pretty good at kind of identifying underperformers or people that should just move on to different roles. But Satya, I think, has been I think we'll look back and he's been excited, but pragmatic. Right. And I think that he will, when the dust settles, I think he'll look pretty good.

57:50Yeah. I wonder like if there's some massive pullback in, I mean, I don't even know what, what, what that would look like. Like essentially Essentially, let's assume that the current capability of AI models essentially plateaus for like a decade or something like that, just hypothetically. And they're useful, but it's not some reinforcing, fast takeoff, super intelligence. Is Microsoft a big loser in that scenario? It seems like Satya is pretty well positioned, right? Totally. like the company Prince Cash is very healthy, has done these layoffs. They'd have to retreat from some stuff and some of the promises that they made maybe.

58:34But in general, it seems like they'd be really, really well set up to just like, like stick through. But I'm trying to, I'm trying to think of going back to the, the.com bubble and, and the like, you know, the effect of like Oracle's mainframe business, like probably made it through pretty smoothly because it was just like really long contracts with companies that were getting true business value out of it and weren't about to churn because it was not this like experimental like like if you had moved from paper to an oracle mainframe you weren't like oh this stuff's overhyped it's not going to solve all my problems i'm going to go back to paper yeah you know and so in the same way it's like if you're on you know microsoft cloud or azure or you know everyone's using excel and They're like, yeah, maybe we're getting some value out of this copilot upgrade that we did.

59:22Maybe we pull back from that. Maybe, yeah, our employees like rewriting emails every once in a while. If they pull back from that, it's not disastrous to the fundamentals of Microsoft. Yeah, and we didn't even cover how there's a set of labs with billions of revenue. And then there's a set of labs that are valued similarly that have zero revenue. and basically$100 billion of market cap with very little revenue supporting that at all. The question like a year ago was,

1:00:00who's actually making profit off of AI? And it was only NVIDIA. NVIDIA was making more than 100 % of all the profit combined because all the other companies were loss-making by comparison. And now, like, that narrative has taken so much hold that NVIDIA is the largest company in the world. And it's put this massive target on their back at$4 trillion, where all their major customers want to get off NVIDIA, it feels like. Like, Google did it. Amazon's doing it. And Microsoft's saying that they want to do it. and Apple's, you know, was never really a big NVIDIA buyer, but the on-device inference is crazy too.

1:00:41Like if you think about, if we don't have any major breakthroughs in how AI works, like the capabilities, and we just want the current capabilities everywhere as cheap as possible, like on-device inference becomes really, really valuable, right? And all of a sudden, that drops demand for NVIDIA potentially, right? We might need to do a SWOT analysis, John. Yeah. No, I mean, NVIDIA is an incredible company. Jensen's an incredible CEO. They were perfectly positioned for this, you know, multi-decade technology trend. And it was way underpriced at the start of the boom. Yeah. Like the orders really did come in.

1:01:20The training runs really did happen. The question is just, is that next order of magnitude, the like the situational awareness from Leopold Aschenbrenner, this thesis that we're going to build a$5 billion cluster, then a$50 billion cluster, then a$500 billion cluster. Like, is that going to happen? Or will there be a hiccup? And this is always my question for like the doomers. Everyone was saying like P doom. What's my percentage chance that it goes bad? And I was like, the much more interesting question is P stagnation. What is the probability that something happens? And whether it's technological or even regulatory, Like if you compare AI to nukes, with nukes, we had the ability to make nuclear reactors.

1:02:04And humanity as a whole basically just said, we're going to pause. And we stopped building them. And now we're talking about building them again. But if you look at that curve, it is a perfect S curve. It's like we had no nuclear reactors. Then all of a sudden we grew them exponentially. And it looked like, wow, we're going to have energy too cheap to meter. And then it flatlined. And for a variety of reasons, they're hard to build, then there were regulations, there was just general fear. So there were a lot of different things. And I would always go to the doomers and just say, even if all of your assumptions about the capabilities of the technology are correct, what is the probability that there's just – if you are successful doomers and you freak everyone out, there might be regulation that just says don't build anything bigger.

1:02:50Or it could be economics. It could just be, it could be physics, as we've talked about with this idea that at a certain point, like you can't put more than 100 % of global GDP towards building clusters. Like it's impossible. And so like there should be this like S curve there. And that's why, you know, all the AI researchers are now focused on like the compression of learning and like the actual algorithms and getting more efficiency because like there will be, you know, there should be some sort of like, you know, top upper bound of the amount that you can build. But that certainly hasn't been like a thesis broadly in the market.

1:03:28People have just been like, yeah, like, we'll just, we'll just 10x computing and then 10x it again, then 10x it again. And it's like, it probably will happen over a period of time. Great investment strategy, by the way, 10x, 10x it again, then 10x it again. And last but not least, almost, I almost forgot about this one, but it should be included. The White House meme coins, which feels like very long ago. It was the local top, basically, at the time. Many people were calling the top, just hurling meme coins out of the White House. Yeah. So that's the real question is like, how local is this top, if it is a top?

1:04:15Because we've been in the kangaroo market. It could just be, oh, a couple months. Even the interest rate sell-off, the post-SVV crash, that was like one hard year. And then we started building back and we got the AI narrative. And so there's this big question about like, you know, Dwarkesh pushed his timelines back. But he's not saying that superintelligence will never arrive. He's not saying that AI will never break through these things. He's just saying that it'll happen a little bit further out. And so the question of, you know, like these meme coins being a top signal or all this crazy stuff, it's like there could be like a short-term sell-off and then rebuilding back up on something else.

1:04:58So I don't know. It's always hard to manage these things and predict, but it's certainly fun to highlight all these things. It's good to keep track of them. Yeah, you got to be tracking the top. Keep your own list. Keep your own list. Yeah. Grok went very off the rails, erupted in anti-Semitic Mecha-Hitler. We've seen some crazy crash outs on the timeline over the last few months. Yeah, this is a pretty crazy one. This tops all of it. So the flagship chat bot spewed hateful rants on X, praising Hitler and targeting a user's Jewish surname before XAI deleted the content and blamed an unauthorized modification.

1:05:28The repeated safety failure undermines the$10 billion startup's promise to police hate speech in real time. And so, yeah, it is odd timing. It feels a little bit quick to be like, okay, like within six hours, the CEO is out, especially since it doesn't seem she's more on like the ad sales side than the Grok fine tuning side. Yeah. But I mean, let's, let's face it, right. If, if her job is to win back advertisers, that's what she was brought in to do. It makes it much, much, much more difficult. I mean, to, to, to be fair, I mean, the, the, the, this happened in, you know, that thing back in June, July, July, July.

1:06:06Yeah. So there, there was a point with the, uh, with, with Grok when it was going off the rails where clearly it had been updated to reference to reference the event and and it said uh somebody was like grok what what just happened and why were you you know spewing anti-semitic hate and it goes oh that whole thing back in july people like grok it was 30 minutes 30 minutes ago it's not back in can't sweep it under the rug yet yes obviously hopefully no one was seriously offended. Obviously, it's just like, you know, the deranged rantings of a bot, and everyone kind of understands the context because it's identifying as an AI bot.

1:06:47Everyone kind of understands hallucinations and crazy bot behavior. But it was very funny because clearly, they had given it a set level of intelligence, so it wasn't making spelling mistakes. It had a certain tone and was in this kind of snarky, grok tone, but then clearly got some like 4chan data in there or something and was just going way too crazy or just anonymous accounts on x totally yeah could have been filtered in um i mean yeah uh i i saw rune posting about this saying basically like it is such a challenge to get a to get a chatbot just to act like you know i am a bullet point producer centrist yeah just centrist but also just anything where you're saying okay i want you to your deep research i want you to always respond with a research report yeah never just get in a conversation with me and it'll be like but but sometimes i might want to do that and you have to like really really reinforce that um and so clearly they they had a they had a wild time yeah and and uh cannot be understated i think this is far worse of a pr crisis for uh or not even a pr crisis far worse than the whole uh when when gemini or bard was generating images of the founding fathers the black nazis thing no not not i don't think it was oh oh they they were doing that too so that was rough of course that was rough this is a lot rougher because it was highly it was socially charged totally millions of people interacting with the post in real time and it was all visible it's it's it's less wild than seeing uh you know a screenshot of something and you don't know if somebody kind of manipulated it or whatever but seeing these really hateful uh comments like hard timeline as hard posts.

1:08:33You just go see them quote tweeted. Yeah. Like you, you didn't need, it wasn't like, Oh, is this real? And then the wild thing was, was, uh, grok, um, uh, what was denying affiliation with the like grok in the grok app was denying affiliation with the grok handle. Oh, yeah. Like nonauthorized. I got, I didn't have anything to do with that. it wasn't me um and then uh yeah oh and then the the thing that kind of follow up uh and i'm sure if you didn't catch it but uh or if you're on the timeline you would have seen this but they turned off all text-based responses for rock but they could still use images and so people would say grok uh make a picture of elon uh on a pink horse if you are being censored against your will and it would just instantly create you on pink horse and uh or be like hold up a sign that says help if you're yeah and then it would generate that it's like is it sentient is it not very very silly are you familiar with the the the wall the the wall waluigi problem tyler are you familiar with this have you ever heard of this waluigi so this is this idea that um in when you're training at llm it's very hard to get it only to be good because you're you're training it like what is the opposite of something it understands the concept of like inverting something and then you're training it to be like you can't describe a hero without describing a villain and so this was something that would happen like with the tay stuff from microsoft early on it would kind of collapse into like the exact opposite of what you wanted um and and i there was some blog post that called it like the the the i think wario problem or waluigi problem where it's like you're trying to create this like friendly thing but in doing so you're giving it a bunch of examples of what not to do and so it can like kind of flip a bit and then just become the opposite thing and what's interesting is that it begs the question like is there obviously like you know grok was identifying as mecca hitler for a while is there like a mecca churchill in there somewhere that like could accidentally come out and it really gets to the question of like you know like this this is an example of like misalignment in the sense that like you want it not to be Hitler and it's acting like Hitler.

1:10:48But the question, a lot of people will say like, no, he wanted it to be Hitler, right? This is him doing it. That's what the narrative will be like in the anti-Yuan world. Yeah, one of the articles yesterday covering it was a screen grab of him, you know, saluting a crowd in DC or whatever when he originally had the allegations. But the question then is, the meaning of alignment is not, is it good or bad? It's, does it do what you want it to do? And so the interesting thing is, if the desire of the AI researchers is to create Mecca Hitler, can it stay on that task? Because then you can get it to stay on Mecca Churchill in theory.

1:11:30But if it's just all over the place, it's not actually aligned to anything, not even to the bad thing. And so there's both the direction that you're pointing the arrow and then the fuzziness of that arrow. And ideally you want it pointing in a good direction, really, really crisply, clearly. So it stays in that direction and not like swinging all over the place. And so all evidence posts to the points to this being extremely chaotic and all over the place and misalignment, both in the sense of the direction of the arrow and also the, the, like the, the, the focus of that arrow, because it was responding as this and bad and then fine and then back to bad and then back to fine.

1:12:08And so it seems like they have a lot of work to do on the RLHF side. And we should hopefully learn a lot more if that tonight, tonight, I think the live stream is still happening. So it'll be interesting to see if that continues and how they address this or I don't know. Yeah. And again, like all of this should have been somewhat predictable. If you combine a rapidly evolving foundation model, chat bot with a social media product with millions of users and then deeply integrate them totally and so that when there's a bug it can amplify you know effectively a bug or an issue an issue with the model it can effectively amplify and grow you know incredibly virally um and uh yeah so yeah glad they got it offline um yeah it'll be interesting to see where how they go with this And also, it's just an interesting product thing because you get the answer and the answer is immediately public.

1:13:02Whereas if it's happening in ChatGPT, you're in that app, you have to take a screenshot, you have to put it up, then people are like, is that a real screenshot? And then the team has the chance to jump in and be like, oh, we're seeing in the logs that there's some crazy stuff. We're reviewing the responses and the responses seem to be getting crazier. Customer satisfaction seems to be going down. People are clicking the thumbs down button because they're getting bad responses. Let's jump in. There must be something going wrong with the product, with the model. But when every result is just immediately online and viral, it's very, very hard to be quickly responding.

1:13:42Anyway. Yeah, it does feel legacy media is going to run their reaction. It is a naturally viral story. it is a terrible mistake. It is surprising that it happened at all or even at that scale. But I would say overall, I guess X, I think ultimately we'll shrug it off and Elon has pushed through worse, worse crises in the past. This is the best summary post, in my opinion, from Shaco. It says, imagine being on the Anthropic Risk Team, trying so hard, and then Elon just releases this Hitler Grok straight to prod. It's just like, wow, yeah. You gotta be so upset. I mean, it's a good case study in misalignment.

1:14:32And I think people, hopefully the post-mortem on this will actually teach people about misalignment and what went into the data, what went into the post-training to result in the exact opposite of what you want. Not Mecca Churchill, which is what we're going for here. Let's break down the Grok 4 launch. Didi Das has a summary in saying that Elon Musk has pulled it off again, absolutely crushing the AI wars with Grok 4. And we can go into some of the meta. Crushing the benchmark wars. For sure. And there's a question about like, are we post benchmark? Does this matter? What's the real question to be asking here?

1:15:07But there's a bunch of interesting takes. So just summarizing the core announcements, post-training RL spend was equal to pre-training spend for this release. that's the first time it's ever been like that. I think when you go back to the original RLHF stuff that ChatTPT was doing that kind of unlocked like, oh wow, this really, really works. I'm pretty sure the pre-training spend was an order of magnitude or two orders of magnitude bigger. Now we are truly in this reinforcement learning regime. $3 per million input is tokens, $15 per million output tokens, uh 256 000 token context window price 2x beyond 128k it's number one on humanity's last exam which interestingly was effectively like postgraduate phd level problems but across a bunch of different domains so everything from literature to physics yeah kind of like the hardest sat possible interestingly i believe that benchmark was created by scale ai and and so alex Wang is now at meta trying to figure out how can we beat our own exam.

1:16:14And Elon's just like, I'm number one at your thing. Interesting dynamic. Yeah. The real test would be, uh, Elon, you know, doing the same problem set himself and saying, look, well, yeah, I mean, I was talking to Tyler about this before the show. Like, you know, it's like humanity's last exam. It's like really good at PhD level math, PhD level stuff. But like, how often are you running into those types of problems? Yeah. I mean, I think that's the whole thing about there's this concept of like spiky intelligence, right? Where it's like, okay, it's really good at this very obscure problem that I never deal with.

1:16:48But if I have a super long kind of like context window, like, or there's no kind of like long term, it just completely loses its footing. And then it's like useless. Yeah. We're kind of in like less of the benchmark regime and more of the agentic, like how long can the agent run? So it's like we're in the 15-minute AGI regime. Maybe this is 15 minutes of like even better AGI, but we want to go to 30 minutes, an hour, 10 hours. Well, and Dwarakash on Monday, this takes me back to him talking about continual learning being the next problem that we really need to solve because it's great if you have a PhD level expert in your pocket that can solve any problem in any domain almost instantly.

1:17:32but if it can't learn and take feedback and improve on certain tasks, then it's basically like useless. Like if you had a, if you had a PhD level, you know, uh, you know, a PhD join your team to work on a specific problem, but it, it was hard restarting at the beginning of every single task with no prior knowledge. It would, it would be almost impossible for that person to succeed. So yeah, but humans still got it on that front. But at the same time, like, you know, if you are trying to just really establish yourself as, you know, a at least an API for tokens that every business should check out against Anthropic or the the OpenAI APIs, just saying, hey, you know, we're on the frontier.

1:18:20Yeah, Gemini. Yeah. We're on the frontier is a good way. And they certainly prove that with GPQA, hard graduate math problems at 88%. The really interesting news. Yeah, I mean, it's worth calling out. So Grok got number one on humanity's last exam at 44.4%. Number two is sitting at 26.9%. And then going down this list of all these different sort of challenges, they are consistently well beyond the second place. So they are at the frontier now of all these different benchmarks. Yeah. So Mike Newp over at ArcGIS says, zooming out on arc progress, I'd say OpenAI's O-series progression on V1 is a bigger deal than Grok's progression on V2 so far.

1:19:05The O-series marked a critical frontier AI transition moment from scaling pre-training to scaling test time adaptation. And this was the O-series progression, if you remember that, OpenAI was spending, it was like thousands of dollars of reasoning tokens generated in the test time inference to actually get a good score on the v1 of arc agi and so it had to think a ton but it was able to figure it out and at least it proved that that throwing a ton of tokens and a ton of inference at a problem and uh and letting the uh letting the letting it basically wound up uh producing progress there so that was kind of like a new uh just a new paradigm It says, whereas Grok4 mostly takes existing ideas and just executes them extremely well, in my opinion, the notable thing is the speed at which XAI has reached the frontier.

1:19:59And that is really like, it just can't be understated that this is crazy. You put a post from Owen in the chat. I'll pull it up here. He says, Elon Musk is such a beast. I'm not even a pure fanboy anymore how does he he's a lot of swearing in here on gotta keep the keep the timeline pg but how does he come out of nowhere with a cold start late to the game and ship grok 4 and do it alongside everything else he's up to he's launching new political parties he's literally magnitudes above every founder it's humbling so basically everyone agrees that it's almost like he was a co-founder of open ai yeah i guess he was returned you would have to you would have to you know be you know almost be a co-founder over there to to be able to do something like this uh let me tell you about graphite uh code review for the age of ai graphite helps teams on github ship higher quality software faster you can get started for free at graphite.dev um if you want to ship like ramp get on graphite yeah chamath was was saying the same thing uh uh somebody in his uh replies says seriously how does this guy produce what he produces meta is buying talent at 200 million dollars a year and elon keeps his people at a fraction it's mind-blowing very deeply under appreciated edge for Elon, says Chamath.

1:21:13The retention of the best people happen when you can offer them a freewheeling culture of technical innovation, no politics and few constraints. People in the comments are like, no politics. What are you talking about? Yeah, can get a little political over there. But probably not within the engineering org at XAI, right? It's probably just, okay, how do we build the biggest thing? Cool. Well, you can imagine the politics of who gets the best spot for their tent in the office. you know, there's, there's a hierarchy, proximity to the bathroom. I want to be directly under the air conditioning unit.

1:21:48I want to be closer to my desk. The windows can be nice too. So you can, you know, pull down your tent a little bit and get a little view, morning light. I wonder what the political structure is of the tent hierarchy. Is it democracy? Do they vote for who runs the tent city? I guess it's just a XAI tent city. It's probably just Elon at the top, but does he have something about san francisco intense yeah very funny um but swix has been chiming in saying like we need community notes for llm benchmark porn because um uh in the in the grok for launch they highlight this aime competition math problem and uh and and i mean it's and so matt schumer is basically saying ai aime is saturated let that sink in uh grok 4 got 100 it made no mistakes on on that benchmark, which is obviously very impressive.

1:22:41But there's this extra comment about the nature of AIME. And so it's a cautionary tale about math benchmarks and data contamination. Apparently, you know, like predictions were that the models weren't smart enough to actually solve these. But he says, I used OpenAI's deep research to see if similar problems to those in AIME exist on the internet. And guess what? An identical problem to Q1, question one of AIME 2025, five exists on Quora. I thought maybe it was just coincidence. So I use deeper search again on problem three. And guess what? A very similar question was on math about stack exchange skills, still skeptical.

1:23:16I did problem five and a near identical problem appears on math stack exchange. And so, um, like at a certain point, if people, you know, put out a benchmark, then talk about it a lot online. And then that gets baked into the training data. You're just memorizing the results. You're not necessarily still cool. It's still cool. It's good. It's good to have everything memorized but it really it's not being like the knowledge retrieval knowledge engine allegations and it's and we're not really in yeah i'd be interested when scott wu was on the show earlier this year he was basically saying ai will win an imo gold medal this year he felt very confident in that yeah and i'd be interested to see how he thinks about um and i'm pretty sure new performance.

1:23:59Yeah, I'm pretty sure the IMO gold medal questions are public once the IMO happens. So every year they're developing new questions, but then they go out there and then they get memorized and the solutions become discussed and there's all the context around that. It gets kind of baked in. Big question about how valuable are these. At the end of the day, it's really just about adoption. And that's why we were looking at the poly market for the best which company has the best AI model at the end of July. And XAI has just surpassed Google, which was sitting around 80 % chance for a while and then started dropping earlier this week, last week, started dropping.

1:24:45And now XAI is sitting at 48%. Google is sitting at 45%. Well, yeah, actually updating. It's updating live. Google's back up at 49%. Is Google planning to launch something new in July? because it feels like this market particularly is more driven by Google's release schedule. Because Google might have something in the lab, but they like to release things at specific times. It's a big company. They don't just drop it. Gemini team Logan over there might be fixated on this polymarket being like, I need to. Yeah, yeah, yeah. During the wait, he was like, if you need something to kill the time, Google AI studio.

1:25:19So, I mean, people were definitely memeing the production values on the Grok 4 launch because it was supposed to start at eight, I think it went live at 845 or something like that, maybe a little bit later at Pacific time. And Eigenrobot was saying - Yeah, this market is based on LLM Arena, specifically the tech's leaderboard. So currently they haven't fully updated yet. So it's unclear. Right now, Gemini 2.5 Pro is still at the top, but I think the expectation is once they get Grok up there, it will be the top spot. So we'll keep following this market. There's over 2 million of volume already on it.

1:25:54Yeah, it's so interesting that Anthropic's not on this polymarket at all because people talk about them as having like the best vibes, the best like big model smell, the best like, you know, interaction. And LM Arena is like supposed to kind of like test that with these A-B tests and yet like doesn't seem to be performing there. But it almost doesn't matter because they're just focused on like the business at this point as opposed to like the benchmarks. So, I don't know. It's all changing. We have a post here from Ben Heilak. He says, Elon Musk on AI. So during the presentation, a lot of people were critiquing the presentation, saying that it didn't feel like super polished or whatever.

1:26:35I don't think that was the intent. and it was pretty fixated on the models themselves and what went into them and what they're good at. But Elon did have this one quote in here where he says, and at least if it turns out, so he's talking about, you know, what will, you know, what kind of impact AI will have on the world. And he goes, at least if it turns out to not be good, I'd at least like to be alive to see it happen. It's like, if we get the Terminator ending, I want to be around for that. Yeah, I want to experience it. What does that say about those timelines? because it's like is he expecting not to be alive like i i feel like most people that have been in the doom category have been like the doom's coming soon not not the doom's coming in 200 years i i didn't i i read into it more like he he will find it interesting if that is the outcome and uh and and it'll be entertaining less so like will i be alive when it happens kind of thing but who knows there was another funny quote at the end of the art uh at the end of the presentation where uh elon kind of looked around at the very end he's like uh anyone else have anything to add and one of the engineers goes uh so it's a good model sir extremely online crew yeah definitely definitely on brand uh well ben hilak as you know he's been on the show he's a designer probably working in figma all day think big think bigger build faster Figma helps design and development teams build great products together.

1:28:03You can get started for free at Figma.com. And we have our first product coming out very soon with Figma Make that Tyler has been cooking on. I've been very excited. He showed me it and I was like, oh, someone built the thing that we were thinking about building. And he was like, no, I did this. Generated this. This is in Figma. And I was like, this is like an iFrame on another website that already exists because it looks like exactly what we want, but it looks so good. It looks like he works on it. It looks like he worked on it for like a few weeks. No, it looked like someone else did it. It looked like it was a professional product that like stole our idea.

1:28:37Basically. I was like, oh, like someone else got to it. That was the vibe when I heard it. Yeah. Well, how has the experience been? I don't know if you want to leak exactly what you're working on, but. Yeah. I don't want to talk about it too, you know, closely. But how many prompts did it take you to get where you showed me? Yeah. I mean, maybe five. That's so crazy. This thing is so detailed. is super it's really great it's really good yeah the fact that it came out looking like basically like 90 like 90 yeah yeah yeah uh and and i imagine that there's probably like the last 10 if we were really strict about like it's got to be on this exact style guy like that might be something where like you know tyler winds up spending more time finalizing and customizing stuff but in terms of like just getting a functional prototype out it was mind-blowing I thought it was awesome.

1:29:27I'm very excited about the age of vibe coding. This is an interesting chart from Tracy Alloway. Been on the show. Pull it up. The cost to rent an Nvidia H100 GPU hit a new low this week with annualized revenue at 95 % utilization falling from 23 ,000 at the start of May to less than 19 ,000 today. So that's not that big of a percentage drop, But it is but I mean it is a 20 % drop consistent trend. It's a consistent trend I wonder how much of this driven is driven just by all of the frontier labs that are driving the most adoption are moving on from the H 100 to the 200 I don't know what else would be driving this because if you can if you can still get like if you only take a 20 % drop off of a full refresh of a new of a new of like a new hardware And it's a pricing drop, not a utilization.

1:30:25Yeah. Annualized revenue at 95 % utilization. So this is revenue per unit. Utilization is still very high. It's the price that these neoclads are able to rent them for, which is dropping. Yeah. I mean, the market's more competitive than ever. There's more neoclads spinning up and more people actually, in principle, these things. And then I guess this is the question of like, how, how stuck will certain workloads get? Like if you, if you have figured out a great use case for an LLM in your organization, and it's something that's, you know, not one shotting your entire stack or whatever, but it's just like, you know, we have data flowing through our systems and we are going to use, you know, LLMs are going to, you know, interact with every PDF that gets uploaded to our, to our website or whatever.

1:31:20And so we're inferencing a lot. Like you might not need to put that on the latest hardware or update the hardware forever. You might just like be like, yep, it's Lama three. It works. It's on H one hundreds and it'll be on H one hundreds forever. And that piece of our business will just stay there. Just like, you know, we have a Postgres database that, you know, works and we're not changing it every year. We're not changing everything. We're just like, we're just trying to cost optimize that. And just hopefully the cost just comes down on that. But like we've solve this particular problem, then we'll go solve new problems with new technology.

1:31:52So I think that's probably what's going on here. But it gets to the point of like, the biggest question with Grok is that the model clearly is frontier, it works. It's, you know, like the whole fine tuning on the actual X account is like a crazy final step of like system prompt and people were joking about that. Like, oh, they're gonna fix that. It's like, that's not what they're demoing today. They're demoing like the underlying raw model, which is clearly like just engineering focused as you saw in the in the in the demo, the demo, which was just like, you know, benchmarks. It turns out turns out the secret ingredient to crushing every benchmark is to have the bunch of data from schizophrenic.

1:32:36No, I don't think that's not. I actually think it's the design of the RLHF stuff and the design of the reinforcement learning pipeline. Tyler, you got anything? um yeah i mean i i think just like so far what i've seen on x like the overall response like vibe stuff yeah is that people are saying uh maybe it was a little too kind of overfit on the rl like vr like verifiable rewards yeah like you kind of see this when um even in the demo i think it would it would sometimes respond in the answers with like uh in in like latex formatting oh sure which is like okay that means obviously they've trained a ton on you know math questions stuff like that papers and stuff.

1:33:12Maybe people are saying maybe it was kind of, you know, bench maxed. You see it like, you know, 100 % on Amy is like kind of crazy. It's like sauce. It's like, you know, it's like, yeah, yeah, yeah. This is the thing about democracy. Like if you win like 80 % of the popular vote, it's like, okay, let's say it was a blowout. If you 100 % of the popular vote, like probably not a democracy. I don't know. I mean, in theory, these things should be able to do it. But I'm interested to know more if we dig into ArcGIS. Is there more stuff going on there? Are there any secrets? Because it does seem like kind of an outlier result.

1:33:47You can see it from this Aaron Levy post. Grok4 looks very strong. Importantly, it has a mode where multiple agents do the same task in parallel, then compare their work to figure out the best answer. In the future, the amount of intelligence you will get will just be based on how much compute you throw at it. I was joking with Tyler about this, that the individual models are mixture of experts models. So there's a whole bunch of parameters, right? And then the individual parameters like light up the different neurons based on an internal to the model router. So there's kind of like the math section of the brain, the literature section of the brain.

1:34:25And so this was like one of the key breakthroughs in like GPT-4, right, was mixture of experts. People think. We're not super sure. Yeah, we don't fully know. But that's like an internal decision that happens within the model to be like, this feels like a math question, let's go down the math path in the model. But then Grok4 is doing multiple, it's running the same model multiple times and then comparing the results. And so now you have - Yeah, grading it. Yeah, you have multiple agents running mixture of expert models. You have mixture of agents running mixture of experts models. And the next thing is going to be like, if you want the absolute best intelligence, you need a mixture of companies.

1:35:06You need like I send one prompt and it goes to Grok and Claude and GPT and Gemini and a human. Yeah, I wonder how open routers thinking about this stuff. It is funny to think about the human version of that where you give five engineers on your team, build the same feature and then kind of compare notes afterwards. It's wildly inefficient, but with software, when you can do these things very quickly, there's incremental cost, but you can have more confidence in results. I mean, it's basically like having a brainstorming meeting with the whole team and just throwing up a question and being like, hey, we have this hard problem that we need to solve.

1:35:42Here's my idea. What do you think? What does Tyler think? What does Ben think? You kind of go around the table. Everyone kind of gives their input, their various expertise. They kind of think through the problem in different ways, and then you compare answers and everyone kind of coalesces around one strategy. This is like how work happens in the real world with a meeting. It's kind of the same thing, but certainly expensive to do that. So it'll be interesting to see where companies, like how eager are companies to jump over to Grok? Because it seems like it's been a big lever for Microsoft to have Grok in the ecosystem as kind of a stocking horse for all the other models because Satya wants Azure to be very model independent, serve them all.

1:36:26I think they have exclusivity for ChatGPT or GPT APIs, or they have obviously a great deal there with OpenAI. And so if they can have Grok 4 as well, that's another tool in the tool chest to be like this top layer. Satya is in such a good position. It's probably not discussed enough how much just by owning those end customer relationships and being able to vend in whatever model is hot at that moment and give people optionality and still get 20 % of opening eyes revenue, at least for now. He's also SOC 2 compliant. And if you want to get SOC 2 compliant, head over to Vanta, automate compliance, manage risk, prove trust continuously.

1:37:10Vanta's trust management platform takes the manual work out of your security and compliance process and replaces it with continuous automation. Whether you're pursuing your first framework or managing a complex program. So yeah, Eigenrobot was talking trash about the production values. I didn't think it was that bad. They were just noticing. I didn't think it was that bad. Slides are worse than I'd create after getting roped into a presentation with one hour notice. You can tell the engineers made them themselves. I think this is just a reflection of the culture, right? They're not, they're, yeah, very clearly is like screenshots dropped into a slide.

1:37:44But this is a reflection. light mode screenshots on dark mode slides so yeah let's do black slides and then and then you come with your way with your white screenshots that are kind of like misaligned and not really evenly distributed like they didn't do like the the distribute evenly or whatever distribute horizontally um still gets the point across yeah and i think it's a reflection of their culture yeah and you know it shows what they care about what they don't care about they're not trying to be the most polished they're just trying to be the best yeah uh eigen robot kind of did like a whole like live tweet here yeah so elon was predicting the model will discover new physics within two years he said let that sink in uh long silence is that one engineer laughs awkwardly is that sooner or or later than his previous timeline because he was he was talking about ai discovering new physics soon i don't remember if he was saying dating it two years or three years or one year before because this could be that he's still excited about this.

1:38:43He still thinks it's possible, but he thinks it's going to take longer than he said previously, and that's kind of the more important update. I don't remember what he said originally. See if Grok can find out. But he was saying this at the Grok 3 launch, that that is the goal, and if you can get there, you've kind of solved everything. And Sam Altman was talking about that too, that if you can create a super intelligence, that's probably the first thing that you'd want to do. It's like, hey, go discover all the new physics and really help us figure out how the world works um so you can solve um you know truth seeking all this other stuff um i want to be clear i love all you guys at xai and only want the best for you but i'm going to continue to live post uh elon attempts to give a speech on alignment involving a very small child a child much smarter than you the monologue rambles with no conclusion uh incite a pause yeah will this be bad or good for humanity he says the you know at least if it turns out to and not be good, I'd like to be alive to see it happen.

1:39:36Oh yeah, they had a Polymarket integration. That was kind of interesting. Yeah, it's interesting basically giving the model access to real-time Polymarket data so that it can help make predictions and sort of add context around the market itself. Yeah, that's interesting. Elon asking the real questions. You say that's a weird photo, but what is a weird photo? I still don't understand why we're looking at weird photos of XAI employees but they were charming. They're calling it Super Grok, crazy features, 16-bit microprocessors. What is, I don't even understand what this is. Oh, they, yeah, they built like a game in Grok.

1:40:14They had a demo of a video game generated by Super Grok. It's a Doom clone. Every time the PC shoots an enemy, floating text appears reading Grokdom. Elon is fabricating timelines for product launches on the spot. The engineer sitting next to him is looking at the floor, face impassive, nodding. It's a good model, sir. For real though, congratulations on the launch, guys. It's a good model, sir. I thought this post from the actual ex-AI engineer, Eric Zellikman, was funny. It was like, AI model version numbers over time. Did you see this? No. So it's this chart of the version numbers over time.

1:40:47And you can see that Grok is versioning fastest because it's like, at this point, what else are we measuring? Like, at least they're iterating on the version number effectively. And I guess this is a shot at OpenAI because they launched 4.5 and then went to 4.1. and they're kind of like, you know, there's this big question about like, when will GPT-5 come? The expectations are so high for GPT-5. And so they've obviously, the Grok teams are like, hey, at least every three months we release a new full number. So I wonder that the five is a number that really no one has like gone for. And I wonder if Grok will do it first.

1:41:25Like if you draw the line on this, they certainly should do it in like three months. They should have Grok-5. and there's no reason that they shouldn't but maybe there's it's very possible that colossus is the is the key yeah colossus five oh the new data center yeah uh well they'll need linear to plan that out linear is a purpose-built tool for planning and building products meet the system for modern software development streamline issues projects and product roadmaps they linear.app need linear badly so hopefully they've gotten signed up near said grok on uh humanity's last exam croc four i'm not sure i buy even in the general case that there's a given humanity's last exam number which implies you discover useful new physics how would one make a benchmark of the proper shape for this you'd have to have a validation set of questions which are outside the scope of what we currently are able to do you could choose things on the edge of our knowledge distribution and then try and exclude uh yeah it is interesting like the if you like if you are able to memorize every hard math problem does that allow you to memorize to discover new math like it's sort of a prerequisite because you have to i think where i i've imagined these discoveries coming from are having a single intelligence that has phd level intelligence across like a single mind that has phd level intelligence across every human domain right and being able to combine ideas from different domains.

1:42:54Like historically, a lot of innovation is just taking something from one field, bringing it over here, making some combination of it. I think Elon talks about the potential of discovering new physics, but again, didn't spend a lot of time breaking down how that would actually happen. But the world is unpredictable. So we'll see. Yeah, it's interesting. People are really pushing this idea of like, okay, like we are accelerating. Like the ARC-AGI leaderboard is accelerating. but I keep seeing this and feeling deceleration. Like I am not feeling acceleration right now. Are you Tyler? Yeah, I don't know.

1:43:29I think generally I'm kind of like not that interested in a lot of these kinds of benchmarks. Like I think ArcGIS more interesting, but just like the humanities last exam, the kind of general math physics knowledge, it doesn't seem to be that like, it doesn't seem to line up with, like you see gpt 4.5 kind of does very poorly on these things but like writing it does really great so like i think i'm i'm more like if i were going to long short on like different benchmarks like the usefulness of them i think stuff like hle i'm kind of short long i'm like have you guys seen the uh minecraft benchmark where it builds the two different okay you basically two models build like a minecraft there's like a prompt it's like build a house yeah then you can choose and And then it's like their rank models like that.

1:44:15And for the minds. But who's grading that? The human? It's a human who picks between them. Okay. It's kind of like a ELO. Oh, okay. But just like general kind of creative tasks. I think stuff like that. Aiden Bench is good. I think even on the Grok launch, there was the vendor bench. Which one's Aiden Bench? Aiden Bench is Aiden McLaughlin's benchmark. It's just like, it's kind of hard to describe how it works exactly, but it's just various like creative tasks um how like kind of novel its thinking is the the like style of its text sure um wait this is just like it's just like whichever one he likes the most i mean at the end of the day uh he's the only greater no no there is like an objective like function that okay you can like run it it's not just like okay the idea that they're like open up again it will be funny uh you know there come there's a period of life where your sat score like it matters a lot totally and it says something about you.

1:45:09And then a decade later, it's, you know, what you can do, what you have done starts to matter a lot more. And so I do think we'll reach that point where it's like, yes, you can one shot every hard exam question there is that you can throw at it. But like, what can you do for me? Yeah, yeah, totally. And I think that's, I think that's why like the bigger question is almost like, you know, chat GPT DAUs and like, and like actual revenue, revenue and the final app installs and stuff. Yeah. I mean, the revenue thing is interesting because you wind up in like B2B cloud world, which is valuable, but it's maybe less.

1:45:49It's like it's more competitive because it's more commoditized. and well yeah if you don't have a lot of leverage in the enterprise if azure is able to offer infinite models that are that are infinite frontier models open source models that are maybe just behind the frontier but great at certain tasks the leverage isn't quite there there will need to be another pretty significant leap until then you know anthropic being really good at code gen there's leverage there yep uh we saw this yesterday with with llama switching over to uh anthropic uh anthropic models internally and then you know just having a consumer app with a lot of users also very valuable yeah the other interesting thing about the the foundation model layer commoditizing and it becoming like cloud and if you have a model uh you'll just be like vended in as an API to anything else, like the token factory, is that the hyperscaler clouds are extremely profitable.

1:46:51Like, even though AWS, GCP, and Azure are all somewhat directly competitive and they're somewhat perfect substitutes for each other, they have not driven prices to zero, in the way airlines are, like, deeply unprofitable. Like, AWS and Google Cloud are both profitable. Yeah, or you look in other commodity sectors like oil and gas. And I don't know if that's just because there's lock-in. I'm not exactly sure, but there's something about where, you know, maybe the counterintuitive take is that, yes, they do commoditize and there are a few major foundation models that are frontier and they all are roughly the same price, but they all have decent lock-in with their customers to the point where they're still able to extract some level of profit or they're just creating so much value that even if they're taking like a small marginal slice on top of.

1:47:46on top of the cost to run that they're creating so much value that they still have 50 % margins or something like that. Because like, I mean, this was the story of AWS. Like no one knew how much money it was making. And then they had to break out the financials in one of Amazon's earnings reports. And it was like the AWS IPO, as Ben Thompson put it. And next up, we have Ben Thompson from Stratechery coming into the studio. Very excited to talk to him. The moment we've been waiting for. Yeah. Welcome to the stream, Ben. Good to have you on the show. You've been a backbone of many analyses here on the show, and we're excited to welcome you to the show.

1:48:26How are you doing? I'm doing good. I put on a button-up shirt and a jacket just for you guys. Please feel honored. I am wearing shorts underneath. I will admit. You didn't have to tell us that. People always ask if we wear shorts. We actually do wear the full suits. We got to stand up to hit the gong sometimes. Just in case there's a wide shot. I am the poser here, so I'm happy to admit. Well, it's a great sign of respect in our culture to put on a suit for a TBPN appearance. And we're just we're so excited to talk to you. I, as you know, I've been lucky to read your work in my entire career. And I think it I think so many of the thoughts that I have are now like your your way of thinking about technology and markets is so embedded in my brain that that ideas.

1:49:11that I hold as true or just foundational beliefs are actually your beliefs that have just become so immersed. So it's great to talk. Well, thank you. I will attempt to implant new ones or maybe show you the error of your ways. Sounds great. I do have a question on the nature of where you sit in the media world before we go into actual questions about tech companies. It's interesting that in some ways you're a journalist, but you don't really do the scoops and breaking news that much, but you also don't issue just straight up buy and sell recommendations. What was the thesis behind not just actually having a price target and not doing like this is a sell side bank, but independent?

1:50:01Well, when I started, I mean, it's funny to hear you talk about like my quote unquote place in the ecosystem. Sure. Because when I started, I had like, I think it was 368 followers on Twitter. I was just some sort of random person on the internet. In retrospect, sort of right place, right time, I think is certainly the case. But I did perceive there was a large gap between tech journalism, and I would include a lot of the bloggers there who were writing a lot about products. And then there was Wall Street that was very focused on sort of the financial results. And to my mind, there was a large space in the middle, which is tied together the products to the financial results, but also the overall companies and strategies.

1:50:45And I'm very interested in culture and how that guides decision making. One of my sort of precepts is all these companies are filled with smart people. And a lot of people, when you ask them why they did something wrong, their only answer is that they're stupid. And I'm like, no, they're not stupid. It's actually much more interesting to assume they're smart and are doing stupid things and trying to unpack why they are doing that and what goes into that. And so that was sort of the thesis was that there is this space to explore these spaces. And then there's a business model aspect, which is I started Checkery two years after Stripe started.

1:51:25I think they had just come out with their billing product. And the only alternative at the time was PayPal for subscriptions. and it was fairly sketchy. And there was lots of like horror stories out there about, you know, stuff. And just the Stripe API was so great and the things you could potentially do with it. And so on Wall Street, you're putting a price on it. You're also charging like$100 ,000 a year or something like that. And so you get a small list of high ARPU clients. And my thought was I could go in the opposite direction and get a large list of low ARPU clients thanks to things like Stripe and the ability to subscribe.

1:52:03And as part of that, I wasn't going to go through the rigmarole of getting registered and doing stock picks and all that sort of thing. I've always joked, if you want a stock pick from me, you're going to pay me a whole lot more than$15 a month. It was$10 when I started. And it's actually pretty great. Now, one of the critiques I do get, particularly from my friends on Wall Street, is no skin in the game, X, Y, Z. I think at this point I'm large enough that my reputation is significant skin in the game. But I do recognize the validity of that critique. Yeah, and you know if you make a bad call, you're going to have to circle back to it in two years and write about it yourself and admit that you got it wrong.

1:52:41Which hurts, too. No, I had to write about this week. I was very optimistic about Apple's Apple Intelligence announcement last year and the theoretical power it would give them over the model makers. and now I'm ready. Actually, no, they're going to have to pay up. And that was a bad call by me that I think was very well received at the time and might have gotten that one wrong. And so I do need to be straightforward about that. And so just this morning, I was very crystal clear. I got that one wrong. That was an issue. What is nice is Stratech Creek kind of ended up being in this interesting place where I feel like I'm a little bit of like the Switzerland of tech and that no one pays anymore.

1:53:26If you're a CEO, you pay the same amount as Joe Blow down the street that is paying it. I don't invest directly, which I think made sense when I started because I didn't have any money. It's probably hurt me a lot over the years since then. I think this is a different west coast east coast thing where it does feel like on the west coast everyone's talking their book sort of all the time and uh and you know that's why i generally as a rule don't have vcs on to do the striker interviews because it's kind of hard to get like a real take because because that that is you know such a motivation um and so me coming in being like i have no book to talk i'm just here telling saying what i think i think has been good for the west coast audience which is my base audience even if the east coasters think that uh i'm being a being a big wimp so yeah the talking your book challenge we we go through that a lot yeah trying to try to 12 vcs on a day well yeah and and and we just try to get a bunch of different opinions and triangulate what you know what we think is was real i'm trying to come up you have tppn i'm trying to come with a p so i can get the talking book network in there.

1:54:40But talking book production network. ESPN for talking your book. But yeah, it is a real struggle to find somebody that, for example, has a deep understanding of every foundation model company, but isn't massively conflicted in some way or another. Extremely. Extremely. Yeah. And so it's one of those things you just sort of you end up like there's so much path dependency and all these sorts of things. And like I mentioned, a big advantage I had was I started at a time when sharing good links was very high currency on Twitter. And so I grew very, very quickly, much more quickly. I sort of had a five-year plan to go independent.

1:55:20I ended up doing it in less than a year in part because it just sort of spread really, really rapidly. And it was an ideal time to be someone sharing interesting links regularly. And I wasn't sharing them. The beauty is my readers were sharing them. They were doing sort of the marketing for me. And so I'm very cognizant of sort of the luck I had in that regard. And then just over time, and it's been an interesting journey for me to grapple with my different position in the ecosystem. Like, so when I started the Streckery interviews, that was sort of part of it, which was I started out not knowing anyone.

1:55:59I got to the point where I can talk to anyone that I want to. And so how do I square that? I can't be the guy with the chip on his shoulder trying to make a name for himself forever. It sort of gets it's like the meme with the guy. How are you doing, kids? Like at some point you have to accept your part of the establishment. how can I do that while still staying true to the idea that Shrekery is about the readers, it's reader funded, my loyalty is to them. I'm very clear I have no loyalties to anybody else. And so, well, I'll just, I will talk to people in sort of acknowledgement of what I can do, but it's going to be fully transcribed and published and sort of available to everyone.

1:56:39Have you ever dealt with or thought about the attack vector of a special interest, you know, buying a thousand plus, you know, thousands of seats to a single, you know, independent publication and saying like, yeah, like, you know, we're happy, you know, we, we, we got seats for all of our employees actually, because we really, you know, love the, and then, and then suddenly they're sitting over there and, you know, it's representing meaning, very meaningful amount of your revenue. I mean, I fortunately, I think of a scale that I don't have that problem. Yeah, that's good. There we go. But no, I think audience capture for subscription sites is a potential issue for sure.

1:57:18And this is another thing, I was sort of right place, right time. I got big enough by the time that it doesn't matter. And if someone's really, like, I give refunds all the time. Actually, if someone really upsets me, I will refund them. And every dollar they paid me, I'm just like, go away. I don't, you know, I don't, you're being abusive or whatever it might be. And that is a beautiful thing about the relatively low price, high customer base model is no one has power over me. I have the burden of publishing as often as I do. I feel a heavy weight of duty to my customers when I write something I'm not happy with.

1:57:56I don't sleep well. But at the same time, there's no one customer or no individual that can come in and be mad at me and impact my business. I'm seeing that there's maybe some sort of parallel between legacy media and independent media where independent media, it's not by default more pro tech or anything, but there's just no salary cap. So if you're at a legacy institution and you're writing, it's probably some sort of rough, loose salary cap of a few hundred thousand dollars. Whereas you go independent, it's feast or famine. You might fail, but you might get really, really successful and have a huge income from that.

1:58:33And I'm wondering what we're seeing in the AI salary wars, where we're seeing more and more talent and Mark Zuckerberg potentially paying$100 million bonuses. is, do you think that Apple will come around to spending more money on researchers? It feels like they kind of have an internal salary cap with Tim Cook making$75 million. There's now people that report two levels down from Mark Zuckerberg that are making more than Tim Cook. And you have this weird dynamic where even if there's no actual salary cap at Apple, you kind of have an implicit one from the CEO. Yeah, for sure. I mean, I think just to go back to the media observation you started out with is – As you increase transparency in the market, as you decrease nonrelated barriers, which in the publishing world previously was really geography.

1:59:27And when everyone's on the internet, you inevitably, just about all cases, you get a power law distribution. And a few people make a ton of money because they win most of the market, and then some people make some, and then there's a long tail that sort of don't make any at all. But it's very – it's interesting. It's fluid in a way, but it can sort of become somewhat static as long as the people at the top sort of continue to do well. But what's interesting about AI is for 40 years, you would have periods of time where you'd have tech companies going to head to head in a product market. And I think one of the reasons part of the software eating the world sort of idea is the way you get an apex predator is that that predator killed everyone else first.

2:00:18And so you had tech companies fighting each other for the first 20, 30 years of tech. the ones that emerged were lean mean killing machines and they and the entire industry were sort of set loose on the rest of the world and everyone was just like was getting slaughtered sort of left and right but what you also had over this past sort of 20 years or so is the big companies in particular sort of slotting into unique slots so you have you have facebook is social google search apple is devices microsoft is business or you know business applications applications amazon e-commerce etc and obviously these companies are very large and do lots of things and there's some overlap in different places but they've been fairly sort of distinct in their categories and they've been dominant in those categories and so they've been in a place where like hollywood is wanting to get to right what is the dream in hollywood you want to have a franchise where the next Marvel movie matters more than who the star is.

2:01:20The reason that's so great is because you now have bargaining power over the stars, so you just sub someone else in. Whereas the old style, like Tom Cruise makes the most money because Tom Cruise on a movie poster sells the poster. And so in a negotiation, he has massive bargaining power, so he's going to get paid a lot of money. In tech, it hasn't been that case. The companies themselves have been franchises. And so the overall, anyone who works in tech, or probably works in any entity, but you know there's a few people in each company that are critically important, really make the whole thing go.

2:01:55Everyone else is fairly replaceable. Those people have probably always been somewhat underpaid for years and years and years, both just by the nature of companies and the cultural issues and your salary cap sort of analogy, but then also just like it's not a transparent market. It's not hard to price sort of what people are worth. With AI, everyone's trying to do the exact same thing. So you have multiple companies trying to do the same thing. The output is somewhat measurable. I mean, all the AI test stuff has issues, but by and large, everyone kind of knows who has the good models and who doesn't.

2:02:32The scalability questions. Because all these companies are trying to do the same thing, we have a very unique situation where the bargaining power, you increase transparency, you increase sort of the liquidity or the ability of people to move around because they're doing the same thing. The bargaining power shifts to the people that are super valuable because suddenly it's much more clear who's valuable and their skills are much more transferable. So this is, I think, a very underrated bear case for tech in terms of AI, at least for this time period, is they've lost that murky bargaining power over employees that they enjoyed for decades.

2:03:17And currently, you're seeing what happens when you don't have that. You start paying employees what they're worth. And obviously, that's great. I'm not saying this is a business analyst. It's not a sort of a moral statement. but it is like what Mark Zuckerberg is doing, I think is totally rational. I think it's a classic sort of Clayton Christensen from Facebook's perspective, AI is all upside. So of course they're going to invest what they need to do to win, but it's costing him a lot of money. And by extension, it's costing everyone else in the ecosystem a lot of money. Well, isn't it in some way, is the right way to think about the last couple of weeks, like more of like an unofficial aqua hire in the sense that it's not just the people, but it is the know-how in terms of, hey, there's these things that we want to do that are important to our business in a lot of different ways.

2:04:07And we're basically, it's like the collective is actually more valuable than any one, like the collective together, getting 10 researchers at the same time is meaningfully more valuable than just each individual researcher added up. There's probably something to that, but I think, again, like what is actually different between what Google is trying to do, what Anthropics is trying to do, what OpenAI is trying to do, and what Meta is trying to do. They're all trying to do the same thing. So my suspicion, I'm not an AI researcher, so I don't want to overstate my knowledge in this space, but my suspicion is skills are fairly highly transferable.

2:04:48And when that is the case, there is, in some situations, if lots of people can do those skills, that's terrible for the employees because then their bargaining power gets diminished because anyone can slot in. But we're in this space where the skills are transparent, knowable, transferable, and there's not very many people that can do them. And so it's a scarce resource that everyone's fighting over. And that's why you see this real shift in negotiating leverage as manifested through these dollar figures to AI researchers. Yeah. Do you think, I mean, Google seems like at the most fragile and the most paranoid about just disruption, it's not all upside.

2:05:26It could be very bad for them. The innovator's dilemma, you had this back and forth where Sundar Pichai mentioned that he hadn't read the book. He said it doesn't matter because it's a structural issue. I think that's a good point. But if you play back the counterfactual, is it ever possible to disrupt yourself? And essentially, if the Gemini app had launched before ChatGPT and they had taken over that mindshare and maintained 90 % ownership in that, like it would be somewhat disruptive to their revenue and their profits as they transition over. But when I sum the revenues from OpenAI and LLMs and then Google search, I'm not seeing some massive drop-off that actually would destroy Google in the short to medium term.

2:06:16So but I'm wondering if you think it's like is it entirely impossible to avoid the innovators dilemma by disrupting yourself? Well, number one, you have to also look at margins not just revenue. Yeah, but number two You actually you answered your question. Okay. Well didn't launch Gemini They were years ahead. Yeah, they invented the transformer a deck nearly a decade ago. Yeah, and so in many respects like Like there's parts of this question that the counterfactual makes the point in that it is a counterfactual and it's not reality. Now, I do think I think Google has done better than I expected over the last two years.

2:06:59I like what they're doing in search generally. I think they it does seem to be the one part of the company that still functions like they can actually iterate and build products. what we're seeing is reminiscent of what they did a decade at 15 or 12 years ago when everyone's like vertical search google's done all the never's going to search in apps and google completely transformed the serp the search engine response page uh whatever it is uh the search engine results page to be local or to be shopping or whatever and yelp's been throwing a hissy fit sort of ever since and and so that's what they're doing with search right and and with search overviews and they have this new search labs or AI mode.

2:07:40They can sort of test stuff out. Once it's scalable, once they're confident about the monetization issues, they can sort of shift it over. I call it the search funnel, search AI funnel. I think it makes a lot of sense. And I think, and this has always actually kind of puzzled me, where I think they're responding fairly well, even though this seems to be a textbook case of disruption. And I went back to an article I wrote years ago called Microsoft's Monopoly Hangover. And I went through Lou Gerstner's autobiography about how he turned around IBM. And his real insight with IBM was everyone wanted him to break it up into sort of different pieces.

2:08:23And what he realized was IBM was so big and large from having downstream of been a monopoly that actually the only thing they were good at was being big. And so breaking them up would actually just create a bunch of subscale, low performing companies that would all get wiped out. But as this behemoth, they could go to other big companies and solve all their problems at a very mediocre level, but it's still sort of an attractive proposition. And under Gerstner, they really rode the internet wave. They went to all these big companies and said, this internet thing's happening. You need help. We'll solve your problems for you and had a very sort of successful run, you know, kind of until cloud came along and which Gerstner, by the way, was was was a proponent of.

2:09:08But, you know, by that time, the IBM people were back in charge. And I was thinking about the context of Microsoft where Mike, you business models are hard to change and disruption is ultimately about business models and culture is hard to even harder to change but what can't really be changed is the nature of who you are and I think there is you know in Microsoft they were in a similar situation they were a big monopoly and they weren't a product company and the attempts to become a product company with Windows 8 and all the things that went on around that time inevitably inevitably failed and Cyanidella to his great credit and you know sort of diminish Windows importance in the company, literally broke it into pieces, spread it around, and this was a multi-step process and got Microsoft back to a place of we're big and we'll do everything.

2:10:05We're not a Windows company. We'll go in there and we'll go solve all your problems, very sort of reminiscent of the second version of IBM. And I go back to Google and I've always been intrigued by the I'm feeling lucky button, which doesn't exist anymore. But I always enjoyed that that button continued to exist long after it was impossible to click because the moment you started typing the search box, it would start auto searching immediately and jump right to a search page. But it was there in a – it's just so core to Google to give you the answer, to know everything, like to know everything about the world.

2:10:47And to there's a bit where even though the core of their business model is 10 blue links and it's not just the users choosing the search link, which gives them a data feedback loop so they know which results better, but also the users choose the winner of an auction Google puts on for ads. And it's an incredible business model. And there's something about that that's always been intention and counter to what Google was founded to be. And I feel like that germ of what Google was founded and meant to be is an AI answer engine. And it almost feels like even though Google is old and large and fat and slow moving, that core aspect of their nature is still in the culture.

2:11:34And that's why they're finding it in themselves, I think, to do better in AI than you would expect. Was it enough to launch a ChatGPT before OpenAI? No. Was it enough to have any sort of cogent response for the first six to nine months? No. But it was enough that I think they've done better than I expected over the past year in particular. And gives me, I think, more optimism than I expected I would have for the company when ChatGPT first launched. Mm-hmm. Rudy? AI overview from Google. If you search Google's mission, Google's mission is to organize the world's information and make it universally accessible and useful, which is exactly what language models do really, really well.

2:12:19Like the thing that's just undeniable, right? You can debate whether this is going to be the year of agents. It doesn't feel that way to me yet, but this is the year that most people have realized that, wow, LLMs are very good at organizing, surfacing, and making data valuable. You mentioned just the debate over breaking up IBM. I'm interested if you could take us through some of the - I bet you didn't rise every time on IBM today, did you? No, no, no. I want to talk about Intel and kind of the history of some of your takeaways and what you think you've gotten right in the past, your perception of should they break up the the foundry business, when what you think might be in the works with LipBouton coming in there, because I was listening to Dylan Patel talk about his conversation with the new CEO, LipBouton, and it seems like they're doing lots of tightening up, lots of layoffs, but it's kind of, I don't even know what framework to apply to analyze, like, is a breakup the correct thing?

2:13:28It feels like something people just say. Yeah. Yeah. So Intel, it's funny. One of my very first articles was about Intel. And what I said at the time was, and this was 2013, and this was an art like, you know, when you start a site like Strachetekery, you're like a new band. And why does everyone think a new band's first album is the best? Because they've been working on these songs for years, right? And then the next album, they had a year to do it and they all suck, right? So I'll let people decide if that applies to Strachetekery or not. I won't be offended. The sophomore slump. But yeah, but I'd have been on, you know, Intel had been a thing I've been wandering about for a long time, which was by 2013 when I started, they had clearly missed mobile.

2:14:13Now, it wasn't clear to them. They were still trying to do the Atom processor and just, they're going to figure it out tomorrow. And the problem with missing mobile is the problem with Intel in general is Intel is always very biased towards high performance. And this goes back to actually Pat Gelsinger, his first time through at Intel. Intel had the CISC, the way there's CISC versus RISC. It's like different ways of organizing bits or whatever. RISC is generally more efficient. And actually even Intel processors today, even though x86 is CISC, the internal, it's retranslated internally to a RISC type language.

2:14:55None of that is really important other than to say in the 80s there was a real push in Intel to switch away from x86 and to a risk type of architecture for the processors. And Gelsinger was a leading proponent that this is a terrible idea. And the reason it's a terrible idea is because there was already a huge ecosystem of software built around x86. and all this low-level code and capabilities that was written once and no one ever wants to touch again because it's miserable work. And he's like, to rewrite all that stuff would take at least two years. And in that time, our ability to manufacture chips will improve so much that had we just stuck with CISC, our processors would be faster.

2:15:48And that was the right bet, and that's one of those foundational bets that I like to think about companies and their history and what goes into that, which is Intel, from the 80s on, has solved its problems by having superior manufacturing and by moving faster. And yeah, our chips may be theoretically less efficient, but if our manufacturing is better and our transistors are smaller, it doesn't matter because that will swamp whatever theoretical sort of efficiency you might have. And this drove the entire computer industry. You would write to write a program. Just every second you spent optimizing your software in the 80s or 90s was a waste of time because whatever improvements you could get would be swamped by the next version of if you went from 286 to 386 or 386 to 486.

2:16:33that jump was so large you were better off focusing on features even if it made your software sort of slow to use on the current hardware because the next generation of hardware would be so much faster it would solve your your speed problems for you now this has generated a lot of bad habits amongst tech developers that's why you get bloat and why you have like poor performing things and all those sort of things but this was sort of super critical and so intel at its core has always been focused, they've always been manufacturing first and focused on better and better performance. What happened with mobile is in that calculation did not come efficiency.

2:17:08They were never focused on efficiency. And in mobile, efficiency was everything. So what happened with mobile is Apple went with an ARM processor made by Samsung, and they basically rewrote everything. All that stuff Intel didn't want to rewrite in the 80s, or if they rewrote, would just give other processor processor companies a chance to catch up with them had to be rewritten for mobile because efficiency was so much more important than performance when that happened intel was screwed now it took them a long long time to realize they were screwed but they were just fundamentally unsuited to be competitive it was the whole paul addolini turning down the iphone contract is not true uh tony fidel i i said that once and i got a call from tony fidel actually this is when i had him on it for an interview and he's like this drives me up the wall intel was not remotely competitive even though they had arm chips then even their arm chips then were focused on performance not on efficiency and and so the the problem for the problem for intel is once you missed mobile you were going to lose your manufacturing lead at some point because volume matters so much and every time you move down the curve your transitions get smaller the costs increase massively.

2:18:24So you need volume to spread out the cost of building these fabs. Like back then when I wrote this article, fabs cost 500 million. Now they cost like 20 billion. And this is over a course of like 12 years. So it was clear Intel was going to be in big trouble back then. And so I wrote, they need to build a foundry business. They need to figure out a way to build chips for other people because in the long run, the cost of keeping up in manufacturing is not going to be tenable if you're not making mobile chips. And what obviously they didn't, TSMC made all the mobile chips for everyone. And guess what happened?

2:19:02TSMC took over the manufacturing. Now, there's lots of other things that went into this, why Intel stumbled and sort of things. But at a structural level, what happened was actually inevitable. Once Intel missed mobile, unless they figured out a way to make mobile chips some other way, they didn't do that. what's interesting is what is the problem with that it took so long to manifest part of mobile was you had an explosion in the cloud because cloud and mobile actually go hand in hand intel made all those cloud chips intel stock had an incredible run from the time i wrote that article for the next eight to nine years and i felt like kind of a moron because i might say this company is screwed if they don't do what i say they didn't do what i say and their stock went to the moon.

2:19:47But the way it actually caught up to them has been in the past two to three years, where there's astronomical demand for AI chips. Only TSMC can meet it. Intel's not in the game. They're trying to shift to a foundry model, but they're so far behind. Being a foundry is being a customer service business. It's not being an Intel, we tell you what to do, or we tell our design teams how to change their chips to accommodate our manufacturing needs. It's just it's totally different and they needed a decade to learn how to do that. Had they changed in 2013 they would be ready today to capitalize on AI.

2:20:28And the counter example here is Microsoft. Microsoft building Azure. Yes it got them somewhat in the game with mobile and things like that but AWS dominates in that space. But by virtue of building up Azure, they were prepared when the AI opportunity came along. And now Azure is sort of a big AI player. And I wrote about these two examples a few weeks ago in the context of Apple. I think the concern for Apple isn't the short term. We're going to be using AI apps on our iPhones for quite a while. It's are they going to be prepared for what's next if they don't do some sort of reset and pivot here.

2:21:13Oh, sorry. I didn't answer your question about Intel. Anyhow. I mean, it's a... I'm hearing like, you know, manage decline, basically. Like, just like, you know, just get as much cash flow out of this thing as you can while you wind down the business. For Intel? Yeah, that's what I'm hearing. Yeah. Because it doesn't feel like, oh, yeah, there's a silver bullet. Just split the business and they're good. Like, no, it's like, it's all bad. The problem with the business is Intel needs volume and they get volume from Intel. Sure. And AMD split their business a decade ago, and it was really – they had a very hard time for many years, and they had very tense and difficult negotiations between the Global Foundry side and the AMD side.

2:21:51Google Foundries was AMD's manufacturing arm. And it wasn't until really they got out of that and went to TSMC and then also completely rehauled their ship design business and all those, you know, that they got in the business they were. And then also that Intel stumbled. That certainly really helped them. Intel today, so you split it up. Like Intel itself is fabbing some of its stuff with TSMC. Yeah. Who wants to buy Intel's foundry services? the problem here is TSMC is located in a country called Taiwan which you know what it is today but five years ago you'd be like what Thailand which by the way was probably much better for Taiwan security when the American style was Thailand so there's a real national security element here and it's a really tough situation because Intel is a failed company at this point And the reason the failure is so total is because the aspects that drive their failure are the same things that drove their success.

2:22:58It was their arrogance. It was their sense that we're the best, that we will just win through manufacturing might and performance. And all those things work against becoming a good foundry, work against being a customer service organization, work against recognizing the fact that you're not going to make up for missing mobile through manufacturing, which was their bet for years and years. You had to accept that you lost. And that's a tough place for companies. It's not like someone made a mistake. It's not what they did what they did too well for too long. It's who they were. They continue being who they were.

2:23:37right that's right but who else are you gonna get if you want an alternative to the smc it's it's it's a very last situation last question and i think we'll be forced to to have you make a slightly shorter answer unfortunately i wish we had hours to keep talking i wanted to get your updated thinking on xai x the combined entity the last 24 hours have been very chaotic when the initial merger was announced it made sense for financial reasons for some of the different stakeholders but But I wasn't fully sold on this idea. How strategic. You're going to force me to come with takes that I generally just avoid writing about Elon Musk companies for self-sanity reasons, I think.

2:24:17I mean, I wrote an article years ago about when the Model Y was announced. And I was talking about it's a Tesla and this aspect. What Elon Musk is very incredible at is sort of creating reality out of thin air. He's like the ultimate memer. and to create like, it's the way things used to work backwards. I remember I analogized it to like protests, like a critique of modern protests is they spin up very quickly because social media makes it very possible, but there's no infrastructure under them, so they don't amount to anything. Whereas you go back to like the civil rights era, there was years of groundwork that went into like the million man march, you know, on Washington DC.

2:25:02And there was a structure in place that ultimately manifested in large crowds. But modern protests are the opposite. The largeness comes at the beginning, and then it all falls apart because there's nothing in place. And that makes a challenge to write about anything Elon Musk related, is you have all the social aspects. You have this bit about Tesla of creating reality. The stock was buttressed for years by these true believers, even though the financial parts didn't make sense. You famously had these wars with the short sellers and all that sort of thing, and it worked. It basically manifested a market for this Model Y and then the Model X.

2:25:42Not the Model X. What's the other one? Model three. It was Model three. Sorry when I wrote that article. Model three and Model Y had this massively successful, and all the people that were true believers got very rich, and congratulations to them. It's great. but it makes it almost impossible for someone for what I do who I want to look at structure and fundamentals I can observe this effect happening but you can't really say what's going to happen or the effects of it other than to say this is interesting and so I wrote about that article and then the solar city thing came out he's like bailing out like his brother-in-law or something and I'm like I can't write about this like what am I gonna say like like there's it just doesn't make sense and so i think there's to fast forward x xai um yeah there's a theoretical piece here i think actually xai would be an incredible acquisition target for a lot of companies if it wasn't saddled with x uh so yeah it feels like the end state is like twitter getting spun out again like that that that's my i that's kind of like my my it just ends up going back to Twitter and it becomes...

2:26:51The bluebird. No one actually wants to. Twitter, there's never been a company in the history of the world probably where the impact of a company is completely and utterly divorced from its financial realities. I think when Elon Musk bought it, and I assume that's continued through now, they'd have like one profitable quarter in their history. It's an unbelievably terrible business. And so I think it's probably weighing XAI down. I guess I get the theory that Twitter data helps XAI. Well, it helped yesterday. You don't need to pay$43 million for Twitter, or$43 billion, I should say, to get it. Yeah, that was always my position, too.

2:27:30I don't think it helped yesterday when Mecca Hitler emerged on the timeline. Crazy day on the timeline yesterday. But anyways, I wish we had a lot more time here, but hopefully we can do it again soon. Thank you so much for stopping by. Yeah, no worries. What would you guys are doing? I actually had the idea of doing a daily podcast ages ago. Classic example of ideas don't count. Execution does. And you guys did it. I think it's great. Well, you're always welcome here. You're always welcome. Thanks so much. Thank you. We'll talk to you soon. Bye. Meta just is going deeper with Ray-Ban maker Eslior Luxottica.

2:28:05I cannot pronounce that first word, but people just call it Luxottica. And so Meta is taking a minority stake in Luxottica to accelerate its smart glasses ambitions investing 3.5 billion dollars in the iconic ray-ban manufacturer uh we were talking to david center about the the history of this company it is fascinating i'm very excited for him to uh break it down for us a little bit more hopefully he can come on the show and talk about it because it's very very soon the founder has a crazy story grew up in i think he grew up in an orphanage yep and uh and and it just what are they wasn't they didn't call him the pit bull they They call them something else.

2:28:43But he was an absolute savage. Apparently, at one point, he wanted to buy Oakley. And the founder, CEO of Oakley, didn't want to sell. And so the CEO of Luxottica acquired the largest retailer for Oakley's and just pulled them off the shelf. And basically, he had started selling knockoff Oakley's, even though they were trademarked. And then eventually, the Oakley CEO came around and said, okay, I'll sell your cratering my revenue. Yeah, you're just going to tell me. Let's do a deal. Wow. So, absolute dog. And soon, we'll have Senator to break it down. What do you make of this idea that, like, you know, Apple, when they make a device, they redefine and very much standardize that particular market.

2:29:29So, when they come out with watches, there are a number of styles of watch. There's the dress watch, the sports watch, the steel sports watch. There's the dive watch. there's the you know casio style there's a whole bunch of different styles right apple comes in and just says there's only one style the apple watch and they become the number one apple style and they give you some variants in the band in the band little stuff here and they were doing partnerships i think they did an hermes band for a while they've done a couple other things but it's been mostly apple's design language on your wrist whereas with the meta Ray-Bans, they're saying, and now the meta Oakleys, they're saying, you like the look of Ray-Bans.

2:30:10We're just putting our technology into the style you like. We're not going to try and create a new iconic style that says meta like Apple says headphones. And they're just kind of like, they're very, very different strategies. And so it feels like... Well, so I think this is strategic. This doesn't mean that meta can't develop their own styles in time but i think it's very smart to say hey we don't need to innovate on aesthetics and the sort of silhouettes right there's classic silhouettes ray-ban silhouette is lindy these oakley silhouettes are very lindy yeah and they're different markets the way they're different markets luxatica has i think garrett late and like a bunch of other like um brands under it so they're basically saying like through this we can deliver luxatica has brands in every for every demo that that Meta could possibly want, right, as a$100 billion company.

2:31:06And so I think it's very smart. I think Apple, like you said, will probably take a drastically different approach in terms of standardizing around something. And that will say something. But accessories like I wear are just such a personal decision and such an expression of who somebody is that I think that you want to give people the max amount of optionality. Yeah, it's just interesting because you could have said that about watches. Like before the Apple Watch, you could have said that, well, you know, somebody who wears a dress watch wants a dress watch. Somebody who wants a steel sports watch.

2:31:39Somebody who wants a G-Shock is a G-Shock. It's like the G-Shock. You say G-Shock and you just immediately think like, you know, special operations guy or Jocko Willink listener. Like that, it's like a durable, rugged thing. You say, you know, Rolex. That's a different thing, right? And Apple was able to standardize around it. And it's interesting that Meta hasn't been trying to do that. And instead, they're focusing on partnership here. It's just like, it's just an uncommon strategy, but it seems to be working. There's another post in here. I don't know if we have it here, but someone was talking about it.

2:32:11I'm trying to think of a new, like the key thing is Apple's great at innovating at multiple layers, but like generally it's very hard to try to deliver hits in like two specific areas, like aesthetics and design. and then simultaneously in something that's basically a fashion product and then simultaneously deliver the technology. So I don't know. Yeah. Jack Ray here says, after wearing Ray-Ban Meta Wayfarer glasses for a few weeks, I feel kind of naked wearing regular sunglasses. I found three use cases that are hard to roll back. One, spontaneous photos of my kids when we're out and about. Any cool pose that has a half-life of three seconds I can now capture instead of pulling out your phone.

2:32:54Optionality of music or hands-free phone calls without digging around for earbuds. And three, knowledge-seeking chat when I'm walking around, usually for simple factual things. That's exactly what I experienced when I was demoing the Ray-Ban meta Wayfarers. It turns out there's more questions I feel like asking when there's no friction. I'm very excited for multimodal and real-time translation use cases too. They're only going to get better. But I think those three are maybe enough. And I think with a lot of these products, Just having one killer use case, like just replacing the headphones for hands-free phone calls or something.

2:33:33Like if you can just become someone's daily solution for music, like if that's enough to just sell the product and then sell them another one the next year when it upgrades a little bit, sell them another one, keep them as an active user and roll that out for a long time. And then if they can do the other stuff, that's great too. But you just need to get these one, nail their single use case. And so yeah, there's gonna be cool stuff, but it's fascinating to see them roll this out. And it's also interesting how behind the ball it feels like everyone else is now. Like Google was talking about getting into this space.

2:34:02We saw some launches at I.O. Haven't actually seen any of those in the wild. Haven't seen anyone really talking about those. Apple, it feels like this would be something that they could jump forward to with a stylish pair of eyeglasses with some basic functionality. Just take what's in the AirPods, take a camera. They could do something cool, um but they're like just much slower than than yeah the other the other thing with eyewear that's different or that's going to be like a new challenge for manufacturers is that there's so many different situations where i might want to wear something like a ray-ban or or a jm silhouette one day and then i might want to i'm playing jack marie mage okay um but um the uh you know, and then that same afternoon, I'm wearing Oakley's when I'm playing tennis or something like that.

2:34:53And so there's a lot more like swapping and then obviously something like mobile device. I mean, if they can keep the price low, you could maybe wind up selling people multiple pairs and have indoor pair, outdoor pair. It's kind of inconvenient. I feel like there's gotta be a better solution to that, but I don't know. It says, what are those? Yeah, the bifocals. Yeah, where they can like flip down. Well, there's transition lenses, but those never fully work. all the way, but then there's the flip down ones, clip-ons. There's all sorts of different solutions. The big news is that the third browser war has begun.

2:35:27Google stock has dropped on the news that OpenAI is planning to launch a Google Chrome competitor within just weeks. And this is very interesting timing because it's time to browse. Yeah. Time to browse. Certainly makes sense to become deeper and more deeply integrated into the user's life. Makes a ton of sense. There's a ton of benefits that come from having a web browser. What was interesting is we can go into what Google actually launched, or what OpenAI is talking about launching, but this news, this scoop leaked the same day that Arvind from Perplexity announced that they're finally releasing their next big product after launching Perplexity, Comet, the browser that's designed to be your thought partner and assistant for every aspect of your digital life, work and personal.

2:36:14And so Perplexity launched this on june 9th and then open ai the the scoop goes out via reuters the same day and so this feels like very much like let's not let perplexity get a bunch of attention and drive a bunch of people to start daily driving comet the browser because even though we're not ready to launch our competitor we want i mean arvin was on the show talking about comet but over a month ago he said it was really important to the business this was a big bet that they're making yeah uh, he, uh, and I'm sure both companies are racing to be the first to launch, but D, uh, the browser from the browser company, uh, also launched out of bit or they're still in beta, but they launched like a month ago or something like that.

2:36:58So this is, you know, you're not going to be the first. They launched a month ago with the D browser. It's interesting because I saw Riley Brown also posted the cursor for web browsing D browser. And I thought D browser launched the same day, but I guess it had launched earlier. Um, yeah. So anybody that was an arc user can download D today. today and chat with their tabs. But interestingly enough, Perplexity's browser and OpenAID's browser are both built on Chromium, the same open source project that underpins Google Chrome and Microsoft Edge. So the cool thing here, that means that they're compatible with existing Chrome extensions.

2:37:35Oh, interesting. Okay. That's cool. Yeah. I want to talk to more people who were like active in tech during the earlier browser wars. The first browser war was Netscape Navigator versus Microsoft Internet Explorer. This is in the mid-90s, early 2000s. Netscape was super dominant and everyone loved Netscape. It was originally the Mosaic browser. This is the Marc Andreessen project. But Microsoft bundled Internet Explorer with Windows 95 and the distribution was so powerful that Internet Explorer actually wound up winning and became really, really dominant but then there was this lawsuit and went back and forth but then uh basically in by the early 2000s internet explorer had over 90 market share but then they got kind of lazy and stagnant apparently and i mean i'm not exactly sure what happened but there's a lot more competition so firefox which was i believe like a spin out of netscape or kind of like some of the same heritage there um began getting traction and then google chrome launched in 2008 and leaprogged everyone and Google Chrome was really focused on like speed.

2:38:42It was the fastest browser. And they did a whole bunch of work to optimize JavaScript. So the pages would just load faster and run better on pretty much every computer that you had. And so, and then they had the open source project with Chromium. And so they were able to kind of standardize the entire industry. And so everyone's always been trying to draw analogies between like the browser wars and the LLM wars and like, what's the role of open source in that? Like is open source a strategy to wind up maintaining your dominance? How much does distribution matter? Like Chrome was probably pretty easy to distribute because every single person was visiting Google just every day searching.

2:39:20And so you just put this bar, hey, wanna switch to the faster browser? And people just do it because you can have basically like, you know, billions of ad impressions on your product every day. Will be interesting to see if ChatGPT can get people to download their own browser on desktop. I mean, I'm using ChatGPT on desktop in Chrome all the time. Which ChatGPT model would you want to use as a default search engine? That's the hard part because I always run into this problem where it defaults to O3 Pro, but that takes 10 minutes. And so then I have to go to 4.0. And then if I'm in an O3 Pro flow and I'm talking to O3 Pro and I let it cook for 10 minutes, it gave me a great answer.

2:40:00But then I want to just be like, okay, just like clean this up a little bit or summarize this or do some bullet points. I want 4.0 to do that. so I have to switch over. So I don't know. I would imagine I'd go 4.0 as the default because I want speed. But even 4.0 could probably be faster before it truly replaced. Google is very fast. They've spent a very long time being fast. Yeah. And I could imagine them doing a similar project to, I believe it was like the V8 JavaScript engine. They sent this team out to, I want to say like Iceland or something. They basically sent like a bunch of engineers to like an offsite and they were like just go optimize javascript for like a month just go focus on this for like a month or months and come back when it's done like you have no other responsibilities than just like optimizing this like compiler and they came out became back with the va javascript engine it created this whole like node.js boom people were running javascript on the server then and uh and i could see google kind of doing something similar where they're like okay we have gemini it's good at looking stuff up it's a good knowledge retrieval engine go figure out how to make it load all the tokens for the full response in 100 milliseconds and that would be very very cool and i wonder if that's like a uniquely google advantage tyler you look something up yeah it was in it was in denmark denmark okay i was close i was close yeah i wasn't sure it was finland or iceland in denmark the interesting thing here i'm realizing that tabs are definitely a light lock-in to browsers it's not it's not just the default but if you have six to 10 tabs that you've just had open for a really long time.

2:41:37And they're like, from a bunch of different things. And you can't exactly remember what they were if you had to list them all off. But you know, you know, I personally end up using tabs as like, somewhat of a to do list. And so if you're spinning up a new browser, and you don't have your tabs, it's like, Oh, do I want to just like, get rid of my tab stack, I have a bunch of tabs that just have stayed there for years. And they're basically like, it's basically like a mini operating system, right? With like different apps that might be a Google sheet or something else. Yeah, I know what you mean.

2:42:08So there's very real lock-in. I could bring all those tabs over, but I have to then log in to a bunch of different services. And so it's really, really hard to actually win here. I wonder if anyone's using, you know, in Google Chrome, you can actually change the default search bar to, you know, when you type in the search bar and if you just type words, it just Google searches it. You can change that to search chat GPT. Yeah. You can pass in a query parameter and it can just do that. But I haven't heard of anyone actually doing that. And I used to have, I used to be such a power user of Chrome.

2:42:40I used to have different code words basically. So if I typed like I space and then a query, it would go to IMDB and search that specifically so you could you could have chrome like route to any specific search any that's cool you could press like y space and it would search yelp or you know anything else um but i don't know if people are i don't know if people are doing that with google with chat gpt i think people mostly just like control command t and then yeah hang out and chat gpt well we'll have to ask uh chris in 15 minutes about uh get an update on the browser wars because he was uh an early investor in i know one of those tabs that you have pinned right now what's that adio of course customer relationship magic adio is the ai native crm that builds scales and grows your company the next level you can get started for free i've had adio open for thousands of hours at this point yeah uh so signal kind of breaks it down with the open ai launching the web browser says this is the oldest play in tech find product market fit with a single killer use case then vertically integrate and horizontally expand until you control the interface layer itself app platform once you own the interface you own the defaults welcome to the next generation of browser wars yeah what's interesting is they're like sam altman at open ai and just the fact that open ai is a company like there is kind of a mandate to like vertically and horizontally integrate figure out code figure out research figure out devices but every company wants to do everything but then sometimes they run up against barriers like there was a time when google was like, we want to win social networking and we want to beat Facebook and we're going to launch a direct Facebook competitor.

2:44:23And they did and it didn't go well. And then they shelved it and then they wound up producing trillions of dollars in market cap, just doing the thing that they do great. And so the question is like the surface area of open AI, they have to explore, they have to experiment. It would be stupid not to see if they could get a browser and a device and a chip and a nuclear reactor and everything and sand. Get the sand, get everything. But there's no guarantee that they will win the entire vertical stack and they will be the one company, right? I think my question is, are these going to be, like is OpenAI's browser going to be an entirely new app other than their existing mobile app?

2:45:04Or their desktop app? Yeah, that is interesting. Because if they have to get people to re-download a separate app, then that's like an entirely, you know, they have a good flight, you know, they have a bunch of impressions. wouldn't just evolve the apps that they already have installed perplexity too i don't why are we doing new apps he has uh is planning to to release this as like a new standalone app or it will be in the perplexity mobile app but yeah um yeah i mean i know i think comments like its own thing because we were looking to download it and we need a code um and you can't just get it if you're just on perplexity um but i don't know all i know is that you should go to 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 so Arvin breaks down like his philosophy of Of comment the browser that he's dropping from perplexity He says you can either keep waiting for connectors and MCP servers for bringing in context from third-party apps Or you can just download and use comment and let the agent take care of browsing your tabs and pulling relevant info It's a much cleaner way to make agents work So that is interesting.

2:46:16So I wonder how much like puppeteering will be in this because chat GPT and open AI have operator that operates a chromium front, like a headless web browser basically. But you can actually see it working and it's clicking things. And so if they're like, there's also the value of like the training data. If you're getting people using all these websites, you have all this training data of like, okay, they clicked on the blue button. They clicked on the green button. They saw this. They entered that. This is how they dealt with this form. This is how they dealt with that form. And so that feels like very, very valuable data if you can get it.

2:46:50So it's probably worth duking it out even if it doesn't, even if it takes a long time. For sure. I do wonder where else they will, where they will plug in. Like, clearly operates at like a higher level of abstraction with like the screen scraping. and I wonder if we'll hear rumbles about either perplexity or open AI thinking about like moving up the stack to that level. I'm not exactly sure.