Loading summary
Transcript106 lines
- [00:02]
A
We had to rush back from Striped Sessions. Fantastic interview with the Collison brothers. Many others yesterday. Go check it out if you haven't already. Fun stuff. We didn't go nearly deep enough on the whole cosian singularity. There's a debate on the timeline about it. We can get into it later. We need to have some more economists on the show to break it down for us. But this is their big pitch. Firms get smaller and the future the coast theory of the firm. Anyway, if you studied economics, you might be familiar, but it's jargon and you're going to be hearing a lot more of it. Singular.
- [00:32]
B
A lot of mumbo jumbo.
- [00:35]
A
I'm not even gonna try. I'm not even gonna try with Jordy. Well, here's some less mumbo jumbo. Tech earnings quad kill recap four in one day. Last time this happened, I think it was 2024 big tech companies all same day. It is random. No one really knows why. It just sort of lines up. It's like a solar eclipse.
- [00:55]
B
Can you use astrology to predict the next one?
- [00:57]
A
Potentially. Potentially.
- [00:59]
B
You would have to imagine it's something around the alignment of the planets that cause.
- [01:04]
A
More like the holidays. More like the holidays. Because books close, then certain things line up also like Nvidia is never in the competition. They're always like two or three weeks later. But anyway, let's go through what actually happened. Google was the big winner. They absolutely crushed the stocks. Up 10% in the last couple days. Really successful. And there's a few reasons. So the core business is still growing. Google Search. This was the question of is AI? Are chatbots going to be eating into the core Google Search revenue, the core business that drives all the other investments? The answer is unequivocally no. Google Search and other revenue was 6.4 billion. Yeah, up 19% year over year. I mean, two years ago there was the whole like Google search is dead thesis that has not been borne out, at least this year. Very, very strong results there. Then Google Cloud is like the major story. So much so that. Take him when he did his earnings recap. He didn't even talk about top lines at Google, Amazon and Microsoft. He just says Google Cloud revenue 63% year over year. Amazon Web Services 28% year over year. Azure 40% year over year. And then meta overall revenue, which we'll get into 33% year over year. So he's not even looking at the rest of the business. All that matters for the three major public clouds are those cloud businesses. Not even Anything specific within the overall business. So long story short, higher capex@google totally justified in a world of cloud acceleration. Backlog growth is huge. Cloud backlog nearly doubled to more than $460 billion. And importantly this isn't like a 5 year, 10 year contract type thing. Half of that, more than half of that backlog is expected to be recognized in the next 24 months. So really good news there. Google Cloud hit 20 billion up 63%. Now it is smaller than Amazon and that's why AWS is people hunting for high 20 growth rates. Maybe begins with 3 in Google it's smaller so it can grow 63% but that's still fantastic. And cloud operating income was 6.6 billion which is very strong margin. So everything is looking good. Search durability, demand for AI infrastructure, it's the full stack AI play right now. Microsoft showed solid execution but they didn't dramatically change any particular narrative. I think a lot of this goes to the just naturally slow deployment of enterprise AI. But this is the focus for Microsoft. They are the enterprise AI play.
- [03:36]
B
Yeah, one thing was notable is they weren't heavily teasing anything related to the next Gemini release.
- [03:44]
A
Yeah, they wouldn't tease that at earnings.
- [03:46]
B
No, no, they have, they have historically. So maybe they want to let the model speak for itself.
- [03:52]
A
Yeah, probably announced Gemini 4 I would imagine. Announced it or a coding specific model. Yeah, yeah, something there. I mean there's certainly demand for it. They have the capacity, there's lots of opportunity. We'll probably be following that in three weeks when Google I O happens. So Microsoft investors wanted to see Azure growth copilot adoption and a solid justification for all the capex the stock's down 2%. Nothing crazy but Microsoft beat on headline numbers. Revenue was 82.9 billion, up 18% year over year and from an analyst perspective it was a clean beat. But there's more nuance here. So in terms of AI adoption, Microsoft
- [04:36]
B
added about this beautiful LED wall million. Yeah, let's give it up for the still underrated. We ordered this bad boy maybe like six months ago but it is fun.
- [04:47]
A
We need to figure out more things to do with it. On copilot adoption this is the big question for Microsoft. 365 like total seats. You know, you think about Microsoft Teams Outlook like the standard Microsoft enterprise seat. I think it's about $30 per user per month, competes with Google Apps enterprise. Right, but Microsoft's been doing this for much longer. They have 450 million paid seats. So in terms of copilot they added 5 million, which is great. They're at 20 million now, so that's solid adoption, but it's still small compared to the 450 paid seats. Ideally, like, every seat would have a copilot alongside of it. I mean, that's the idea. Right. And so the question is, how quickly can they get that 20 million copilot seats up to 200 million, 400 million. Ideally, everyone has one of these, add on subscriptions, and that's lifting. The overall business market reaction was a bit choppy as the new OpenAI relationship gets digested. And so there's two sides to the Microsoft OpenAI deal right now. Of course, OpenAI is no longer limited to selling just through Azure. Azure no longer has exclusive access to OpenAI models. OpenAI is now available on AWS as well. And it cuts both ways. So on one hand, Microsoft, they used to be the exclusive provider. If you're a sales guy at Microsoft on Azure, you're probably really excited to call somebody up and say, hey, you want to use GPT 5.5? We're the only place in town. Like, you should come over here. You should migrate more of your infrastructure. You should be using Azure. But it was probably frustrating because a lot of people say, no, we're locked into aws. We are stuck in that cloud and we'd love to, but we'll just use the API directly or we'll do, we'll figure out some other workaround, but we're not moving for this.
- [06:32]
B
Yeah. Or people could be like, yeah, we're open to trying it, but it'll take a couple quarters realistically to get ramped up.
- [06:40]
A
Yeah. And so from an Azure perspective, losing exclusivity is a negative. But from a business perspective, the More growth for OpenAI is good for Microsoft's equity stake in the company. And so these two tensions, like the, the capital balance sheet side of the business versus the Azure sales and acceleration side of the business, like these two,
- [07:02]
B
not to mention Copilot versus Codex.
- [07:05]
A
Sure. Yeah. And so these two, these two sides of the business are sort of butting up against each other. And Satya Nadella ultimately has to go and un on, you know, renegotiate the contract. And they did. And it seems like they're in a good place because a lot of people were sort of saying like, oh, well, is this going to get messy? Is there some going to be some lawsuit here? It seems like there was a very clean negotiation renegotiation. And they OpenAI got rid of the AGI clause that allows them to stop sharing models, but they are sharing revenue.
- [07:33]
B
Basically The I think 2032 end date, yeah, is just the IP share instead of this like arbitrary AGI.
- [07:42]
A
So over at Amazon, Amazon still of course the CapEx king. They top the CapEx forecast. But as scary as that would be, AWS is re accelerating and so that's very good news. So the stock moved slightly upwards. Q1 2026 sales for Amazon overall were 1 81.5 billion for the quarter. For the quarter. They're like on track to make a trillion in revenue soon. Up 17% year over year, beating expectations. AWS is expected to see 25 growth. That was the expectation, 25%. That's sort of where they've been. They came in at 28% so they beat on AWS growth and that's really, really important, especially for their scale. This was very good news. The ads business is still cooking, churned out $17.2 billion in revenue and the chips business just crossed 20 billion run rate, which is very, very good when they have big deals. AWS is in this interesting position. Less of this like Google Pure Play Full stack and more focused on working with both OpenAI and Anthropic at this point. And so that all justifies more investments in Trainium and AI infrastructure broadly. And the market seems to like it. Huge capex numbers but reasonable response from the market and the stock. So Meta Meta had a rough go. Huge drop, nearly 10% down following earnings, sort of jumping up and down now. But the core business was extremely strong. So Q1 revenue was 56.3 billion, up 33% year over year. So high growth for the overall business. Ad impressions rose 19%, average price per ad rose 12%. They raised the CapEx outlook from they have this range, it's within $10 billion of 125. They added 10 billion to both. So instead of saying, hey, we'll be between 115 and 135, now they're saying we're going to be between 125 and 145. Will they actually get more compute for that or is that just a reflection of rising prices for all the inputs that go into compute spending? That's a big debate because there's one bull case where, oh, they're optimistic they're going to buy more compute, but it's like, no, there's a possibility they're just spending more to get the same amount of compute as they thought. But the big problem for shareholders, the big answer is it's not as clean of a Can you imagine?
- [09:54]
B
Take one second and imagine that Meta is a private company and their pitch is that we make AI agents for selling products. They're at a $200 billion run rate, growing 33% a year. How's it price? 20 trillion probably with those operating margins.
- [10:13]
A
Yeah.
- [10:13]
B
Joking. It truly is one of the most incredible businesses of all time. But they don't have the cloud people. Yeah, yeah, they don't have, they don't have the cloud thing. He's kind of acting like the most AGI pilled CEO. Like hyperscaler CEO, right? Yeah, he was talking about like basically talking about like RSI on the earnings call. He's like, it's not just about like code, you know, coding. Like he really seemingly believes and he doesn't want to be in a position where someone else has actual super intelligence and he does not. And so, and that's just like a little bit too fuzzy because when you look in the rest of the market right now, it's like productivity tools.
- [10:55]
A
Yeah.
- [10:55]
B
Right. And so the market just like can't really read into it.
- [10:59]
A
Yeah.
- [10:59]
B
And they're just kind of like not even giving him full credit for the business that he has.
- [11:03]
A
Yeah. And I forget if he's ever mentioned the idea of, of selling tokens, selling AI infrastructure, getting in the cloud market, but it's way easier for an aws.
- [11:14]
B
He kind of loosely mentioned that last year when he was like starting to ramp stuff up. Like basically saying like, maybe we could do this at some point.
- [11:23]
A
Yeah. Because it's way easier to justify the capex at Amazon when you say, oh well, like, you know, if, you know, regardless of how this market plays out, maybe Anthropic takes some of the compute. Maybe OpenAI takes the compute. Maybe there's a whole bunch of new AI labs that take compute. Maybe we'll use it for our services, our own models. If models commoditize, it's still valuable. If they don't, it's still valuable. There's a whole bunch of different ways to, to shift that capex around and justify it. Meta. It's got to show up in the AI figures. It's got to show up in the ad business, at least right now. And so there's are questions about how the big capex spend feeds back into cash flow when the company doesn't have a platform. With big enterprise AI contracts, massive RPO figures, there's not an easy place to just stuff compute necessarily and at least get a market standard return on investment. Of course they could spin up reselling they could become a Neo cloud or something like that. But at their scale, it's a whole new host of challenges if you, if they go that direction. And I think that that's not the plan. I think you're dialing into it that they are in fact recursive self improvement pilled at this point. The other sort of tremor in the system was the fact that daily active people, which is something that they were. It's not users anymore, it's dap. Daily active people declined sequentially since the first time that meta began reporting the metrics. So less people were using meta platforms this quarter than previously. But they had some good explanations for it. They said that there were Internet disruptions in Iran and WhatsApp restrictions in Russia. And so this was not representative of a real change in Internet user behavior. And I think we all recognize that that is reality in the sense that there's not some up and coming social network that's eating Meta platform's family of apps. Lunch, right? Like the TikTok threat came and sort of went. The Snapchat threat came and sort of went.
- [13:23]
B
Sora.
- [13:23]
A
Sora came and went. And so there was a different post. I saw that social media use is declining, especially under young people, but it was already so high it's not really showing.
- [13:37]
B
It will truly be insane if we get this new alien technology, we get LLMs which in digital form can convince you that they are a person and we end up getting no new big
- [13:50]
A
social platforms might just strengthen the current ones.
- [13:52]
B
Well, yeah, no, I mean, I think that's the base case right now, but it's still wild. I think you would expect, I think again, if you rewound to like 2020 and you said, hey, we're gonna have this technology where you can effectively have a digital twin of yourself, you would think that somebody would figure out some new.
- [14:07]
A
Yeah, maybe. I don't know. I continue to think it just strengthens the incumbents. Roblox, Instagram, WhatsApp, Facebook. Like these platforms will just get more content on them that will probably be LLM generated, LLM enhanced. Even if it's not LLM written or developed. It's, it's, you know, a creator taking less time doing other boring things and spending more time on their core output. So we'll keep tracking it. So overall, what did we learn this earnings season? One possible takeaway is that there's a little bit of a fracturing in the overall AI narrative. So the AI narrative over the last 12 months has basically been like, say the biggest number, biggest Capex number, biggest deals just grow, grow, grow. Now every hyperscaler reports a big capex number but each company does have a different story. So Google's the full stack platform. Microsoft's focused on enterprise adoption and distribution. Amazon's most aggressive on infrastructure and partnerships with OpenAI and Anthropic. They have the cleanest less. It's a less competitive relationship with Anthropic and OpenAI because they don't have the DeepMind equivalent necessarily. Meta is all about ad optimization. And Meta is also this interesting like high risk option on Frontier AI. Like it's possible that if TBD Labs, MSL, if that pays off, you wind up getting an OpenAI sized or anthropic sized business out of it. In the oligopoly world where they have a breakthrough, they get to something that's 5, 5 or 4, 7 level and then they're competing in an equal level and they're playing in that oligopoly. And so you like the metatraders are sort of or the investors are sort of weighing both of those sides. You have a very solid ads business and then you have this call option on potential Frontier AI which is still early but showing some really solid signs. So the market likes short term revenue evidence right now and there's still many strategies that will play out over the rest of the year. But this quarter was focused on can you justify your CapEx this quarter. And so we'll keep following it. Let's go over the timeline and see what people were saying. The headline numbers of course were huge. Cloud growth AWS 28%, GCP 63%, Azure at 40%. Absolutely massive numbers. Remember people were hoping for GCP to be at 40% and it went up 63%. Absolutely massive. AWS people are hoping for 30, they got 28. Not too bad. This poster ren is comparing the price to earnings multiples today versus the dot com boom. So today's mag4pe ratios, the companies that reported earnings meta's at 16x price to earnings, Google's at 17x, Amazon's at 24 and Microsoft's at 25 price to earnings ratio. Compare that to during the dot com Microsoft is at 73x, Cisco is at 200x plus, Yahoo was trading at 800x earnings. The Nasdaq as a whole traded at 200 times earnings during the peak of the dot com bubble. Today's bubble is trading at 16-25x earnings on companies generating hundreds of billions of dollars in real cash flow. Now they're drawing down on that Cash flow. But even in some bizarre world where AI is a bubble and it's fake and we all go back to pencil and paper, we'll probably go back to digital advertising and spreadsheets in the cloud and SaaS and all of these companies will still benefit and be in fantastic positions.
- [17:42]
C
Yeah.
- [17:42]
B
The only thing that, or one thing that's imperfect with these comparisons is when you look at the peak, you can say, okay, Meta, Google, Amazon and Microsoft have relatively reasonable PE ratios. But if you look at the rest, you look at all the companies that have been started in the last like five years, especially in the private markets that have been marked up from 20 to 100 to 500 to a billion and beyond, many of them have absolutely no earnings. And so you have to look at like maybe something more like a Yahoo as like a decent comp where Yahoo was trading a peak at 800x and
- [18:20]
A
so, and there is some, there is some pushback in the, in the replies here, Reasonous Forces. AMD is at 130, Tesla's at 350, Palantir is at 220 and intel is at 900 times earnings.
- [18:34]
B
I know. So it's like very, very convenient to just take four of the greatest companies ever in the history of the world and say there's no bubble, they're reasonably priced.
- [18:44]
A
Yeah. I think it is useful to go back to the fact that there's a lot of hype in tech, but what's driving the vast majority of the market cap right now is still fairly low. PE companies like intel is so much smaller on a market cap basis than Microsoft, Amazon, Google, Meta that even if you're overpaying on earnings for that, the overall market is anchored to a pretty decent earnings engine still. And that's like the bull case here.
- [19:16]
B
George Hotz is back with a white pill.
- [19:19]
A
What'd he say?
- [19:20]
B
May 1st, he says AI will create jobs. It's nice to see Jensen talk about this. And it's super obvious when you think about it. AI and immigration are fundamentally the same. There's new people showing up and hopeful. Hopefully everyone understands how and why immigration creates jobs. Wants are effectively unlimited. It's classic Jevons paradox that if we make something more efficient, we end up using more of it. Or a cool aphorism I learned at Facebook if you make the site 10% faster, people spend 5% more total time on it.
- [19:47]
A
He worked at Facebook for. He was like an intern.
- [19:49]
B
He's part of the machine now. Just like you get the wawa crying people about how immigration lowers wages for native born Americans and we got to keep the hard working immigrants out because you have some right to be lazy or something. You'll get this about AI. AI will out compete some humans at some jobs. But protectionism is for losers. The important thing is that the overall pie grows and inequality stays somewhat in check. Not by redistribution, but by design. There will be more to do than ever before.
- [20:16]
A
More to do than ever before.
- [20:18]
B
Can we pull up this video of Jensen? Yeah, let's do it in the production chat for you.
- [20:24]
A
Yo, you have it?
- [20:25]
B
Yeah, we can play it.
- [20:26]
A
Great. I'm excited for Eric Soufert's deep dive here. We'll get into that later.
- [20:30]
B
The Soufinator
- [20:33]
A
and Reminder. We have the Isaacinator coming on in just seven minutes. Mike Isaac has been covering the Elon Musk versus trial.
- [20:43]
B
Is it like a four day work week with trials?
- [20:45]
A
It is.
- [20:45]
B
Are they living in the future?
- [20:46]
A
It is. Wait, wait wait. So can we pull up this screenshot? Because did you read Mike Isaac's full thread on his experience covering the trial? Did you read the whole thing? Because he's been live tweeting it and it's very funny because he will post. So we gotta zoom in here. This is a.
- [21:05]
B
Did you turn this into.
- [21:06]
A
I turned it into an infographic. So this is a minute by minute chronological log of off topic slash personal, minor inconvenience moments. Because Mike Isaac will talk about. Oh, Elon Musk just said this on the stage. Oh, the lawyer fired back with this. The jury said this. The judge said that. Right. But then in between that he has been logging very, very minor inconveniences that he's experienced. So he talks about his lunch, the roasted tomato and corn pizza. He says he forgot a pillow for his sit on. Wait, he said corn pizza? I don't know. He said he bought water.
- [21:47]
B
That's a very Oakland thing.
- [21:48]
A
At 16:07:45. This is transcribed from X. I think the times are UTC or something. He brought a water bottle but forgot to fill it in the rush for a seat. Now he's increasingly thirsty. During an intense moment of testimony, someone's laptop suddenly blasted a loud YouTube video in the gallery. This is a garden hose nozzle. It said courtroom mishap. So he's going through it all. It's very fun to track his and we'll dig into his experience both in the court and actually with what's going on with the case. But let's play this Jensen clip. This sounds fun.
- [22:23]
C
The things that are said are very counterproductive. And in fact, hurtful. On the one hand, maybe I don't
- [22:30]
B
know who he's talking about.
- [22:30]
C
Maybe a scientist thinks that by warning people that AI is going to completely permeate and proliferate across radiology and therefore radiologists are going to get wiped out. On the one hand, that might be considered warning and therefore helpful, but in fact the counter would have been hurtful if we convinced everybody not to be radiologists and we now need radiologists. That actually is hurtful to society. It is hurtful if we convince all the young college graduates to not be software engineers and it turns out the United States needs more software engineers than ever. That's hurtful. And so we have to be mindful of how we communicate the importance of this technology and what it's able to do to advocate for policy and advocate for guardrails on the one hand. On the other hand, scaring people with things like saying nonsensical things which are not going to happen, that this is an existential threat to humanity. There's 20% chance that is existential. That's ridiculous. That it's going to wipe out 50% of new college grad jobs. That is going to completely destroy democracy. I mean, these kind of comments are not helpful. They're not based, they're made by.
- [23:50]
A
He's subtweeting in real life. He's not vague posting, he's subtweeting. I do think the 50% numbers are so funny because it's like 5050 is the most neutral you can be about some prediction. It feels like, I mean, I get that it's not. In that case, it's like 50% of jobs, but whenever you issue like a 50, 50 proclamation, it sort of tells you nothing. Like we were looking at like the prediction markets on the trial and we were like, oh, it's 50 50. Okay. So that actually doesn't give me any information about what's going to happen in the future. And it's very easy to make a prediction that's like, oh, I predict this will happen, 50, 50. And then if it doesn't happen, you can be like, well, yeah, 50% chance it wasn't going to happen. Like, I had it really high that it wasn't going to happen. And then you can also say, you know, I predicted it.
- [24:30]
B
Yeah, it's so. It is, it is truly universal. When we talk to these companies that are building AI products or selling AI products, it seems like the ambition level in every single industry is just going up. Even, even Gabe from Roga was saying yeah, there's an opportunity to cut costs or there's an opportunity to get aggressive and like take market share. And Henny said, like a lot of firms are saying, like, let's just do more work, let's grow, let's grow our revenue, let's take on more clients.
- [24:57]
A
Right.
- [24:58]
B
It's harder and harder to doom about the job market.
- [25:02]
A
Yeah, you need to doom about the task market. If you're. It's over for tasks. Tasks are getting automated. No, it's serious. Like there are tasks that, like the reading of the X ray or whatever it was, the radiologist scan, what do they actually read? Is that an X ray that they're looking for tumors in? You know what I'm talking about, right? The famous Hinton. Yeah.
- [25:24]
B
I don't know if it's technically an X ray.
- [25:25]
A
It's not. Right. It's a CAT scan.
- [25:27]
B
It's some scan where you can look.
- [25:28]
A
Yeah, they do a scan and of course image generation and image models are very good at detecting cancers. But that is just one of several tasks, many tasks that a radiologist does throughout the day. And so radiologists are currently making more money than ever and they are in more demand than ever. And I was telling you before the show about truck drivers. They don't just drive the truck. There was a survey that said 70% of them carry weapons because they are effectively, they are security guards for the truck as well as probably a bunch of other things that you don't think about. Light repairs on the truck, you know, stopping for different things, rearranging things in the back, making sure that the load is like balanced and whatever.
- [26:08]
B
You're driving with kids in the car and you drive by a big truck on a road trip, you can go like this.
- [26:12]
A
Yes. And then that's another thing that the, that the self computer could never.
- [26:16]
B
A computer could never look out the window and see some kids in the car and.
- [26:20]
A
Yeah.
- [26:20]
B
And honk.
- [26:21]
A
But. Well, yeah, I mean, we'll see data center beautification. And he summed it up perfectly. He says make data centers aesthetically beautiful. And so people have been, quote, tweeting, posting, riffing on this, sharing different ideas. More people would be pro data center if every data center had a beautiful open to the public heated pool. That's an interesting twist. Full water slide. I think half pipe is a big idea. Monster truck rally constantly going on. These are ways to win people over. Sort of a nitro circus going on in every data center.
- [26:59]
B
Truck rally on the roof of a data center.
- [27:01]
A
Yeah, sort of a universal Basic Nitro Circus. So you get. Are you familiar with Nitro Circus? Is this a three fingers moment for you? You're like, I'm familiar.
- [27:11]
B
No, I'm just imagining they're building.
- [27:13]
A
Just say you've never actually been to Nitro Circus.
- [27:15]
B
I'm imagining they're building the dirt bike ramps over the data set.
- [27:19]
A
Exactly.
- [27:20]
B
Back flipping over.
- [27:20]
A
Exactly. Julie Young shares a trick. This one neat trick that Los Angeles uses to hide oil deckers. We have a few oil derricks in LA that are disguised as synagogues. They are literally just sitting in the middle of the city. Lowell. Surprisingly few people are aware there's a number of data centers that are disguised in Los Angeles as well. Usually you can tell because there's no windows in the building. But these exist in downtown usually for edge computing, for like the content delivery network. You put store in your Netflix videos there, that type of stuff. Not doing training runs, not gigawatt clubs,
- [27:55]
B
clusters, just in skin exams are getting automated. Oh yes, this is a very cool product.
- [28:00]
A
Okay, break it down for us.
- [28:02]
B
Let's pull it out. We have video here.
- [28:04]
A
Yes.
- [28:06]
B
I think anybody that's been to a dermatologist, it's like looking for potential, you know, skin cancers, things like that, abnormalities. Has had the experience of like the, you know, when you're, you know, that video or the meme where the, the guys like walking through security and the security guys just like, you know, does this like really pat down. It's like that's basically what like a lot of dermatologists are doing where they just like take a quick glance and they're like, oh yeah, looks good.
- [28:34]
A
Yeah, they might be busy, they might
- [28:35]
B
be sleeping in like a couple years.
- [28:37]
A
But this robot.
- [28:37]
B
So a robot that just does it like really, really precisely consistently. Ideally, hopefully. Hopefully a lot more accessibly. I can see this making a lot of sense. And then you actually. It doesn't even replace the job of the doctor. It just helps them speed up the process.
- [28:55]
A
Where is this company? What stage are they? Do they have to go through FDA medical device approval process? That could be a couple of years maybe. But this seems like once you get it through, it's going to be massively successful. I wonder where they are in the FDA process. Hopefully they can move along quickly. Arcprize put out an update on Ark AGI v3, which. Tyler, are you still on the human leaderboard for that?
- [29:21]
B
I don't think I'm on the leaderboard anymore. That was still early on. I think there's actually more games now.
- [29:27]
A
Roasted. Well, Arcade GI V3 has been very tough for even the most frontier AI models. But the good folks over at ArcPrize have benchmarked the two latest and greatest AI models from OpenAI and Anthropic GPT 5.5 scored 0.43%. Opus 4.7 scored 0.18%. It is state of the art. This is the weird thing about these ARC AGI tests is that they start with such a low baseline 0% 1%, and then they start shooting up as the models get more capable. But it is very interesting and reassuring to see these powerful models that you sometimes hear these narratives. It can do everything and then there's something that you know to continue working for, to look forward to in the future. There's more applications here. And so they found three total failure modes. True local effect, false world model, wrong level of abstraction from training data. Solved the level but didn't reinforce the reward. And so the ArcPrize account goes through a little bit of what's going on when an AI model tries to play Arc AGI V3, which is basically a 2D video game. Simple enough for any human to progress through, but increasingly difficult for AI models.
- [30:48]
B
Okay, yes. We gotta talk about one of the most exciting new products created in the last hundred years.
- [30:54]
A
Tell me.
- [30:55]
B
According to Katie at Business Insider, Amazon will now create an AI podcast about their products where two AI hosts discuss the products and take your questions as if it's a call in show. Yes, let's play.
- [31:07]
A
We don't need to play this one. This one is so rude. But you can imagine the type of products that people are generating AI podcasts for. Only the silliest things will be generated. I don't know if people want to listen to a full podcast about every product they buy on Amazon, but you have had some very strong opinions about paper towels. You wanted paper towels from like the 1950s or something. Remember this whole thing you didn't want?
- [31:32]
B
Well, I just wanted a filter to be able to buy. Only shop for things on Amazon now
- [31:36]
A
instead of a filter. Or you can listen to an hour long podcast about every possible skew and then you can make the most determined decision available. Go enjoy your weekend, enjoy your Friday. We will see you on Monday. Goodbye.
- [31:52]
B
Love you.
- [31:53]
A
Flashbang.