The Exchanges

Q What do you think it says? About either meta or maybe the technology in general, but like that you cannot hire a bunch of the absolute top talent in the industry and still deliver a model on par with these other massive companies.

A My thought here is that, um, ironically, I got this from Scale AI, Wayne's former company, that most model training has moved from pre-training to reinforcement learning. Um, and so I think when it came to pre-training, which again is just dump huge amounts of text, uh, in and then get the models to predict the next word, and then you can fine tune it afterwards. Um, now models are being really trained on doing tasks. So you just give it a goal and then it goes out and figures out how to accomplish it in and of itself. And that is much more specified than these pre-training runs where they gained some level of general attention intelligence. So I think you can very easily, I mean, they did get people on images and reasoning and all that stuff, but I think you could very easily, um, get great people, but I think it also takes a mass of people to do this reinforcement learning work and the models have started to diverge. Like open AI is great on health and anthropic is great on code. Um, and so you got to pick what you're best at. So that's one, one thing. And then I do think there's certainly culture issues there, which were predictable, right? The new, the new people were going to come in and annoy the people that were there and probably not be able to work together and probably not to be able to work with each other because they got their money. Uh, so that's, that's probably …

AI assessment note: “there's certainly culture issues there, which were predictable”

Q Yeah, I think so. I, I have to admit, like, I know, did you feel this was a big launch? Did, did you kind of, across your feeds, in your bones, did you feel this was like, oh, like a big launch?

A I think, yeah, this is an important moment for me to reflect on something that I've been meaning to share on the show for a while, which is that we can't really, you know, first glimpses of models sometimes don't tell the full story, and we slammed GPT-V and said it was something that just did not live up to expectations. No, I did. You, you actually called it. And you said it was a big advance, and I said this, this stinks, and I mourned O-III. Actually, it was a big advance, and so I'm going to hold my, my, uh, assessment for, for a moment, because sometimes it might seem like a small blip and turn out to be massive.

AI assessment note: “I'm going to hold my, my, uh, assessment for, for a moment”

Q actually potentially unlocking tens of billions of dollars, it's back again. And, uh, and again, we have our benchmark here on the podcast. Waymo's operating in New York City officially will mean AGI is here. But, but I think, I don't know, did you, like the complexity of the funding, do you call this, do you genuinely call this a hundred ten billion dollar round? There's so many stipulations here.

A No, it's not, it's not, certainly not that. It's, I think it's a very important point that you're calling out here. I mean, remember when OpenAI and NVIDIA said that they were going to do a hundred billion dollars together, and it was another one of these, well, it's ten billion now, and in time, turns out that hundred billion dollars was actually thirty billion, which is what NVIDIA will be investing. Although Jensen Wanga said, we hope they invite us, you know, to come back, but certainly a intent to invest a hundred is very different from an actual action to invest Uh, to me, the interesting thing here is where, you know, what happens as a result of these major deals? So this is from, uh, CNBC. OpenAI said it's expanding its existing thirty-eight billion dollar agreement with Amazon Web Services by one hundred billion over the next year, eight years. Uh, so So it's gonna get 50 from Amazon, but it's gonna put back either 100 or one 38.

AI assessment note: “No, it's not, it's not, certainly not that.”

Q again, like a matter of timing because there is like always going to be a downturn. And if there's a downturn, there's is, do you really believe that like big tech is going to be insulated by it from it? Because, um, You know, the future of AI. I mean, that would be the hope that they would project, I would imagine, but that's just not how markets typically work.

A Right. And I, I've mentioned earlier that I have this list of risks and it just keeps getting longer. The more we see the, the bets grow and I'll read a little bit from it. I mean, here's some risks that could really make things, uh, go sideways. One is that compute gets really cheap. Like maybe a company like grok, not Elon Musk's grok, but grok, the, uh, inference chip. Um, makes it so cheap, and then all of a sudden you have a company like OpenAI with a trillion dollars of build out, um, uh, contracts, and I don't know what happens there, but like a lot of companies stock prices are hinging on that. Um, another thing that could happen is that training doesn't yield improvements anymore. Like we've been running on this idea that bigger models will get you exponential, but exponentially better, um, you know, AI, sorry, bigger, bigger training runs will get you exponentially better models. Um, but that might be leveling out. If that doesn't happen, you know, there's going to be a sort of road runner off the cliff type of moment. Um, there could also be no economic advantage in holding AGI. And I think we're going to start talking about that, uh, much more on the show, uh, as we, you know, maybe get closer to this, but like if you have all powerful AI, um, and then, you know, you could have, it's not going to be a long time until a lot of others catch up there. And so then what …

AI assessment note: “here's some risks that could really make things, uh, go sideways”

Q comfortable with the assumption that they don't, or that they're somehow usable and responsible. So does this, is this gonna completely backfire on them? Or is this gonna, because trust is kind of paramount to the central. Every time you ask chat GPT a question, you're assuming it's At least relatively correct and, and responsible about how it's answering you, like, does this actually potentially hurt their, their regular usage?

A So, okay, I have two thoughts on that. First, let, let's not underestimate or, or de-emphasize the fact that these models have gotten much, much better. Uh, just, if you think about the level of hallucination, like I'm going to read out a list of Apple researchers, uh, that have left meta. Uh, over the, sorry, that have left for meta, uh, over the past couple of months. And, you know, I did a chat CPT query. It had all the researchers names. It was accurate. Uh, God help me if it wasn't, but it had links. I followed the links, confirmed the links, and it was right. So we have seen much better, uh, that these models have, as they've gotten better, they have hallucinated a lot less. They are much more trustworthy and it, it bears out in the data, at least Matthew Prince from Cloudflare talked about how, Uh, people, people don't need to go to the footnotes as much as they did anymore. Uh, the problem is of course becoming too trusting of it, like if it's getting 95% of the things right, and you trust it like it's a hundred percent, you're gonna make some big mistakes. So, but they, let's, I don't think we should downplay the increase, uh, in, in trustworthiness there. And, and then the second thing is, um, I expect this erotic or loving companion feature to be extremely popular. And your, your, I think this is important. We shouldn't glance over it. Your relationship with technolo…

AI assessment note: “I don't think we should downplay the increase, uh, in, in trustworthiness there.”

Q of concern. I really think this is a shakeout. It's kind of like a back to reality, come down to earth moment, but I don't think there's gonna be massive spillover effects from any kind of, uh, Any kind of rational analysis on what's really happening right now, and investors just start to, start to come back down a little bit. I think it's gonna be okay. What about you?

A I'm at a five. Uh, I do think that there's a non-zero chance that, uh, progress stalls, and a lot of this hype around AI just Translate. It just fades. Um, people realize that it's harder to implement than a lot of the hype was making it out to be, and it takes longer. The, the timeline is longer than it's anticipated, and, uh, and maybe that just leads to stagnation for a while. I don't know. I'm not convinced that's gonna happen, but I, I definitely appreciate the possibility. Um, all right, well, let's talk a little bit about OpenAI's announcement this week. They had this Dev Day, Developer Day. It seems like everybody just wants to build. Have you heard this before? OpenAI held its annual Dev Day on Monday, where the company rolled out its plan to build apps into ChatGPT. The demo showed how programs like Spotify and Figma can be called or discovered without leaving the ChatGPT window. With so much of the tech world barreling towards AI integration, OpenAI's demo was the best picture yet of what an AI-first internet might look like with interfaces like ChatGPT querying information and executing commands directly. I'm, I, I don't know why I can't get excited about this, or maybe I do feel like I've heard this from Google, from Amazon, from Apple, and OpenAI again, um, and, uh, and I'm not, like, gonna lose my mind over the platformization over, of ChatGPT. Am I underplaying …

AI assessment note: “I'm at a five. Uh, I do think that there's a non-zero chance”

Q an overall better way to develop and code and engineer. So, so I think It, I think that, yeah, the two are inextricably linked, but, but I'm curious, what did you take away in terms of where Anthropic is going? And we're going to talk about the funding in just a moment. Clearly there's momentum, but, but what does he see, really see as the future in, in your takeaway?

A I mean, I think that one thing that's clear is that Dario, this sort of does link to the funding believes that he's the firmest believer in the scaling laws, which is that if you add more compute data and increase the model size, you're going to get predictable exponential results. And so I think that's basically the way that they want to move is they have some near-term business interests to get them to a point where they can build bigger models and more efficient models and effectively trying to build a supermodel. Like I think everybody is like, It's one thing that's going to come up. I have MG Siegler coming on. We have MG on the first Monday of the month, and he writes this great newsletter called Spyglass. Um, and we're going to talk about the fact that all these companies are effectively working towards the same thing, like call it super intelligence, call it, you know, AGI. Effectively, they're working on scaling up large language models and adding sort of the scaffolding around them to make them useful. So I think that's really where they want to go is They believe that they can build the, the models best. They want to raise the money and build the biggest and best models. And then sort of, does anybody guess what happens from there? But we've seen these predictions from Dario that he, he has a belief that there's going to be like real economic value there. Um, and als…

AI assessment note: “labor augmentation and labor replacement is going to be where their business is.”

Q a huge variety of tasks and can't just kind of regurgitate and piece together the sum of human knowledge that they've harvested, harvested from the internet, but can actually come up with new things. You know, new ways of building rockets, new, new ways of generating power. Uh, and I think that's kind of, did you not feel recently that that's where the goalpost has been shifted? ASI from AGI?

A Yeah, it's definitely the new jargon term, um, which has even made it easier for a company like OpenAI to say, hey, you know what? We're going to call this AGI. And here's another quote from the Theo Vaughn podcast where Sam says, GPT-V is the smartest thing. GPT-V is smarter than us in almost every way, you know, and yet we're here. Dude, this is, this is it. It's coming. I, I, I, I mean, I am, again, I'm not stating this, Uh, conclusively, I'm leaving myself open to be wrong. I'll admit I'm wrong if this is what happens. Um, but if he doesn't come out straight up and say this is AGI, then there's gonna be lots of winks to it. I, I think that you're gonna see a tidal wave of commentary calling it that when this comes out.

AI assessment note: “Yeah, it's definitely the new jargon term, um, which has even made it easier”

Q will have to generate content. Like, I think maybe they start buying up, even if it's like the more kind of like informational type stuff that's very straightforward, sports scores and analysis or whatever else, like, I think they have to start buying up some kind of small media properties Because they're gonna have to feed in real-time content from somewhere, and maybe, is this the future of news, Alex?

A I think so. I mean, I think you could see it take shape in a bunch of different formats. Uh, the one way you could do it, you could do it is you could potentially have, let's say, you know how the White House has a pool report? Uh, so basically reporters from different agencies follow or different, uh, Publications follow the president and then write up this report, uh, that's, you know, sort of shared with the pool, and that's how we get a lot of our reporting on like what the president was doing is because they're relying on, uh, the pool report. Instead of having to have 50 reporters, they have one that distributes it. So, do we have OpenAI, for instance, paying for the pool report? Uh, and then just using that to surface real-time insights. Do we have it contract with individual journalists or publications and say, when you have a scoop, just like you would file it on. Yeah. I mean, this is similar to what you're saying, just like you would file it on your website. Can you file it into chat GPT? So I think the integration is going to be a lot more, um, A lot. Uh, it will just disintermediate the website. And in fact, like, um, we did a story on big technology a couple of weeks back, maybe a month back now with, uh, about world history encyclopedia, which is this site, uh, the second, the second biggest history site in the world. And it's CEO was like, yeah, we're seeing a 2…

AI assessment note: “I think so. I mean, I think you could see it take shape”

Q the energy in the industry is from the fact that everyone has had enough light bulb moments that they get. This is going to actually work at a certain point, but how we get there, is it the God model? Is it just some better implementation people? Come on, just Get your processes in place, but however we get there, I think most people have gotten it that we will.

A Well, I think we, I mean, we've been debating this as an either or, but in this certain, ah, use case, I think it's both. And, I mean, I think about the fact, so I've uploaded, ah, my podcast analytics to every subsequent model of, ah, OpenAI's GPT series, and said, here's the raw numbers, give me the trends. And those reports have gotten so much better as the models have gotten better, to the point where O-three was spinning some, like, unbelievable business intelligence. Uh, based off of the, uh, raw data. Like everything, the episode names, the listens, um, geographies, all this stuff. And so that's the thing. If we, if we're at the point where all these models, uh, have, have run into a wall, or are getting close to it, I don't think we're there. I think there's still room to go. But the fact that you have trouble in meta, and in, um, in Anthropic, and in OpenAI, in terms of pushing out the biggest models, and that, that Increase in size, which they thought would lead to exponential results is not delivering them. Uh, that's an issue. I mean, I'll speak with, uh, with DeepMind about it next week, but, um, it just seems to me to be a problem.

AI assessment note: “we've been debating this as an either or, but in this certain, ah, use case, I think it's both.”

Q what they're trying to do with this, and, and again, In one episode, I still love the fact that this represents kind of like cutting edge frontier technology relative to bad Rudy and Anna, Annie, and those kind of like anime characters, but, but I think that this is, it's an interesting move and we'll, we'll see if the most important thing, does it work and does it work well?

A Yeah. I mean, I think it seems like people are saying really good things about, uh, perplexity comment and, uh, I just got, uh, access to it. So I'll come in with a report next week on it, but, um, there's been trouble to get this done. I think, I mean, everything from Apple intelligence to Alexa plus, um, just doesn't seem like these agents are able to do the full range of things that people want to get them to do, including, uh, Operator, but again, like as this technology gets better and as they build better scaffolding or tool use, you know, those are those jargon words that matter a lot, basically giving them these capabilities to use these programs. I think we're going to see someone crack it eventually. This is from the TechCrunch article that sort of gets to the complexity. The launch of the ChatGPT agent represents OpenAI's boldest attempt yet to turn ChatGPT into an agentic product that can take actions and offload tasks for users rather than just answering questions. In recent years, Silicon Valley companies, including OpenAI, Google, and Perplexity, have unveiled dozens of AI agents that have promised to do just that. However, these early versions of AI agents have proven to struggle with complex tasks and seem less compelling as products than the ultimate vision tech executives pitch around AI agents.

AI assessment note: “just doesn't seem like these agents are able to do the full range of things”

Q Have we both just convinced ourselves that a hundred million is reasonable for these engineers? Because I think I am starting to be convinced of it.

A Absolutely. I mean, absolutely. Even when we spoke about it the first time, right? Once we, once Zuckerberg brought Alexander wing, what did I say on the show? There's going to be more. And this is a strength, a sound strategy because you have everybody talking about how pre-training is hitting diminishing returns. You have everybody talking about how data is hitting a wall. And so what do you need? You just need these algorithmic developments. Now let me ask you this. So, so I would say that this is a good bet, but I'm going to ask you this. Do you think this is a sign that like, and I, okay, I think I haven't answered this before I ask you, but that the, this AI moment is sort of in the last throes and sort of just grasping for anything that will allow for improvement, given that like the mechanisms that brought it here are starting to tap out.

AI assessment note: “Absolutely. I mean, absolutely... I would say that this is a good bet”

Q Okay. No, no, but explain to me, what would be your take on, you know, in a world where AI gets cheaper and models interoperable and application layer, et cetera, how do they win?

A Uh, I don't know. I mean, I think, I think this, this has always been, the problem here has always been you're building on something that it's very difficult to hoard. And I don't know if there's an answer to that yet. So they've, they've started to build products, right? And, you know, maybe get to go back to the laptop example from last week, you know, you're building something that's like a laptop can be used by a personal folks, can be used by people for personal reasons, can be used by people for business reasons. Um, And so, like, your, um, your, your, you can, there can be multiple laptop makers, but ultimately we see what happens. They compete based off of cost and, um, based on cost. And, you know, Apple's, you know, doing well in its MacBook business, and there are some others. Uh, but, you know, it's, it's not like world beating businesses. So that to me is, is like the real question here, um, in terms of where they go. But, um, I mean, the other side of it is you could say, let's say we take your line right here, that the product matters most. Uh, they both have built compelling products. You know, OpenAI with ChatGPT, Anthropic with CloudCode and CloudCowork, and those will continue to grow, and they've been the engine behind their, um, behind their growth, and maybe that won't slow down, even if the pricing power goes down a bit. So that's kind of the way I think …

AI assessment note: “they both have built compelling products. You know, OpenAI with ChatGPT, Anthropic”

Q out of the goodness and the altruistic nature of the comms professionals at Anthropic, that's why they came up, or maybe the PR agency that's who hired it, or maybe Claude was so good that it came up with this strategy on its own. Is it for the good of humanity, or is it because they raised a three hundred eighty billion dollar valuation round a month, two months ago?

A Now, let me tell you what I think is actually going on. And it sort of maybe is in the middle of all these. And is it a little tinfoil hat type of theory, potentially? Maybe it's somewhat conspiracy minded, but I don't care. I think, I legitimately think there's a chance that this is what's happening. Okay. Think about what we've seen with Anthropic and OpenAI recently. Remember, these companies released Claude and ChatGPT originally as demos, as ways to show off what their technology is capable of, so you might buy some intelligence metered from their API. Over the past three or four months, both of them have gravitated toward building a super app. Uh, something that uses the most advanced intelligence to control your computer, that will, to help you get things done, to in some cases even build new software for you, which has created this big saspocalypse moment, and also on the other hand, Has helped them raise globs of money, under twenty two billion in OpenAI's case, thirty billion in Anthropik's case. This has effectively enabled the build out, uh, that they are embarked on, which is going to help them, um, raise more money and grow bigger and build bigger, bigger models. And so, as these models get better, I think there is a question that is taking place within these labs. Do we take the intelligence, the most intelligent models that we've built, and do we keep them exclu…

AI assessment note: “Now, let me tell you what I think is actually going on.”

Q event going on, and my god, that was just something crazy to see. But, like, overall, I mean, I don't know, are you hearing or seeing, like, it's just kind of quietly fading off. And which I think is possibly one of the most dangerous things because it really depends on kind of an energy to keep it moving. So yeah, what do you, what do you think is happening?

A I don't know. Uh, I, I just think it's one of those things that sort of it's built on. I don't want to say group think, but built on a story, right? And I don't know the story, the story, when the story doesn't work, Bitcoin drops. We've seen this so many times. I mean, the story works. The story is, okay, like, you know, the president is a, you know, a supporter of this, and now you could be able to own it from mainstream financial institutions, then it works. But I, I, I mean, I've, and I hold some, I'll admit it, but I've always felt that there was weakness in the underlying story, which is, you know, if it's not going to be useful as a currency, and what is it? Store value? Okay. That just depends on people's belief in it.

AI assessment note: “it's built on a story, right? And... when the story doesn't work, Bitcoin drops.”

Q Explain to me the flywheel with search. Like we have personal context. We have personal intelligence, but, uh, charging advertisers to show up in your search results, which are presented in some kind of order is blue links. And to be higher on that page, that is the search business model. It's not actually just searching for stuff. That's the business model. So how is it a flywheel for that?

A Well, I think the, okay, the, the, there's the business and then there's the product. And so this, I think Pierce is talking specifically about product. Now, one interesting thing about generative AI is that the queries are much more varied than search. And I think something like one third of searches are brand new to Google every day. I would argue that many more, uh, of these questions to generative AI, uh, models are novel. And so I think to become good at doing what you do, You have to be able to really appreciate the broad range of queries and figure out where you're doing well and figure out where you could get better. And so now Google is going to put its, its, um, It's product in front of the billions of Apple users who are now going to be giving it data back. Now, maybe Apple's going to build some sort of fly, uh, some sort of firewall. But the fact is that this partnership is going to make the Gemini product better because it's going to really understand a much broader range of user behavior and optimize based off it. Whereas again, with ChatGPT, they got to fight for every single user. They don't have the distribution of their own products like Google does. And they certainly don't have this Apple deal anymore. In fact, Google is supplanting ChatGPT in Siri.

AI assessment note: “partnership is going to make the Gemini product better because it's going to really understand”

Q about what their business looks like and breaks down, especially as all these new ideas, the pharma, Idea, drug development that Sarah Fryer had mentioned, like, it's clearly on their mind. You got to have some plan. It's December, 20, 25. If you're talking about two to three years now, there's some plan in place. What do you think that business is going to look like at a hundred billion?

A Well, I've told you that, like, we asked, like, where did they get these projections from? And I think that they've been sprinkling some fairy dust on the, uh, on an Excel spreadsheet and hoping that gets them there. I don't know. I don't think they know exactly what's going to go there. I think they're betting that rapidly, that they've bet it previously, that, you know, you scale up these models, you add more data, you add more compute, uh, and you will get Rapidly, rapid improvement in your ability in this, your ability to, to do things with these models, and they will lead to economic activity. I just don't, I, I don't see how they get to that number that quickly, because once we both know this, because you're working in this, I've done reporting on this. Once you get into enterprises, what happens? Even if you have a product that works really well, uh, you have to engage in business transformation and even worse approvals. And that does not move fast.

AI assessment note: “I don't think they know exactly what's going to go there.”

Q of not necessarily obfuscation, but other ways of framing it. It's like, okay, we agree that, uh, we shouldn't spend tens of billions of dollars on NVIDIA server farms, but we need to build out, um, you know, our, our in-person, um, AI robotics arms, right? In order to, to keep these models and, and keep sort of the next phase going as we march towards AGI and yada yada.

A So markets just opened, NVIDIA opens up down 11%, so still above three trillion dollars, so it's not like the AI revolution is over, but down 11%, so just a cool, you know, couple hundred billion dollars shaved off the market cap in a morning. Let me talk to you a little bit about what these companies are saying back to Wall Street, or actually talking to Wall Street, about to allow them to keep spending. So Satya Mandela is doing his tweets, he says, he's talking about Jevon's paradox. He calls, he says, Jevon's paradox strikes again. As AI gets more efficient and accessible, we will see its use skyrocket, turning it into a commodity we just can't get enough of. Um, let's say that happens.

AI assessment note: “what these companies are saying back to Wall Street... to allow them to keep spending”

Q an incredibly lucrative, high valuation IPO. Like, there's no way if this, like, was a real thing. Like, if everyone, if every banker who's on the Anthropic deal truly felt that this company could literally destroy the financial system and I've seen indications of what it will and can do and drain my own bank account and ruin my own standard of living, would they work on the deal? Maybe.

A Can I, can I, can I give you the reason why they would? Um, I mean, if the, if, if this power is truly the direction that we're heading towards, right? And it almost always starts, like, we know the history of the internet. It starts with a game or it starts with some crazy interaction and then our crazy, like, you know, sort of mistake, and then it ends up being something that everybody uses. If we're seeing the type of power of these AIs to do what they did in this situation, I mean, just imagine what they could do if you could actually harness that power and use it for economic activity. It kind of goes back to what we were talking about earlier. The fact that these benchmarks are being hit and we're not really seeing the difference, the change, um, if there's a way to pro, like to, for good, to harness this activity for good, and of course that's economic, it could be used for health, it could be used for all different types of things, Um, it obviously changes the world, and, ah, potentially if you can use it in a productive sense, can change the world for good.

AI assessment note: “Can I, can I, can I give you the reason why they would?”

Q frontier or even like, I'm just pulling a name out of my head, but like a Jane street, if you're going to spend two million dollars per head, but you're going to generate like twenty million on each, you know, um, bet like on each investment, then that money is justified to Jane street probably. But I think, you know, it goes back to how do you justify that cost?

A Right. And I think you also, I mean, you, you effectively have the answer in your piece or, or maybe a different answer, uh, or maybe an answer that expands upon this in your piece that I think is, you know, sort of says it all. Right. So you say, uh, and this is your reaction to Kimmy K-III, you say Kimmy K-III, but you can basically say this about this entire open source, uh, moment. It, you say it's the latest evidence that the AI frontier is becoming more contested, more global, and potentially less proprietary. And I think we talked about this a little bit last time also. What happens when the AI intelligence itself becomes more proprietary, becomes less proprietary? The thing that matters is the product, right? So I asked earlier, what do the AI labs do? And we've been on this a bit on the show recently. If you're, if you're, um, if the intelligence that you've developed becomes less proprietary, um, if it becomes more of a commodity and less of something that you can hoard, Your products are, are just what matter. And we, we saw this with, with Deep Seek and we're seeing it again here. Um, I think that if there was a belief that you could build a trillion dollar company just by selling intelligence from the API alone, uh, not gonna happen most likely. Uh, but what could happen is you, if you're the developer of the intelligence, you can build products with that intellige…

AI assessment note: “you can build products with that intelligence, uh, and, and sort of return to your investors”

Q Recent weeks, week, month. Um, but yeah, I mean, I think, do you, what do you think is happening? And what do you think will happen? And should we extrapolate based on what's been happening?

A Well, okay. I don't want to lose sight of the bigger picture here because I think I brought this in to really talk about how this added compute and this teaming up of SpaceX and Anthropic is going to be Answer one of Anthropik's biggest problems and has made the AI race much spicier, right? Because now you have Anthropik and its key models against OpenAI, and they're both less constrained by compute than they were previously. I mean, OpenAI was less constrained, and now Anthropik is also less constrained. So this is going to be a heavyweight battle. I do see them as being the likely winners. I do think they'll be able to get the models to act more efficiently. I mean, goodness gracious, after my experience with Force Evan, I hope, I hope Anthropik does. Um, and so therefore this is, this is a, we've seen a consolidation and a solidifying, uh, of this race. And yeah, I mean, are there going to be bumps in demand? Sure. Right. I mean, you, you are working in enterprise AI. I'm doing a lot of reporting on it. The number one thing I hear is that a lot of this is just boards telling CEOs you need an AI strategy and them spending money. Right. And without really figuring out what the problem is or diagnosing how they, how they would solve it. So yeah. There's part, part of that does exist. Um, but I think the broader story is, can't really deny, I mean, neither of us will deny the pr…

AI assessment note: “So this is going to be a heavyweight battle. I do see them as being the likely winners.”

Q it can go haywire and kill the wrong people? Or, or I guess, actually, is the argument that not that it will Go haywire and go kill a bunch of random people? Or is it, are they implying that the Department of War, it's still weird for me to say that, ah, actually will use it for nefarious purposes and that's the risk? What do you think's implied in there?

A Again, like my perspective here, and I understand why, why, why Anthropic would not want to sign this away to the Pentagon to like have like full use and whatever it, because if it's a company comes in with values, it has its values. But again, like, I don't know if there's a concrete worry here. That's what I'm trying to say. I think it's mostly just like, you know, blanket no, this is against our values and, and let's go. And even Emil Michael was on, I think he was on Fox business or Fox news saying that like, we're in the middle of this discussion and this blog post comes out. You know, again, like, I, I do think, and again, I, I don't want to feel too cynical about this, but I do think that this is, you know, sort of a PR opportunity. But back to your point, I mean, like, you know, I try, I would trust, think about this, I would trust a Waymo, I would get in a Waymo, I would trust it to drive me. Um, I know it's probably good on, like, 99.7% of, of, of rides or whatever. Much better than humans. But I don't want Waymo to be the police. Like, I'm not giving Waymo a gun and saying, if you see a crime, go arrest somebody. Like, I don't trust it to that extent. That's the difference I'm trying to make here.

AI assessment note: “I don't know if there's a concrete worry here... this is against our values”

Q whatever it's gonna look like. Maybe they'll, Rufus, actually, I've been using a lot more, I do find pretty good, but like, I think they're gonna, there, there's no, it's hard to see what's the big growth story With Amazon right now. Do you, do, do you see any, if you're, if you're to be outlining at Amazon investor day and you are Andy Jassy, what are you talking about?

A The one I just cited is that it's the biggest cloud services company and it just grew 24% after going like 17, 18% for a bunch of quarters in a row. That's pretty good for Amazon. But I think there is starting to be some concern about Jassy and, um, This one Twitter user discounted trash flow. I'm just reading. I don't agree with this, but sort of the extreme view on Jassy calls him a bean counting bureaucrat who lucked into the CEO chair after brown nosing Bezos for decades inherited the greatest e-commerce machine ever built and immediately started breaking it oversaw the biggest layoffs in Amazon history. 27,000 heads, uh, chopped in waves while he pocketed two hundred million in comp packages forced a brutal five day return to office mandate that sparked internal revolts. Leaked memos and mass quiet quitting. Led retail bleed billions in over expansion hangover. Closing warehouse he helped overbuild. AWS growth decelerated hard on his watch as Azure and Google Cloud closed the gap. Stock flatlined or lagged behind peers for years. Post-transition until AI hype and rate cuts bailed him out. Face his massive FTC antitrust lawsuit. For the monopoly abuses he pretends were in his fault, uh, zero vision, infinite corporate jargon, toxic spreadsheet obsession, preaches day one culture while turning Amazon into a soulless cost-cutting machine, just another over-promoted middle man…

AI assessment note: “The one I just cited is that it's the biggest cloud services company”

Q mean, we might mean different things by agent. And so I'm just curious, Alex, what do you, what? Yeah, I'm sure we're grounded in your definition because I highlight just in passing, I mentioned yesterday, my own, Part of, uh, part of our event, you know, that, boy, everyone just uses this term agent, and I do think people use it in different ways. What does it mean to you?

A Yeah, it's such a great question because I do think that in some ways that agent has been used to rebrand automation, uh, We've been seeing automation demos forever. I mean, even, so just to give you one example, I wasn't trying to shave the Amazon demo. I was just to give you one example. We were all, I mean, we, a lot of folks watching the tech world were at Google I.O. when they demoed a voice assistant that will go, will call a restaurant for you and book you a table. And like, they did the actual conversation, and the, the assistant has like, human utterance goes, um, well, maybe we could have a table for, and it's like, and then it would actually go and book you, book you the restaurant.

AI assessment note: “in some ways that agent has been used to rebrand automation”

Q great article, but it really gets at this issue. And so could he, could he go after, uh, some individuals? I wouldn't be surprised if he weren't, and I wouldn't be surprised if he even went after an Elon Musk. I wouldn't be surprised if he tried to, um, go after some of the people, uh, some of the higher profile people involved in SPACs. Just to make a point.

A Yeah. Oh, it'll be interesting. I mean, the, the, uh, order of operations on that is going to be important, right? Because if you're, if you go only after like the folks involved in this retail trade, and I know there's lots of implications there and you leave Elon and you leave the SPAC guys alone, you could have a problem on your hands. So bingo, let's talk a little bit about, uh, another speculative asset, although I'm going to get in trouble for saying that Bitcoin. Um, so when I hear you talk about Bitcoin, you seem somewhat amused Pretty skeptical. Um, obviously the price over the last couple of months has gone wild, you know, up to around 60,000, down to 30,000, back up again, down again. Where do you think the freight train's heading on that front?

AI assessment note: “if you leave Elon and you leave the SPAC guys alone, you could have a problem”

Q my parents' place who still have cable, and it actually kind of made me miss cable, and I don't know, I think they're spending like a 180 bucks, which I think I'm spending more now. But is, is all this AI hype going to end up with us longing for the days of basically what streaming has done, make me wish and miss cable? Is that what's going to happen?

A Well, first of all, it's just going to get worse for two reasons. Uh, one is we're starting to see consolidation in the space. Like you have Netflix as this clear winter and then, um, and then Paramount and Warner Brothers Discovery are going to tie up. Right. So, like, you're gonna see, well, it's actually good that the two will balance each other out, um, but, like, the days of every streamer competing with every streamer kind of on even footing in price matters, uh, seems to be away, seems to be going away, and then, of course, um, with AI trained on human screen behavior, um, they now are required to watch at least five hours of Netflix during the workday, and that is a demand signal that we're all gonna get screwed by.

AI assessment note: “Well, first of all, it's just going to get worse for two reasons.”

Q America is that place? If not in America, where, where do you go? Do you go to the Middle East? Do you, like, get some aluminum refinery in Canada? Because those consume a lot of energy, and you can just, like, buy out the aluminum refinery and turn that offline, and I don't know, but you can try out different ideas, but I think energy will be the biggest strain.

A And you basically, Zuckerberg talked about the fact that you might need a moderately sized nuclear power plant to be able to, to do this. And you asked, well, what about Amazon? And he said, ask Amazon. And I asked Amazon, did a little research, right? It actually, Amazon actually has purchased a nuclear, a small nuclear power plant in Pennsylvania. Correct me if I'm wrong here. And it's 960 megawatts. So close to that gigawatt. Uh, size, and they're gonna use 40% of the energy there, I, I assume for AI training. So is that sort of what this energy, uh, battle looks like moving forward? And is that even enough energy given that everybody wants to add more compute and more data and more energy into the process to actually be able to build these models?

AI assessment note: “Amazon actually has purchased a nuclear, a small nuclear power plant in Pennsylvania.”

Q got even more than that. Like to the point of, you know, like you show up and there's aliens, uh, all of a sudden, if, if we're like, you know, human race says whatever, six or seven billion people, if there's all of a sudden a hundred billion bots on these, on these networks, like, what does that look like? Uh, and how do you, Possibly hope to control that.

A I don't know if you can. I really don't. I mean, I hope we can, but this is definitely uncharted territory. So, uh, another big story that that's going on this week that we definitely shouldn't miss this, you know, speaking about today is that Nvidia had this one hundred billion dollar investment in OpenAI, uh, and it's seeming like it's pulling back here. This is from the Wall Street Journal. Nvidia plans, uh, Nvidia's plan to invest up to a hundred billion in OpenAI has stalled After some inside the chip giant expressed doubts about the deal. The companies unveiled the giant agreement last September at NVIDIA's Santa Clara, California headquarters. They announced a memorandum of understanding for NVIDIA to build at least 10 gigawatts of computing power for OpenAI and to invest up to one hundred billion to help OpenAI pay for it. But in recent months, NVIDIA CEO Jensen Wang has privately emphasized to industry associates that the original one hundred billion dollar agreement was non-binding And not finalized. He also said, he also privately criticized what he described as a lack of discipline in OpenAI's business approach and expressed concerns about the competition it faces from the likes of Google and Anthropic. Much of the recent concern about OpenAI has come from the success of Google's Gemini app. Uh, Anthropic is also putting pressure on OpenAI thanks to its popular AI c…

AI assessment note: “I don't know if you can. I really don't... So, uh, another big story”

Q and browse versus having a personal shopper that you just talked to while you're sitting at your desk and they go out and buy stuff for you. But, uh, yeah, I think it's, there is definitely a large universe of transactions that will work in this way. And this is going to happen. I do believe that. Just if it, is this the predominant way people shop? I don't know.

A I think it could be, and it's going to sound crazy, but let me give you a couple of data points here. So first of all, when you're shopping on Amazon prime, people have become conditioned to just basically take Amazon's choice and buy it. And that is because they've had enough trust and enough positive experiences within Amazon that they believe that they're going to get the best deal on the internet when they're on prime and they don't need to go to too many sites. I think it's become a natural behavior. Now, when you trust ChatGPT, when you're, let's say you're married to ChatGPT. No, just kidding. But, but, uh, let's say you're talking to, I mean, maybe you are, um, when you're talking to ChatGPT. No judgment. Do your thing. Get married to ChatGPT and then go shopping with it. Buy nice things.

AI assessment note: “I think it could be, and it's going to sound crazy”

Q can't you also do a music service? Why can't you also do movies? And is, is I know that Waymo comes out of Alphabet, but is this, if it's a success, 10 years from now, is it a story of like, well, yes, um, I don't own a car anymore because I have a Waymo or a cruise subscription, but that's sort of walled off from the larger tech story?

A I think it's a great tech story. And I think that like we sometimes can judge the, the impact of tech in terms of the way that it rearranges our world. Right. And for, for better or worse, I think this is mostly a better situation. And, you know, despite the fact that everybody tags me when a cruise falls into a puddle of cement in San Francisco, which like not great, but come on. Um, You know, we, you know, anyway, we're going to talk about it more on Wednesday about all these, these headlines about the cruise hating a purse or someone being critically injured under a cruise. It turns out it was a human driver that slammed into them and they just landed under the cruise. But, um, but I think that this could really reorient the way and not to get too deep into this because it's going to take a while and these promises have been made and, and delayed. Um, but it will reorient the way that our cities work. It will change, you know, you know, if you don't, if nobody owns these cars and you're just hailing them, then you don't really need space for parking. I think it will create a much better world and, and will probably be, I mean, I put it up there with iPhone, chat GPT and Waymo cars, the three most amazing experiences I've had with tech in my lifetime.

AI assessment note: “I think it's a great tech story... judge the impact of tech in terms of”