Demis Steps Down, Apple’s Memory Problem, Microsoft’s Clever Trick

**Alex Kantrowitz** (0:00)
Hey, everyone. I had a feeling that there would be an emergency podcast of some sort on my break this summer. As AI news has been progressing so quickly, there seems to be something emergency podcast worthy every week. And so, of course, while I was flying from New York to Indonesia on the plane, I learned, or actually when we landed, I learned that Demis Hassabis had stepped down as CEO of Google DeepMind, which is some of the most seismic news of the year in AI. And I figured that what I would do would be to share a couple notes on Demis stepping down and what it means for Google and what it means for AI, and then play the full Friday show that we have recorded. It's going to be with me and MG. Siegler. We're going to talk about Apple's memory issues. And then we also, at the end, talk about some of the misalignment of interest within Google that a couple of days later seems to have manifested with Demis' departure. So, a quick couple of minutes for me here, talking a little bit about what it means, whether it's good for AI and where Google goes from here. And then I'm going to play the full episode from MG. And I'm going to try to do this as quickly as possible. So, first from The Wall Street Journal, Demis Hassabis is stepping down as chief executive of Google DeepMind to become chairman and chief scientist, Google CEO Sundar Pichai said in a post on X. Google DeepMind technology chief Corai Cavucciolo is taking on responsibility for all AI model development and Jeff Teen, Google's current chief scientist, is leaving with three other company veterans to co-found a new AI startup. Okay, let's talk about why this happened. If you think about what Demis and his perspective has been over the past, let's say, year, year and a half, and he's been on the show a couple of times, it's been very different from some of the others in the AI industry. You speak with Dario Amande, you speak with Greg Brockman.
Their perspective is that the current LLM technology is potentially unlikely, according to their opinion, a straight shot to AGI. With Demis, it was always there is a couple more breakthroughs needed in order to get to human level artificial intelligence and so his interests really diverged a little bit from where a lot of the industry and a lot of the money is going in AI. And that money part is important because right now, this is key for business. The companies that have been able to build strong LLMs have been able to make a lot of money off of it. And if you didn't think that progress was going to continue apace and you think that there are other research areas more promising, it would almost be tough to continue to go along even if there was the business interests aligned with... It would be tough to go along even if it made sense for your business interests if you thought the research, most promising research is going to be elsewhere. And this is what the Financial Times reports today. It says, while Hassabis is widely admired for his scientific leadership, several people familiar with Google's thinking said senior executives had become frustrated by what they saw as his lesser focus on the commercial demands of the company's AI business. So basically, Demis is thinking his goal is to develop super-powerful AI, and he seems potential in areas outside of LLMs, where while the whole technology industry is moving towards large language models with the GPT model and CLAWD. And you actually have seen this kind of manifest with Google, where the breakthroughs have been on the more world model type of applications and applications of AI that resemble and help us understand the physical reality, things like Ioffe-Fold. And in fact, when I spoke with Demis in January last year, one of the things he told me is he wanted to make an artificial cell. You just don't do that with a large language model, you do that with other types of AI models. Okay, that brings me to my second point. Is this good for AI? Yes, it is. There are trillions of dollars that are going into developing LLM technology right now.
One of the things with AI progress is that if you go all in on one mode of the technology and it doesn't manifest results, you could end up having an AI winter, a drawdown people believe less in it. Of course, you don't want to invest entirely in one technique. I actually think it's good for AI that Demis will be focusing more on the scientific side, probably more on the world model side, and not deeply in LLM. So you'll have basically the talent at OpenAI and Anthropic and potentially what's left, the AI organization at Google, pushing the LLM methods forward and Demis going out and looking for the one or two breakthroughs that are needed to push AI to where he wants to go. And look, if one of them succeeds, whether it's the Anthropic OpenAI group or it's the Demis group, ultimately the technology progresses. So that's something to keep an eye out for. Okay, lastly, where does this leave Google? Look, obviously this is not what you want in the middle of the most high stakes fight in technology history. Things are moving exceptionally fast. Google is clearly behind on the coding agents. They don't really have something that is in a league with the latest large language models, but also with the applications with the Codex and the Cloud Codes. Google has not made noise on that front. And it is a shame, honestly, for the company that it could not make it work with Demis because clearly this guy knows what he's doing with AI. So you don't want this to happen if you're Google. But the other side of it is, if you feel like you need to make a move, you have to make it. If you feel like you're being held back, if you feel like your efforts are not progressing maybe because of diverging interests or differences of opinion, you can't just stick with it just because, you know, that's the way things have done. So it's a high stakes move. And then of course the question is whether Google has the talent to continue to stay in league with the others because of course you're going to see, now Demis is staying on, he's going to become the Google DeepMind Chairman and Chief Scientist.