Marc Andreessen: evaluation

Transcript context

…Almost for fun. Let me ask a friend of yours, Yann LeCun, what are your top 10 favorite things about Yann LeCun? I think he’s a brilliant guy. I think he’s important to the world. I think you guys disagree on a lot of things, but I personally like vigorous disagreement, I, as a person in the stands, like to watch the gladiators go at it. And- No, he’s a super genius. I mean, look, I wouldn’t say we’re super close, but casual friends. I worked with him at Meta. He was the chief scientist at Meta for a long time and still works with us. And obviously is a legendary figure in the field and one of the main people responsible for what’s happening. My serious observation would be it’s the thing I’ve talked to him about for a long time, and I keep trying to read and follow everything he does, is he’s probably, he is the, I think, see if you agree with this, he is the smartest and most credible critic of LLMs as the path for AI. And he’s not, there’s certain, I would say, troll-like characters who are just cropping everything but Yann has very deeply thought through, basically, theories as to why LLMs are an evolutionary dead end. And I actually, I try to do this thing where I try to model, try to have a mental model of the two different sides of a serious argument. And so I’ve tried to internalize that argument as much as I can. Which is difficult because we’re investing it behind LLMs as aggressively as we can. And so if he’s right, that could be a big problem, but we should also know that. And then I sort of use his ideas to challenge all the bullish people to really test their level of knowledge. So I like to grill people. I got my CS degree 35 years ago, so I’m not deep in the technology, but to the extent I can understand Yann’s points, I can use them to really surface a lot of the questions for the people who are more bullish. And that’s been, I think, very, very productive. So it is very striking that you have somebody who is that central in the space, who is actually a full-on skeptic. And again, this could go different ways. He could end up being very wrong. He could end up being totally right, or it could be that he will provoke the evolution of these systems to be much better than they would’ve been. He could be both right and wrong. First of all, I do agree with that. He’s one of the most legit and rigorous and deep critics of the LLM path to AGI know. His basic notion is that there AI needs to have some physical understanding of the physical world, and that’s very difficult to achieve with LLMs. And that is a really good way to challenge the limitations of LLMs and so on. He’s also been a vocal and a huge proponent of open source, which is a whole nother, which you have been as well.…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.