Yann LeCun

All right.

So I'm going to go a million different directions with this, but it sounds to me, if I can play this back to you, there was an epiphany moment where as you were, and I'm presuming you're looking at things like looking at Jan LeCun's work and some of those early, of what we would now consider to be the modern sort of way of thinking about machine learning and large language models and so forth.

You're looking at that research, and I'm assuming it's like 10 years ago, give or take, and you see something in there that gives you, is it alarm? You're sort of sitting there thinking, I see a new attack surface developing.

What was the aha for you?

There was therefore no strong reason to assume that it would persist as models became more capable.

Researchers such as Yann LeCun have argued for years that more capable AI will require architectures with explicit world models and planning mechanisms rather than simply scaling autoregressive language models.

Astra does not prove that Lacoon is right, though I think he is.

And declining chain of thought monitorability does not, by itself, show that large language paradigm was a wrong turn.

You can Google it.

So remember that I often bring up Yann LeCun and JEPA, which is a new type of architecture that's not an LLM.

Yeah.

Next concept modeling.

And that's why RL is so exciting.

Do you buy Yann LeCun's post-transformer world theory? And how does that change everything?

I do think we will be in a post-transformer world.

I don't think the transformer is the be all end all.

real-world models.

Yann LeCun and Fei-Fei Li are working with that.

It's really more about teaching them the laws of physics and so on.

The contrarian thing I'll say about this is that if we all end up using AI to some extent, and the use of it is pretty high, and if we use it for important things like medical questions and such, then it's in no one's interest to have stupid AI.

We need that from Europe.

We need that from investment in more places.

We need that from places like Yann LeCun and his AMI labs and Fei-Fei Li, now bought by a big company, to give us different paradigms and models.

And if I were going to talk to regulators, as you do regularly, I would wish for large investment in universities to do research around this, to do more foundational work on AI, to take away that cachet from the companies themselves, to enable watchdogs of various sorts.