Carl Yeh

So how do you collect that type of knowledge?

And that's in all enterprises, all businesses, because we know there's always an exception.

There always is a nuance.

There's always edge cases, and an agent can't take care of that.

And it goes back, actually, to the discussion on artificial general intelligence.

Until it can do that, I don't think we can call it that way.

Yes.

I think they also mentioned like in the end, you're still giving it to a prediction model.

Like it's still predicting.

Well, hence, this leads to the argument that Jan LeCun had, Ilya Seskova has, that the LLM can only go so far.

You need something else that is able to take into account, hence, decision traces and that stuff.

And I don't know.

And we're just talking about regular, maybe a corporate business decision.

What about external decisions based on visual cues?

You're like, oh, if this happens, if I see it, then this happens.

How do you do that?

If you go into a business context, there's certain cues that people would have, but then sometimes that's why there's like, there's a person who is your expert at this, but it's based on a lot of times I was talking to one client and he's like, why do I ask him, why do you make decisions?

Like a lot of times it's by intuition and gut because I've seen it before, but

But I've seen it.

Like, how do you take that and put it to an agent?