AI与下一个词预测揭示了什么是智能

One thing that this whole era of AI is teaching us is what-- is our idea of what intelligence is, is not really accurate.

The, uh, the history of AI has been, here's a task that only humans can do. Like maybe it is read natural language or win at chess or solve a math problem. And then one by one, someone finds some AI algorithm that also does that.

Like, okay, we can now recognize faces. We can, we can now understand speech and-- But you look at how it's done, and it doesn't feel like intelligence. Like it's, oh, it was some trick.

You just, you just cobbled together these neural networks and random algorithm.

Mm-hmm. And we were looking for some elusive, intelligent way of thinking, and we don't see it in the tools that actually solve our goals.

But maybe it's actually because intelligence is not what we think it is.

So like large language models in particular have become very successful, and a lot of what they're doing is just predicting the next token, predicting the next word in a sentence.

And it, it doesn't sound like something which is intelligent.

Like you ask someone to improvise a speech, and they don't have no preparation, and at every moment they're just saying the next word that comes to their mind.

Stream of consciousness dump. Uh, you don't think that this is-- this could actually work, but it does for LLMs, and maybe that's actually a lot of what humans do as well.

This response I'm giving you right now is not rehearsed.

I haven't-- I didn't think about it that deeply, but it's-- You can get very far just by being in the moment and reacting.

Um, it's-- Certainly, there's not... There's some things that actually do require careful, deep thought, but you, you can do a lot of what seemingly seem like intelligent tasks almost by instinct or by autopilot. Um, um, that's I think one thing that the age of AI is teaching us, that maybe we're not quite as smart as we think we

are sometimes.