Yann LeCun bet a billion dollars that it can. He left Meta arguing today’s chatbots are a dead end, and that real intelligence comes from “world models,” systems that learn how the physical world works rather than just predicting the next word.
Two things nag at me.
First, how do we even measure it? Every famous AI test is basically a language exam. But a world model doesn’t write essays, it predicts what happens next. So either these systems slip past the tests we trust, or we have no good way to score them yet.
Second, LeCun says you can’t reach real intelligence through language alone. Probably right. But isn’t the reverse just as true? Could anything that masters physics but can’t grasp language really be called intelligent? So much of human thought, math, planning, culture, rides on words.
My gut says neither pure chatbot nor pure world model gets us there. The winner is some marriage of the two.
So maybe the question isn’t chatbots versus world models. It’s how the two work together.
Is language the engine of thought, or just a handy way to talk about it?
[link] [comments]