The biggest companies in the world are spending billions and billions on AI — specifically on “large language models” like the kind that power the likes of ChatGPT, Google’s Gemini, and Meta’s Llama.
That’s a mistake, says the man who, until recently, headed Meta’s AI push.
“LLMs are great, they’re useful, we should invest in them — a lot of people are going to use them,” said Yann LeCun, who’s still an AI researcher at Meta for the moment, speaking at an event in Brooklyn Sunday night. “They are not a path to human-level intelligence. They’re just not. Right now, they are sucking the air out of the room anywhere they go — and so there’s basically no resources [left] for anything else. And so for the next revolution, we need to take a step back and figure out what’s missing from the current approaches.”
That’s a bracing critique of his employer’s strategy. It’s also not a new one: LeCun has been a consistent critic of LLMs for years, arguing that real computer intelligence won’t come from language models that hoover up text from the internet. He thinks breakthrough AI will come from “world models,” which rely on visual data instead.
But the timing gives his remarks extra oomph. There has been ongoing speculation about LeCun’s future at Meta for many months. That speculation got much more intense last spring, when Meta started spending billions to buy and employ an all-star roster of AI experts focused on LLMs — essentially a repudiation of LeCun’s approach.
And last week, news broke that LeCun was likely to leave Meta and launch his own AI startup.
LeCun made a point of not addressing those reports directly on Sunday at Pioneer Works, an artsy/tech cultural center that was overflowing with Gen Xers who remember using dial-up modems, as well as Gen Z types raised on TikTok. But his comments sure sound like an explanation for a yet-to-be announced move. He thinks LLMs aren’t the future. Mark Zuckerberg does. So it would be weird for LeCun to stay at Meta. (I’ve asked LeCun and Meta for comment).
I think LeCun’s commentary also gets at something more meaningful than an outspoken employee clashing with his employer. It’s a reminder that things we think of as givens today can be upended tomorrow — particularly in tech.
And particularly in AI. For years, LeCun was one of the standard-bearers in AI, which is why Zuckerberg hired him in 2013. Now all of the momentum in AI has moved away from his point of view and toward LLMs — particularly after OpenAI debuted ChatGPT three years ago, which kicked off a massive investment cycle in LLM infrastructure and talent. (If you think that has created an AI bubble, you’re not alone.)
I certainly don’t know if LeCun is correct. Maybe LLM boosters like Google’s Adam Brown, who was also onstage with LeCun Sunday, and who thinks that LLMs are going to create human-level intelligence, have got it right, after all.
But the fact that this science isn’t settled — and that new conventional wisdom around new tech could form quite quickly — certainly ought to give you pause. If the smartest people in AI can’t agree on what “smart” means, it’s going to be very hard to forecast how any of this shakes out.
