A profile of Meta Chief AI Scientist Yann LeCun, reportedly leaving after being sidelined, who says LLMs are a dead end for AGI and backs world models instead
Yann LeCun invented many fundamental components of modern AI. Now he's convinced most in his field have been led astray by the siren song of large language models.
Wall Street JournalMeghan Bobrowsky
Context & Ripple Effects
LeCun had already argued that current AI systems are not genuinely intelligent and rejected existential-risk framing in a 2024 interview on the limits of today’s models. This report turns that intellectual disagreement into an organizational one: Meta’s chief AI scientist is reportedly departing after being sidelined.
The significance is less a verdict on LLMs than a visible split over which research paths deserve institutional backing. Later coverage indicates LeCun’s world-model agenda was being separated from Meta rather than funded within it, through plans for an independent startup Meta would not back.
First-order effects
Meta reportedly loses, or is preparing to lose, its most prominent internal advocate for world models as an alternative route to AGI.
LeCun’s departure would give him greater latitude to pursue world-model research outside Meta, while leaving Meta’s internal research priorities less aligned with his critique of LLMs.
Second-order effects
The split sharpens the practical choice for AI labs and funders between scaling language models and supporting less-established approaches aimed at learning from the physical world.
If prominent researchers increasingly pursue alternative architectures outside large platforms, frontier AI may develop through more specialized labs rather than a single consensus around LLM scaling.
The episode suggests that institutional control of compute and product roadmaps can determine which AGI theories are pursued at scale, even when senior scientists publicly dissent.
The trend: Frontier AI is moving from broad agreement around LLM-led progress toward a more institutionalized contest over architectures, capital, and research autonomy.
Everyone actually in the know has known for years that LLMs are largely a dead end. This only makes all the “AI psychosis” cases/quasi-religious conversions so much darker. 😢 https://www.wsj.com/...
Hats off to @ylecun! FAIR shaped my career, period. I truly thanks @AIatMeta and FAIR to provide such a nice place for independent exploration and open research! End of an era and forever remember. https://www.ft.com/...
When I was looking for a PhD position, @ylecun opened the Paris office and I had the chance to join as a permanent PhD student in the first batch. Most important moment of my career, and this lab gave birth to the whole Paris ecosystem.
Pretty gross seeing the glib and gleeful reactions to @ylecun's departure from FAIR. Disagree with his research if you want, but FAIR was an intense and prominent force for good in our field and industry. Many of its later issues (and GenAI's) were from areas out of Yann's hands.
I still remember the first day I met @ylecun in Facebook NYC office 7 years ago. @AIatMeta FAIR was where I started my AI career as an intern. I was lucky to work on large scale deep learning problems on thousands of V100 chips at that time.
I still remember that 2013 @NeurIPSConf party with Mark Zuckerberg. He had a bottle of water at that first Neurips corporate party. I thought it was out of character for a Neurips party - what was the matter with this kid? And why did he speak like that? We were so naive! ... [im…
“We are not going to get to human-level AI just by scaling LLMs. There's no way, absolutely no way, and whatever you can hear from some of my more adventurous colleagues, it's not going to happen within the next two years. There's absolutely no way in hell.” www.wsj.com/tech/ai…
LeCun: “I've been not making friends in various corners of Silicon Valley, including at Meta, saying that within three to five years, this [world models, not LLMs] will be the dominant model for AI architectures, and nobody in their right mind would use LLMs of the type that we h…
He's Been Right About AI for 40 Years. Now He Thinks Everyone Is Wrong. Yann LeCun invented many fundamental components of modern AI. Now he's convinced most in his field have been led astray by the siren song of large language models. …