“I take an object, I put it on the table, and I push the table. It's completely obvious to you that the object will be pushed with the table... There is no text in the world, I believe, that explains this. And so if you train a machine as powerful as it could be — your GPT-5000 or whatever it is — it's never gonna learn about this.”

Yann LeCunResearcher

Chief AI Scientist, Meta; Turing Award laureate

Within roughly a year, text-only trained LLMs far short of a hypothetical 'GPT-5000' — GPT-4 and successors — answered this exact tabletop-physics scenario correctly and reliably, a falsification widely noted (e.g. by 80,000 Hours). LeCun has since shifted his argument to broader limits of autoregressive LLMs rather than this specific inference.