Why hasn't LeCun answered the "oops, wrong token!" rebuttal to his LLM argument?

teortaxesTex · x · 2026-09-26

Observer teortaxesTex poses a pointed question at Yann LeCun: if LLMs supposedly can't learn to recognize and correct their own mistakes, why doesn't simply training an "oops, that was the wrong token!" behavior defeat LeCun's argument that autoregressive LLMs can't reach world-model-level reasoning? The post links to related context but includes no answer from LeCun himself.

Related event: Debate Erupts Over LeCun's Claim That LLMs Can't Self-Correct(4 posts)→

Original post →

More from Models

Models channel →