Yann LeCun says strong code generators are not just pure LLMs

suchenzang · x · 2026-08-04

A post pushing back on the habit of rewriting AI history to claim every system idea was obvious all along. The screenshot shows Yann LeCun clarifying that he was talking about pure autoregressive token prediction, and that strong code-generation systems are not just plain LLMs doing next-token prediction.

The broader point is that using Chinese open-source components inside systems is not some new revelation; the author argues it was always on the table, and the new framing is just a way of calling that path "neurosymbolic."

Related event: LeCun: Strong Code Generation Requires More Than Autoregressive LLMs(4 posts)→

Original post →

More from Models

Models channel →