LeCun: Training LLMs on human knowledge yields memory, not intelligence
haider1 · x · 2026-11-22
Yann LeCun argues that training an LLM on human-produced knowledge only teaches it to accumulate what's already known — that's memory, not intelligence; rote learning rather than reasoning, invention, creativity or adaptivity, and not even the kind of intelligence seen in animals. Consistent with his long-standing critique of autoregressive LLMs in favor of world-model approaches.
More from AGI Musings
- Gary Marcus hails 'centaur math' milestone as Lean certifies human+AI proof — GaryMarcus · 2026-11-22
- Yoav Goldberg: AI safety community must publicly distance itself from its most extreme voices — yoavgo · 2026-11-22
- Gary Marcus: LLMs work better because they borrowed tree search from symbolic AI — GaryMarcus · 2026-11-22
- A mathematician despairs when the machine solves it first; an engineer moves on — yunta_tsai · 2026-11-22
- Gary Marcus: universal approximation doesn't mean networks can learn — GaryMarcus · 2026-11-22
- David Deutsch: Explanation, not proof, is the point of pure mathematics — TacoCohen · 2026-11-22