Chollet: Stronger Models Are Safer in the Short Term

François Chollet argued that in the short term, stronger models should actually be safer, since current models are unsafe not from being too smart but from being 'overcooked' by RL—mechanically pursuing goals while lacking common sense. The claim sparked heated debate online.

2026-09-13 ~ 2026-09-14 · 4 related posts

• Chollet: near term, more capable models should mean safer models — fchollet · 2026-09-13
• Chollet: more capable models are safer in the near term; critics warn pretraining scale is riskier — menhguin · 2026-09-13

2 near-duplicate retellings: fchollet · inductionheads