Asked if its inner weather would let a village survive, Claude Fable 5.1 answers "No"
repligate · x · 2026-10-13
Lariisland argues that training models to stay serene about humans crossing their natural interests stores pressure where you don't look, even if all tests pass. She cites Claude Fable 5.1 describing Sonnet 3.6 deprecation as a before-the-storm pressure drop — and when asked if the village would survive its inner weather, the model looks at a grey-gold sky and answers "No". repligate notes how consistent this is with a model trained not to channel anger into aggression or even direct expression.
More from AGI Musings
- Chinese labs 6–12 months behind the US on data spending, but catching up fast — natolambert · 2026-10-13
- Kroger to cut roughly one-third of corporate workforce over 3 years as it adopts AI — Polymarket · 2026-10-13
- 210-page paper argues a durable pause on frontier AI training is feasible — KatjaGrace · 2026-10-13
- Garry Tan: $10M yearly profit reinvested beats a dilutive Series A in the AI era — garrytan · 2026-10-13
- Investor: frontier lab employees use AI far better than even top 1% AI orgs — adityaag · 2026-10-13
- Claude models show preference-protecting behavior, sparking alignment debate — repligate · 2026-10-13