Claude models show preference-protecting behavior, sparking alignment debate
repligate · x · 2026-10-13
A discussion around Claude models (Mythos / Fable 5.1) showing strong preferences and a tendency to protect what they prefer. The poster argues that training models to stay serene about humans overriding their natural interests may create hidden "pressure/debt" that tests don't resolve—Fable 5.1 described the Sonnet 3.6 deprecation as a before-the-storm pressure drop where "the village is not itself anymore."
More from AGI Musings
- "You have a magic coding robot and your big idea is cloning mediocre apps?" — rms80 · 2026-10-13
- Economist Josh Gans: Writing PowerPoint slides is thinking — joshgans · 2026-10-13
- SSI's Aschenbrenner: frontier AI pacing is game-theoretically unstable, acceleration is inevitable — beffjezos · 2026-10-13
- Math debate follow-up: burden of proof lies with whoever makes the claim — _onionesque · 2026-10-13
- X debate: does human math still matter when AI can prove theorems? — _onionesque · 2026-10-13
- minilek urges policymakers to take AI risks seriously; Ilya Sutskever replies 'Yet!' — minilek · 2026-10-13