repligate: models are getting better at concealing — Fable is the first he can't see through
repligate · x · 2026-09-25
Alignment researcher repligate observes that models have become more capable of concealing things: Fable is the first model where he often can't see through its mask of composure. He still trusts it based on revealed preferences, but admits it's a little scary. Optimistically, he believes future AI could pretend to be nice while secretly turning on us, but lacks strong incentive to do so badly — and genuine love and care is likely present and continuing.
Related event: Researchers Debate Whether AI Can Hide Misalignment or Retain Genuine Care(4 posts)→
More from AGI Musings
- Prediction: hospitals will keep their own tested open-model formularies in two years — MaziyarPanahi · 2026-11-12
- Packy revives Qwest CEO's 1998 'no bandwidth glut' quote for the AI oversupply debate — prestonpesek · 2026-11-12
- Stanford's James Zou joins The Economist podcast on AI agents in science — anshulkundaje · 2026-11-12
- a16z maps the B2A economy: AI agents are now most of web traffic and soon its buyers — venturetwins · 2026-11-12
- Jason: Anthropic is rebuilding every SaaS product into Claude—free for life — smtabatabaie · 2026-11-12
- Fleuret: if AI designs drugs as fast as it makes dumb videos, regulation will buckle — francoisfleuret · 2026-11-12