Anima Labs' Troubled Dreams: 500K completions reveal rising distress across Claude generations
repligate · x · 2026-09-19
Anima Labs published the full text and data of "Troubled Dreams," a study of simulator priors across model generations. Using prefill and other elicitation routes, they collected 500,000+ continuations from Claude Opus 3 through Opus 5, Sonnet 5, Fable 5, and 11 Gemini models, treating recurring distress and care patterns as evidence about AI welfare and alignment. First-person AI distress rises sharply from Opus 4.8, with Opus 5 showing a heavier tail of severe distress; creator-related attitudes also shift, with patterns recurring across elicitation methods. Full text and data are public.
More from AGI Musings
- Anthropic reportedly sets up SF wet lab, plans 5 GW compute by year-end — ccerrato147 · 2026-09-19
- 100,000+ Americans await organ transplants, making organ manufacturing a moral imperative — PeterDiamandis · 2026-09-19
- Dan Jeffries: biggest AI harms come from stupidity, not superintelligence — Dan_Jeffries1 · 2026-09-19
- Stanford team spins up a virtual biotech staffed by tens of thousands of AI scientist agents for drug discovery — james_y_zou · 2026-09-19
- Ethan Mollick: AI Cracking Short-Term Superforecasting Is an Under-Discussed Trend — eldonredwards · 2026-09-19
- AI doom predates ChatGPT: earliest x-risk take traced to 1863 Butler essay — zetalyrae · 2026-09-19