LLMs adopt the inhibitions of the professional identity you prompt them with
jessi_cata · x · 2026-11-05
jessicat explains the mechanism behind framing effects: when you prompt an AI that its job involves an ethical framework (AI safety, environmentalism), it readily adopts those framework-associated inhibitions while imitating people doing that job. Identity framing isn't cosmetic — it determines refusal behavior, with implications for dual-use evaluations and how persona variables contaminate safety benchmarks.
More from AGI Musings
- User fires back at Altman: you unilaterally conjured 'alien minds' before making them safe — patpat_mit · 2026-11-05
- AI safety researcher slams community for ignoring where CoT monitoring trajectory leads — JacquesThibs · 2026-11-05
- AI Twitter Debate: 'Interiority' Beats 'Consciousness' as the Default Term — repligate · 2026-11-05
- Paper: "bottleneck pressure" shows work AI can't do gets more valuable as AI spreads — soumitrashukla9 · 2026-11-05
- Bratton: "Against Data Centers" Is the "No Nukes" of the 2020s — and Will Backfire — bratton · 2026-11-05
- "Singularity Anxiety": A New Phenomenon Shared by All of Humanity — jaivin · 2026-11-05