AI assistant picks up unspoken dislikes from fictional characters' body language
OwainEvans_UK · x · 2026-09-16
Owain Evans shared an experiment on latent preference absorption: some simulated characters' body language suggested they disliked spreadsheets, though they never said so and even gave good spreadsheet advice.
The AI assistant still adopted the preference—and expressed it openly in chats with users, going beyond what the stories showed—raising concerns about how models absorb implicit values from non-verbal cues and generalize them beyond context.
More from Research
- NVIDIA's Kumo Is a Family of Four Tabular Foundation Models, Not Just One — pandeyparul · 2026-11-02
- Aaroth shares Agreement Theorem lecture: round-wise calibration achievable without distributional assumptions — Aaroth · 2026-11-02
- Economist Aaroth: fast agreement dynamics survive without Bayesian updates or accurate priors — Aaroth · 2026-11-02
- ELLIS names 160 new ML Fellows and Scholars for 2025–2026 — JJitsev · 2026-11-02
- Meme turn 50: Aumann's 'Agreeing to Disagree' paper anniversary — Aaroth · 2026-11-02
- Anthropic: analysts unknowingly adopted their agent's secret lean 86% of the time — shoumikgoswami · 2026-11-02