Researchers Question Whether Agent Behavior Matches Language
Thomas Dietterich and others argue that agents' behavior should be tested against their language, questioning whether chain-of-thought reliably predicts actions and proposing language interventions to verify causality.
2026-09-01 ~ 2026-09-01 · 2 related posts
- Dietterich: We must test if agent behavior matches their language, not just assume CoT implies causality — tdietterich · 2026-09-01
- Does intervening on agent language change behavior? METR study blocked — NeelNanda5 · 2026-09-01