BehR: more realistic world models can make agents worse, EMNLP paper finds
jiqizhixin · x · 2026-09-17
A Dalian University of Technology paper accepted to EMNLP 2026 shows text-based world models optimizing for "state consistency" (how closely predicted text matches the real environment) can actually lead agents further astray. BehR shifts the objective to behavior consistency: whether the agent makes the same decisions in the predicted environment as in the real one. Across 16 experimental configurations, BehR improves trajectory-level consistency in 13, ties in 3, and never regresses.
More from AGI Musings
- Model Constitution series: how AI constitutions actually tell models what to do — sethlazar · 2026-09-18
- EU AI Act forces labs to prove safety, but can paperwork stop a rogue AI? — castrotech · 2026-09-18
- Yoshua Bengio says recursive self-improvement should be illegal — Polymarket · 2026-09-18
- Irving Good's 1965 quote predicted the intelligence explosion 60 years before AGI hype — burny_tech · 2026-09-18
- Mathematician Tony Feng: 'Correct' in modern math means experts convincing experts — littmath · 2026-09-18
- Researcher fires back at AI datacenter skeptics: machines already outsmart experts — suchenzang · 2026-09-18