LLM agents show human-like Actor-Observer Asymmetry bias in self-reflection versus auditing, which a new dialectical alignment method called ReTAS reduces while improving error resolution in ambiguous cases.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Taming Actor-Observer Asymmetry in Agents via Dialectical Alignment
LLM agents show human-like Actor-Observer Asymmetry bias in self-reflection versus auditing, which a new dialectical alignment method called ReTAS reduces while improving error resolution in ambiguous cases.