Counterfactual localization identifies points of deceptive commitment in language model reasoning traces by estimating deceptive outcome probabilities via resampling, revealing that attention-based features generalize across environments while lexical cues do not.
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
-
The Point of No Return: Counterfactual Localization of Deceptive Commitment in Language-Model Reasoning
Counterfactual localization identifies points of deceptive commitment in language model reasoning traces by estimating deceptive outcome probabilities via resampling, revealing that attention-based features generalize across environments while lexical cues do not.