Answer, Refuse, or Guess? Investigating Risk-Aware Decision Making in Language Models
Published in arXiv, 2025
This study formalizes the task of risk-aware decision making in language models, examining how models should adapt their decisions (answer or refuse) based on contextual risk levels. The researchers identify key failure modes where LMs make irrational decisions in both high-risk and low-risk scenarios and propose skill decomposition solutions that separate downstream task performance, confidence estimation, and expected-value reasoning. Results show that prompt chaining significantly improves performance in high-risk settings across multiple models and datasets.
Recommended citation: Wu, C.K., Tam, Z.R., Lin, C.Y., Chen, Y.N., & Lee, H. (2024). “Answer, Refuse, or Guess? Investigating Risk-Aware Decision Making in Language Models.” arXiv preprint arXiv:2503.01332.
Recommended citation: Wu, C.K., Tam, Z.R., Lin, C.Y., Chen, Y.N., & Lee, H. (2024). "Answer, Refuse, or Guess? Investigating Risk-Aware Decision Making in Language Models." arXiv preprint arXiv:2503.01332.
Download Paper