Answer, Refuse, or Guess? Investigating Risk-Aware Decision Making in Language Models

Cheng-Kuang Wu, Zhi Rui Tam, Chieh-Yen Lin, Yun-Nung Chen, Hung-yi Lee · 2025

Read Answer, Refuse, or Guess? Investigating Risk-Aware Decision Making in Language Models by Cheng-Kuang Wu, Zhi Rui Tam, Chieh-Yen Lin, Yun-Nung Chen, Hung-yi Lee (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-01332.txt.

Description

Cheng-Kuang Wu, Zhi Rui Tam, and colleagues study when a language model should answer, refuse, or guess in a 2025 study. Agents built on these models act step by step toward goals, and some wrong steps carry severe costs. An agent that never defers will eventually blunder; one that always defers is useless, and the right balance depends on how bad a mistake would be.

The authors investigate risk-aware decision making, where the tendency to hold back should scale with the stakes. Calibrated deference is a safety property that benchmarks rarely grade, yet every deployed agent depends on it.

Catalog record

CREATORCheng-Kuang Wu, Zhi Rui Tam, Chieh-Yen Lin, Yun-Nung Chen, Hung-yi Lee
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE59 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-01332

Browse the full library · About the Tower of Babel