Uncertainty-Aware Generation and Decision-Making Under Ambiguity

This study evaluates uncertainty-aware decision-making algorithms based on Bayesian decision theory and risk-averse approaches for LLM tasks like tutoring and peer reviewing. The authors use conformal prediction to provide guarantees over strategies and scores, finding that these methods can improve generation utility but require careful implementation under high ambiguity.

• The work applies Bayesian decision theory and risk-averse decision-making to tutoring and automatic peer reviewing tasks.
• Conformal prediction is utilized to provide guarantees over tutoring strategies and review scores.
• Risk-averse rules can degrade performance by optimizing for generic outputs, whereas Bayesian methods tend to perform better.
• High ambiguity requires careful implementation of these algorithms to avoid negative impacts on utility.

The authors consider this important because it uses decision theory techniques to improve LLM-based decision-making and outlines open challenges for the community.