Verbalizing LLM's Higher-order Uncertainty via Imprecise Probabilities

Researchers propose new uncertainty elicitation techniques for large language models using imprecise probabilities framework to better capture higher-order uncertainty. The approach addresses systematic failures in ambiguous question-answering and self-reflection by quantifying both first-order uncertainty over responses and second-order uncertainty about the probability model itself.

• →Current uncertainty elicitation techniques for LLMs fail systematically in ambiguous scenarios and self-reflection tasks.
• →The research introduces imprecise probabilities framework to capture both first-order and second-order uncertainty in LLMs.
• →New prompt-based techniques directly elicit and quantify multiple orders of uncertainty from language models.
• →The approach aims to improve credibility and support better downstream decision-making from LLM outputs.
• →The method addresses fundamental limitations in classical probabilistic uncertainty frameworks for LLMs.