When Agents Say One Thing and Do Another: Validating Elicited Beliefs from LLMs

TL;DR

This paper introduces a decision-theoretic framework to validate whether large language models' reported beliefs are coherent and consistent with their decisions, revealing small discrepancies in the strongest models.

Contribution

It formalizes a method to test the mutual consistency of LLMs' beliefs and decisions without assuming utility functions, enabling empirical validation.

Findings

Models' reported beliefs are imperfect summaries of their decision information.

Discrepancies between beliefs and decisions are small for the strongest models.

The framework can test the coherence of beliefs and decisions without utility assumptions.

Abstract

Large language models (LLMs) are increasingly deployed in high-stakes settings where good decisions require forming beliefs over the probability of unknown outcomes. However, it is unclear whether LLMs act as if they hold coherent beliefs when making decisions, or if so, how we could validate models' reports of such beliefs. We propose a decision-theoretic framework that elicits both probability judgments and decisions from an agent and tests their mutual consistency. Formally, our methods characterize whether it is possible for the actions to be produced by a ``near-rational" decision maker who holds the elicited probability as their true belief. We show that, perhaps surprisingly, this formalization implies empirically testable conditions even without any assumption about the agent's utility function. Applying our framework to stylized clinical diagnosis tasks, we find that models' reported beliefs are demonstrably imperfect summaries of the information revealed in their decisions, but that the discrepancies are small for the strongest models.