Exploring the Potential for Large Language Models to Demonstrate Rational Probabilistic Beliefs
This work addresses the problem of ensuring trustworthy and explainable performance in LLMs for information retrieval and decision systems, but it is incremental as it builds on prior research on reasoning and uncertainty in LLMs.
The study investigated whether large language models (LLMs) can represent rational probabilistic beliefs, finding that current models fail to adhere to fundamental properties of probabilistic reasoning, as demonstrated through a novel dataset and established uncertainty quantification techniques.
Advances in the general capabilities of large language models (LLMs) have led to their use for information retrieval, and as components in automated decision systems. A faithful representation of probabilistic reasoning in these models may be essential to ensure trustworthy, explainable and effective performance in these tasks. Despite previous work suggesting that LLMs can perform complex reasoning and well-calibrated uncertainty quantification, we find that current versions of this class of model lack the ability to provide rational and coherent representations of probabilistic beliefs. To demonstrate this, we introduce a novel dataset of claims with indeterminate truth values and apply a number of well-established techniques for uncertainty quantification to measure the ability of LLM's to adhere to fundamental properties of probabilistic reasoning.