Tracking how AI models change their answers reveals reasoning patterns
Answer-Distribution Trajectories: A Stochastic-Dynamics View of LLM Reasoning
Artificial IntelligenceComputation and LanguageInformation TheoryMachine Learning
Summary
Sometimes AI language models give answers after a series of thought steps, but we usually only look at whether the final answer is right or wrong. The authors created a method that watches how the model’s guesses change during this thinking process instead of just checking the last answer. This way, they can see if the model was exploring different ideas, changing its mind, or committed confidently to one answer. Their approach helps understand why some answers succeed or fail and shows differences in how various models think over time. This matters because it gives a clearer picture of AI reasoning beyond just final accuracy.
large language modelschain-of-thought reasoningstochastic dynamicsanswer distributionentropyreasoning processpredictive distributionmodel uncertaintybenchmark evaluation
Authors
Mar Gonzàlez I Català, Haitz Sáez de Ocáriz Borde, Davide Murari, Carola-Bibiane Schönlieb, Pietro Liò, George Montañez
Abstract
Chain-of-thought reasoning provides a structured computation between a model's input and final answer. Yet it is often evaluated through endpoint accuracy, which ignores the path taken to reach that answer. An emerging line of work addresses this limitation using entropy profiles, which track how uncertainty evolves over the reasoning process but do not reveal which competing hypotheses account for that uncertainty. We introduce answer-distribution trajectories, a stochastic-dynamics-inspired representation that tracks the model's full predictive distribution over answers as reasoning unfolds. As a strictly finer representation than endpoint and entropy summaries, answer-distribution trajectories enable us to characterize a trace through a dynamical reasoning profile spanning exploration, revision, motion, and commitment, and to distinguish different dynamical mechanisms of reasoning success and failure. Across sixteen open-weight language models and four reasoning benchmarks, we show that traces with the same endpoint and similar entropy profiles can exhibit substantially different reasoning dynamics. We further find substantial variation in these dynamics both within and across models and tasks, with different objectives favoring different dynamical profiles. Additionally, we show that training and inference choices systematically reshape these profiles. Our results suggest that answer-distribution trajectories provide a rich framework for analysing and evaluating the dynamics of LLM reasoning.