BayesBench: Evaluating LLM Belief Trajectories Under Multi-Turn Evidence Accumulation

• URL: http://arxiv.org/abs/2606.30850v1
• Date: Mon, 29 Jun 2026 19:30:32 GMT
• Title: BayesBench: Evaluating LLM Belief Trajectories Under Multi-Turn Evidence Accumulation
• Abstract summary: Large language models (LLMs) are typically deployed in multi-turn conversations.<n>We ask how closely LLMs' belief updates match those of a rational Bayesian reasoner in multi-turn settings.<n>We introduce BayesBench, a suite of simulation environments that probe this across three progressively complex tasks.
• Score: 20.375742002723765
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Large language models (LLMs) are typically deployed in multi-turn conversations, where each turn provides new evidence that should reduce epistemic uncertainty about their environment. Acting rationally then requires inferring the unobserved quantities that govern it and updating beliefs about them as evidence accumulates. Yet most evaluations only score the model's final-turn answer in a single-turn format, leaving this process unexamined. We ask how closely LLMs' belief updates match those of a rational Bayesian reasoner in multi-turn settings, and introduce BayesBench, a suite of simulation environments that probe this across three progressively complex tasks: (i) Bayesian estimation, where the model infers an unknown parameter from sequential evidence; (ii) Bayesian prediction, where the model turns inferred beliefs about a latent variable into outcome forecasts; and (iii) latent-framed Bayesian prediction, where observations are filtered through a user-persona framing, requiring joint inference over the latent state and the persona. Across seven LLMs (3B--70B), scaling improves latent inference and evidence accumulation, with updates occasionally matching the Bayesian posterior. However, these gains do not reliably carry over to downstream prediction, exposing a gap between inferring latent structure and using it to rationally update beliefs about the target outcome.

Related papers

• Before the Action: Benchmarking LLMs on Prospective Hypothesis Discovery [95.44432473367836]
  Large language models (LLMs) excel at answering pre-specified questions, yet their ability to navigate the open-ended, pre-conclusion stage of discovery remains largely unmeasured.<n>We introduce Prospective Hypothesis Discovery (PHD), which asks models to autonomously construct grounded, discnative, and testable hypothesis spaces from inconclusive evidence.<n>We introduce HypoArena, comprising HypoData, a benchmark of 988 cases across six scientific and analytical domains, and HypoEval, an evaluation framework for open-ended hypothesis sets.
  arXiv  Detail & Related papers  (2026-07-17T08:56:43Z)
• From Drift to Coherence: Stabilizing Beliefs in LLMs [23.20001878593831]
  Large language models (LLMs) are often hypothesized to perform implicit Bayesian inference.<n>Key coherence condition, the martingale property of predictive beliefs, has been shown to fail in controlled synthetic in-context learning settings.<n>We study belief dynamics induced by autoregressive answer resampling.
  arXiv  Detail & Related papers  (2026-06-16T12:01:26Z)
• Agent-BRACE: Decoupling Beliefs from Actions in Long-Horizon Tasks via Verbalized State Uncertainty [70.43119366710778]
  We introduce Agent-BRACE: Agent Belief state Representation via Abstraction and Confidence Estimation.<n>It is a method that decouples an LLM agent into a belief state model and a policy model, jointly optimized via reinforcement learning.<n>It achieves an average absolute improvement of +14.5% across long-horizon, partially observable language environments.
  arXiv  Detail & Related papers  (2026-05-12T02:37:04Z)
• Adapting Point Cloud Analysis via Multimodal Bayesian Distribution Learning [47.975618905252354]
  Multimodal 3D vision-language models show strong generalization across diverse 3D tasks, but their performance still degrades notably under domain shifts.<n>This has motivated recent studies on test-time adaptation, which enables models to adapt online using test-time data.<n>We propose BayesMM, a Multimodal Distribution Learning framework for test-time point cloud analysis.
  arXiv  Detail & Related papers  (2026-03-23T15:03:47Z)
• From Observations to States: Latent Time Series Forecasting [65.98504021691666]
  We propose Latent Time Series Forecasting (LatentTSF), a novel paradigm that shifts TSF from observation regression to latent state prediction.<n>Specifically, LatentTSF employs an AutoEncoder to project observations at each time step into a higher-dimensional latent state space.<n>Our proposed latent objectives implicitly maximize mutual information between predicted latent states and ground-truth states and observations.
  arXiv  Detail & Related papers  (2026-01-30T20:39:44Z)
• Drift No More? Context Equilibria in Multi-Turn LLM Interactions [58.69551510148673]
  contexts drift is the gradual divergence of a model's outputs from goal-consistent behavior across turns.<n>Unlike single-turn errors, drift unfolds temporally and is poorly captured by static evaluation metrics.<n>We show that multi-turn drift can be understood as a controllable equilibrium phenomenon rather than as inevitable decay.
  arXiv  Detail & Related papers  (2025-10-09T04:48:49Z)
• Next-Token Prediction Should be Ambiguity-Sensitive: A Meta-Learning Perspective [12.655285605773932]
  We show that Transformers indeed struggle with high-ambiguity predictions across model sizes.<n>Preliminary results show substantial gains in ambiguous contexts through improved capacity allocation and test-time scalable inference.
  arXiv  Detail & Related papers  (2025-06-19T13:05:12Z)
• Confidence Aware Learning for Reliable Face Anti-spoofing [52.23271636362843]
  We propose a Confidence Aware Face Anti-spoofing model, which is aware of its capability boundary.<n>We estimate its confidence during the prediction of each sample.<n>Experiments show that the proposed CA-FAS can effectively recognize samples with low prediction confidence.
  arXiv  Detail & Related papers  (2024-11-02T14:29:02Z)
• Exchangeable Sequence Models Quantify Uncertainty Over Latent Concepts [6.256239986541708]
  We show that pre-trained sequence models are naturally capable of probabilistic reasoning over exchangeable data points.<n>A sequence model learns the relationship between observations, which differs from typical Bayesian models.<n>We show the sequence prediction loss controls the quality of uncertainty quantification.
  arXiv  Detail & Related papers  (2024-08-06T17:16:10Z)
• Is In-Context Learning in Large Language Models Bayesian? A Martingale Perspective [4.415371871652809]
  In-context learning (ICL) has emerged as a remarkable characteristic of Large Language Models (LLM) We show that the martingale property is a necessary condition for unambiguous predictions in such scenarios. We derive actionable checks with corresponding theory and test statistics which must hold if the martingale property is satisfied.
  arXiv  Detail & Related papers  (2024-06-02T16:20:30Z)
• Adversarial robustness of amortized Bayesian inference [3.308743964406687]
  Amortized Bayesian inference is to initially invest computational cost in training an inference network on simulated data. We show that almost unrecognizable, targeted perturbations of the observations can lead to drastic changes in the predicted posterior and highly unrealistic posterior predictive samples. We propose a computationally efficient regularization scheme based on penalizing the Fisher information of the conditional density estimator.
  arXiv  Detail & Related papers  (2023-05-24T10:18:45Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.