BayesBench: Evaluating LLM Belief Trajectories Under Multi-Turn Evidence Accumulation
Quick Answer
BayesBench evaluates LLMs' belief updates in multi-turn conversations, revealing that while scaling improves latent inference, it doesn't consistently enhance downstream predictions.
Quick Take
The study assesses seven (3B-70B) across Bayesian tasks, highlighting a gap between inferring latent structures and rational belief updates.
Key Points
- BayesBench introduces three tasks: Bayesian estimation, prediction, and latent-framed prediction.
- Seven LLMs (3B-70B) were tested for belief updates in multi-turn settings.
- Scaling improves latent inference, with updates sometimes matching Bayesian posteriors.
- Downstream predictions show a gap between inferring latent structures and belief updates.
- The evaluation method shifts focus from final-turn answers to belief evolution.
Paper Resources
📖 Reader Mode
~2 min readAbstract: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.
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2606.30850 [cs.AI] |
| (or arXiv:2606.30850v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2606.30850 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Ankur Samanta [view email]
[v1]
Mon, 29 Jun 2026 19:30:32 UTC (852 KB)
— Originally published at arxiv.org
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.AI
See more →HOBA: Hierarchical On-Policy Bidding Agents for Adaptive Online Advertising
HOBA (Hierarchical On-policy Bidding Agents) is a novel hierarchical reinforcement learning framework that enhances online advertising bidding systems by improving adaptability and reducing hyperparameter tuning costs. It utilizes a for hyperparameter inference, a SARSA agent for expert model selection, and a dynamic expert pool for bid execution, achieving a +3.6% increase in target cost during large-scale deployment and outperforming state-of-the-art baselines on AuctionNet.