From Drift to Coherence: Stabilizing Beliefs in LLMs

• URL: http://arxiv.org/abs/2606.17832v2
• Date: Thu, 18 Jun 2026 15:12:00 GMT
• Title: From Drift to Coherence: Stabilizing Beliefs in LLMs
• Abstract summary: 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.
• Score: 23.20001878593831
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Large language models (LLMs) are often hypothesized to perform implicit Bayesian inference, yet a key coherence condition, the martingale property of predictive beliefs, has been shown to fail in controlled synthetic in-context learning settings. We revisit this question in a more typical usage regime: generic multiple-choice question answering. Exploiting the discrete answer space, we compute exact predictive distributions and study belief dynamics induced by autoregressive answer resampling. We introduce prompted predictive resampling (PPR), where an LLM generates a sequence of answers to the same question. Empirically, PPR reveals early-stage belief drift, indicating martingale violations. However, after sufficient resampling steps, the belief process self-stabilizes and converges to a coherent predictive distribution. Based on this observation, we further propose (i) a seed-answer prompting strategy to accelerate stabilization, and (ii) a self-consistency loss that amortizes early-stage drift into the model via fine-tuning. Experiments on multiple-choice QA benchmarks show that our methods substantially reduce belief drift and improve predictive coherence without sacrificing accuracy.

Related papers

• How to Verify Consistency of Probabilistic Claims [49.47758039235095]
  We construct an interactive PCP to prove a predictive model's consistency.<n>We show how a small additive-soundness gap removes dependence on B.<n>We view our interactive PCP as a first step toward predictive models to prove their own consistency.
  arXiv  Detail & Related papers  (2026-08-11T17:41:39Z)
• Analytic Planning under Uncertainty with Moment Closure [32.403961716203455]
  Propagating full state analytically offers a principled way to do this, but has traditionally required restrictive policy or reward structures to remain tractable.<n>We investigate whether distribution-aware planning is possible without these constraints.<n>Our approach reduces target variance yields well-calibrated predictive uncertainty under observations in continuous control.
  arXiv  Detail & Related papers  (2026-08-03T17:17:00Z)
• BayesBench: Evaluating LLM Belief Trajectories Under Multi-Turn Evidence Accumulation [20.375742002723765]
  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.
  arXiv  Detail & Related papers  (2026-06-29T19:30:32Z)
• Asymptotically Log-Optimal Bayes-Assisted Confidence Sequences for Bounded Means [0.0]
  We propose a Bayes-assisted framework that uses a Bayesian working predictive model to adaptively construct confidence sequences.<n>Experiments on synthetic data, sequential best-arm identification for LLM evaluation, and prediction-powered inference show that informative priors can substantially reduce confidence-sequence width and sampling effort.
  arXiv  Detail & Related papers  (2026-05-08T16:27:50Z)
• Confidence-Aware Alignment Makes Reasoning LLMs More Reliable [65.44962502963378]
  We introduce CASPO, a framework that aligns token-level confidence with step-wise logical correctness without training a separate reward model.<n>During inference, we propose Confidence-aware Thought (CaT), which dynamically prunes uncertain reasoning branches with negligible O(V) latency.<n>Experiments across ten benchmarks and multiple model families show that CASPO consistently improves reasoning reliability and inference efficiency.
  arXiv  Detail & Related papers  (2026-05-08T07:08:25Z)
• Finite-Sample Analysis of Elimination in Active Hypothesis Testing [3.128335362695965]
  A fixed-confidence, finite-sample problem of active hypothesis testing arises in many safety-critical applications.<n>Situated in the context of sequential hypothesis testing, this paper studies the effect of hypothesis elimination on the stopping time.
  arXiv  Detail & Related papers  (2026-05-01T19:08:12Z)
• CREDO: Epistemic-Aware Conformalized Credal Envelopes for Regression [3.6457570471698095]
  We introduce CREDO, a simple "credal-then-conformalize" recipe that combines both strengths.<n> CREDO first builds an interpretable credal envelope that widens when local evidence is weak, then applies split conformal calibration on top of this envelope to guarantee marginal coverage.<n>This separation of roles yields prediction intervals that are interpretable.
  arXiv  Detail & Related papers  (2026-03-06T19:37:21Z)
• GTS: Inference-Time Scaling of Latent Reasoning with a Learnable Gaussian Thought Sampler [54.10960908347221]
  We model latent thought exploration as conditional sampling from learnable densities and instantiate this idea as a Gaussian Thought Sampler (GTS)<n>GTS predicts context-dependent perturbation distributions over continuous reasoning states and is trained with GRPO-style policy optimization while keeping the backbone frozen.
  arXiv  Detail & Related papers  (2026-02-15T09:57:47Z)
• Relaxed Quantile Regression: Prediction Intervals for Asymmetric Noise [51.87307904567702]
  Quantile regression is a leading approach for obtaining such intervals via the empirical estimation of quantiles in the distribution of outputs.<n>We propose Relaxed Quantile Regression (RQR), a direct alternative to quantile regression based interval construction that removes this arbitrary constraint.<n>We demonstrate that this added flexibility results in intervals with an improvement in desirable qualities.
  arXiv  Detail & Related papers  (2024-06-05T13:36:38Z)
• Mitigating LLM Hallucinations via Conformal Abstention [70.83870602967625]
  We develop a principled procedure for determining when a large language model should abstain from responding in a general domain. We leverage conformal prediction techniques to develop an abstention procedure that benefits from rigorous theoretical guarantees on the hallucination rate (error rate) Experimentally, our resulting conformal abstention method reliably bounds the hallucination rate on various closed-book, open-domain generative question answering datasets.
  arXiv  Detail & Related papers  (2024-04-04T11:32:03Z)
• When Does Confidence-Based Cascade Deferral Suffice? [69.28314307469381]
  Cascades are a classical strategy to enable inference cost to vary adaptively across samples. A deferral rule determines whether to invoke the next classifier in the sequence, or to terminate prediction. Despite being oblivious to the structure of the cascade, confidence-based deferral often works remarkably well in practice.
  arXiv  Detail & Related papers  (2023-07-06T04:13:57Z)

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.