Skip to content
AI Atlas
PaperActive

LLMs Are Not (Consistently) Bayesian: Quantifying Internal (In)consistencies of LLMs’ Probabilistic Beliefs

Applearxiv.org/pdf/2605.06915

quality84

Updated 2 h ago · first seen 11 Sept 2026

paper_01M294AHM2Z9JAPGE6SY4ZN50T

Published
28 Aug 2026
T1 · 2 h ago
arXiv
2605.06915
T1 · 2 h ago

As of

Rewind the record: see this entity's attributes exactly as AI Atlas knew them on a given day.

Claim history

6 claims · 6 properties

Paperpaper_url1

Claim history for Paper
ValueValid from → toStatusSourceConfidenceExtractor
https://machinelearning.apple.com/research/llms-not-consistently-bayesiancurrentcurrentApple Machine Learning ResearchT1highdeterministic

Abstractabstract1

Claim history for Abstract
ValueValid from → toStatusSourceConfidenceExtractor
Modern AI systems are being deployed in complex domains such as medicine, science, and law, where there is often not a single correct answer given the observed evidence. Such systems must be able to represent and update uncertain beliefs about the world as new evidence arrives to make rational decisions. We introduce the novel technique of studying LLMs as information processing rules and utilize the information processing gap—the deviation from Bayes updates—to study the internal (in)consistencies of how LLMs update their probabilistic beliefs from evidence. Our extensive experiments evaluate…currentcurrentApple Machine Learning ResearchT1highdeterministic

arXiv idarxiv_id1

Claim history for arXiv id
ValueValid from → toStatusSourceConfidenceExtractor
2605.06915currentcurrentApple Machine Learning ResearchT1highdeterministic

Authorsauthors1

Claim history for Authors
ValueValid from → toStatusSourceConfidenceExtractor
Chacha Chen, Matthew Jörke, Adam Goliński, Masha Fedzechkina, Guillermo Sapiro, Sinead Williamson, Nicholas FoticurrentcurrentApple Machine Learning ResearchT1highdeterministic

PDFpdf_url1

Claim history for PDF
ValueValid from → toStatusSourceConfidenceExtractor
https://arxiv.org/pdf/2605.06915currentcurrentApple Machine Learning ResearchT1highdeterministic

Publishedpublished_at1

Claim history for Published
ValueValid from → toStatusSourceConfidenceExtractor
28 Aug 2026currentcurrentApple Machine Learning ResearchT1highdeterministic

Claims are temporal and append-only: a new observation closes the previous claim (valid_to) instead of overwriting it. Conflicting claims from different sources are kept side by side and flagged — never averaged. Methodology →