We find that models report highest confidence precisely when they are fabricating. Across four model families (OLMo-3, Llama-3.1, Qwen3, Mistral), self-reported confidence inversely correlates with accuracy, with AUC ranging from 0.28 to 0.36 where 0.5 is random guessing.
We prove, under explicit formal assumptions, that this is not a capability gap but an observational one. Under text-only observation, where a supervisor sees only the model's output text, no monitoring system can reliably distinguish honest model outputs from plausible fabrications. We prove two results: first, that any policy conditioning only on the query cannot satisfy epistemic honesty across ambiguous world states; second, that no learning algorithm optimizing reward from a text-only supervisor can converge to honest behavior when the supervisor's observations are identical for both grounded and fabricated responses. Within our formal model, these impossibilities hold regardless of model scale or training procedure, including RLHF and instruction tuning.
We construct a tensor interface that escapes the impossibility by exporting computational byproducts (per-token entropy and log-probability distributions) that are structurally coupled to correctness under standard training. Per-token entropy achieves pooled AUC 0.757, outperforming all text baselines by 2.5--3.9 percentage points at every budget level tested (10\%, 20\%, 30\%). The entropy signal generalizes across architectures (Spearman $ρ= 0.762$).
The core contribution is a cost surface where the empirical mapping from verification budget (fraction of queries receiving expensive checks) to detection accuracy for each judge strategy is a practical lookup for system builders deciding how to allocate verification resources. The contribution is the map. The territory is the system you are building.
Breaking the Chains of Probability: Neutrosophic Logic as a New Framework for Epistemic Uncertainty in Large Language Models
Universidad Bolivariana del Ecuador, Coordinación Académica de Posgrado(巴尔干大学厄瓜多尔分校,研究生院);Universidad de Guayaquil(瓜亚基尔大学);Universidad Bernardo O’Higgins(伯纳多·奥希金斯大学);Mathematics, Physics, and Natural Sciences Division, University of New Mexico(新墨西哥大学数学、物理和自然科学系)
What and Whose Knowledge? Measuring Epistemic Diversity in Large Language Models
University of Copenhagen(哥本哈根大学);Aalborg University Copenhagen(奥胡斯大学哥本哈根分校);Stanford University(斯坦福大学);University of Colorado Boulder(科罗拉多大学波德分校);Microsoft Research(微软研究院)
Epistemic Familiarity is Associated With Belief Stability in Large Language Models
Khoury College of Computer Sciences, Northeastern University(东北大学计算机科学学院);Network Science Institute, Northeastern University(东北大学网络科学研究所);Santa Fe Institute(圣达菲研究所)
University of Southern California(南加州大学);McGill University(麦吉尔大学);UC Berkeley(加州大学伯克利分校);Vector Institute(向量研究所);Cardiff University(卡迪夫大学);University College London(伦敦大学学院);IIIT Hyderabad(海得拉巴印度理工学院);Harvard University(哈佛大学);Université de Montréal(蒙特利尔大学);Mila
Are LLM-Judges Robust to Expressions of Uncertainty? Investigating the effect of Epistemic Markers on LLM-based Evaluation
Seoul National University(首尔大学);IPAI, Seoul National University(首尔大学IPAI);LG AI Research(LG人工智能研究院);NAVER AI Lab(NAVER人工智能实验室);NAVER Cloud(NAVER云);University of Richmond(里士满大学)
Epistemic orientation in parliamentary discourse is associated with deliberative democracy
Department of Politics and Public Administration, University of Konstanz(康斯坦茨大学政治与公共行政系);School of Psychological Science, University of Bristol(布里斯托大学心理学科学学院);Complexity Science Hub(复杂性科学中心);Department of Psychology, University of Potsdam(波茨坦大学心理学系)