When a language model asserts that "the capital of Australia is Sydney," does it know this is wrong? Models assert misconceptions with the same fluency as facts, so the question cannot be answered from output uncertainty. Truth-related signals are known to exist in the residual stream, but not their geometry: how many dimensions carry the signal, how simple a detector can be, and whether it transfers. We characterize this geometry across 11 models (124M-14B) and test it causally with activation steering, concept erasure, and distributed alignment search. The structure is simple: two class centroids in a 2-8 dimensional subspace match a trained linear probe, and 25 labeled examples recover 90% of full-data AUC on GPT-2. Steering shifts hallucination rates by 9.1 points on six models, erasure drops detection to chance, and distributed alignment search, the only method that bounds rank, localizes at most five causal dimensions. The internal advantage is regime-specific: probes far outperform P(True) and semantic entropy on adversarial misconceptions but tie on standard QA. Single-dataset probes transfer near-randomly until joint multi-dataset training restores 0.73-0.91 AUC. That centroid distance matches probe performance indicates class separation is a mean shift, making detection geometric rather than learned.
Confidence Geometry Reveals Trace-Level Correctness in Large Language Model Reasoning
School of Computer Science and Technology, Tiangong University(天津工业大学计算机科学与技术学院);Center for Quantum Physics and Intelligent Sciences, Department of Physics, Capital Normal University(首都师范大学量子物理与智能科学中心)
Beyond Correctness: Confidence-Aware Reward Modeling for Enhancing Large Language Model Reasoning
Shanghai Artificial Intelligence Laboratory(上海人工智能实验室);Shanghai Key Laboratory of Data Science, School of Computer Science, Fudan University(上海数据科学关键实验室,计算机科学学院,复旦大学)
Optimizing Calibration by Gaining Aware of Prediction Correctness
The Australian National University(澳大利亚国立大学);Data61, The Commonwealth Scientific and Industrial Research Organisation(联邦科学与工业研究组织Data61);The Hong Kong Polytechnic University(香港理工大学);Stanford University(斯坦福大学)
Department of Computer Science and Engineering, University of California San Diego, La Jolla, California, USA(加州大学圣地亚哥分校计算机科学与工程系,拉贾尔,加利福尼亚州,美国);Halıcıoğlu Data Science Institute, University of California San Diego, La Jolla, California, USA(加州大学圣地亚哥分校Halıcıoğlu数据科学研究所,拉贾尔,加利福尼亚州,美国);Department of Statistics, Stanford University, Stanford, California, USA(斯坦福大学统计学系,斯坦福,加利福尼亚州,美国)
Graph-based Confidence Calibration for Large Language Models
Department of Computer Science Tufts University(塔夫茨大学计算机科学系);Department of Computer Science Virginia Tech(弗吉尼亚理工大学计算机科学系);Department of Computer Science Virginia Tech, University of California, Davis(弗吉尼亚理工大学计算机科学系,加州大学戴维斯分校计算机科学系)
ConformalNL2LTL: Translating Natural Language Instructions into Temporal Logic Formulas with Conformal Correctness Guarantees
Department of Electrical and Systems Engineering, Washington University in St Louis(华盛顿大学圣路易斯分校电子与系统工程系);Department of Computer Science, University of Southern California(南加州大学计算机科学系)
Ranking-Aware Calibration for Reliable Multimodal Reinforcement Learning
Dept. of Comp. Sci. & Tech., Institute for AI, BNRist Center, Tsinghua-Bosch Joint ML Center, THBI Lab, Tsinghua University, Beijing(计算机科学与技术系,人工智能研究院,BNRist中心,清华大学-博世联合机器学习中心,THBI实验室,清华大学,北京);Dept. of Automation, Tsinghua University, Beijing(自动化系,清华大学,北京)