The paper investigates how to improve confidence calibration for large language models (LLMs) used in automated code revision (ACR). It proposes applying local Platt-scaling to three fine-grained confidence scores, rather than the conventional global method, and demonstrates that this approach consistently reduces calibration error across multiple tasks, metrics, and model sizes. The study shows that fine-grained calibration, especially when combined with global scaling, yields more reliable confidence estimates for ACR tasks.
By Hong Yi Lin, Chunhua Liu, Haoyu Gao, Patanamon Thongtanunam, Christoph Treude
The paper introduces Counterfactual Fragility Certificates (CFC), a model‑agnostic audit protocol that maps each prediction to an evidence‑failure trajectory, summarizing it with metrics such as greedy flip budget, margin‑collapse area, degradation thresholds, and fragility dominance score. CFC is shown to identify brittle high‑confidence predictions on seven tabular benchmarks with an AUROC of 0.915, outperforming existing scalar scores by up to +0.405. The method remains effective across various perturbation and review‑budget scenarios, and can also inform fragility‑aware regularization and temperature correction.
By Filippo Cenacchi, Longbing Cao, Runze Yang
arXiv:2610.17747v1 Announce Type: cross
Abstract: Failure-attribution benchmarks ask an LLM judge which step or agent caused a failure. A high score should mean the judge checked the faulty step agai...
By Tianyang Xu, Jing Gao
The paper investigates how AI systems that perform best‑of‑n search require different validation strategies as the search width changes. It shows that auditing only small search widths leaves a gap in reliability estimates for larger widths, and proposes retaining candidate ranks and truth labels to estimate reliability across all widths up to N. The authors derive theoretical bounds on the minimax mean‑squared error, design procedures that achieve these bounds, and demonstrate that a shared audit can significantly reduce maximum error across many widths in practical CodeRM pools.
By Ricardo Fitas
The paper investigates how memory systems can answer a current query correctly yet fail to retain distinctions needed for later updates. Using a paired‑history audit, the authors evaluate 24 history pairs across six synthetic mechanisms and two model backends, achieving perfect reveal accuracy on DeepSeek and high accuracy on GLM. Record‑level audits reveal specific failures in structured reveal memories and frontier late‑reference adequacy, and the authors test a label‑equivariant repair that only partially restores correctness.
By Guangzhe Zhang
arXiv:2607. 18278v1 Announce Type: cross Abstract: Calibration is usually evaluated in aggregate, but the most dangerous failures are often local: predictions that remain highly confident despite being wrong.
By Filippo Cenacchi, Longbing Cao, Runze Yang
arXiv:2609.24194v1 Announce Type: new
Abstract: Evaluation scores used around LLM systems -- including reward models, rerankers, and LLM judges -- can track surface form instead of the quality they c...
By Daein Weon, Dongho Kang
arXiv:2609.07095v1 Announce Type: new
Abstract: LLM assistants often produce more answers than humans can review before users see them. Most evaluations ask whether an answer is wrong, unsupported, o...
By SangJin Park, Myungsub Choi, Jineok Kim, Minseung Kang
The paper introduces a risk‑controlled framework for using large language models (LLMs) as judges in tasks without reference answers. By calibrating uncertainty thresholds on a held‑out set, the method ensures that the false discovery rate of accepted verdicts stays below a user‑specified level α with high probability, using finite‑sample Clopper–Pearson intervals. When the parametric judge lacks confidence, the instance is routed to a retrieval‑augmented mode with a second calibrated threshold, preserving the error guarantee while achieving higher coverage than single‑mode baselines.
arXiv:2608. 02786v1 Announce Type: new Abstract: AI systems can fail silently.
By Priyanka Bajaj (Independent Researcher)
arXiv:2606. 31630v1 Announce Type: new Abstract: Language models increasingly write probabilistic programs (in NumPyro, Stan, or Pyro), but a program that compiles, runs, and passes every unit test can still be \emph{statistically} wrong -- a Gaussian likelihood for heavy-tailed data, a Poisson for over-dispersed counts, an invalid prior support, or a pathological parameterization.
By Jian Xu, Delu Zeng, John Paisley, Qibin Zhao
The paper introduces Alignment Forecasting, a method for predicting whether fine‑tuning a language model on a given dataset will increase specific alignment failures such as deception or sycophancy. It presents ALIGNMENTFORECASTBENCH, a benchmark of over 5,000 forecasting questions across many models, datasets, and failure modes, and shows that a simple forecasting scaffold using an LLM’s assessment of dataset bias can outperform baseline forecasters. The authors demonstrate that filtering out high‑risk training examples identified by the forecaster can improve alignment in multiple‑choice evaluations, though benefits in open‑ended conversations remain uncertain.
By Chen Yueh-Han, Bruce W. Lee, Ilia Sucholutsky, Tomek Korbak