The study investigates how varying the capability of a reviewer model in a large‑language‑model (LLM) execute‑review‑revise pipeline affects rejection decisions on 100 olympiad mathematics problems. A mid‑tier reviewer improves final accuracy by 12 percentage points (from 52 % to 64 %) without damaging answers, while a self‑reviewer detects errors best (85 % recall) but rejects too often and harms correct solutions. Below a certain capability threshold the reviewer becomes inert, changing none of the answers and doubling token cost.
By Faizan Tanveer
The paper investigates the reliability of ranking tables produced by small-sample evaluations of large language models (LLMs). Using LLM‑inferred prompt structure across eight model variants, the authors find that prompt‑structure recovery is highly unstable, with only the bottom of the ranking consistently reproducible. They demonstrate that standard evaluation practices can misrepresent model performance and propose reporting practices to improve transparency.
By Dipankar Sarkar
arXiv:2607. 08065v1 Announce Type: new Abstract: LLM-as-judge (Zheng et al.
By Kaihua Ding
arXiv:2606. 15887v1 Announce Type: cross Abstract: Large language model (LLM) systems are increasingly proposed to assist peer review, yet most evaluations judge the prose of machine-generated review text, not the validity of the numeric score a system assigns.
By Costa Georgantas
arXiv:2606. 09046v1 Announce Type: new Abstract: Useful audits reveal not only how often a model fails, but also where its failures concentrate.
By Vyzantinos Repantis, Ameya Gawde, Harshvardhan Singh
arXiv:2608. 08514v1 Announce Type: new Abstract: We independently reproduce two recent methods for making large language model (LLM) reasoning more reliable, and stress-test them across domains and models (RPC across four new task domains with Qwen3-8B, LCF across four 7-8B models).
By Minhan Cho, Jimin Kweon
arXiv:2606. 27383v1 Announce Type: cross Abstract: Large language models (LLMs) are increasingly used as research assistants, yet it remains unclear whether they can calibrate research takeaways to the strength and scope of the supporting evidence.
By Yu Fu, Yongqi Kang, Yong Zhao
The paper introduces Janus, a method for validating error patterns in language models by comparing error rates across predefined yes/no properties and using shuffled decoy labels to set significance thresholds. Janus requires that a pattern’s error difference surpasses the decoy-derived threshold and is replicated on held‑out data before reporting. Experiments on a controlled code‑finding task confirm several meaningful error patterns, while on MuSiQue and LongBench v2 Janus reports no confirmed patterns for the tested properties, contrasting with standard shuffling tests that sometimes confirm patterns.
By Vyzantinos Repantis, Ameya Gawde, Harshvardhan Singh
arXiv:2607. 00276v1 Announce Type: cross Abstract: Current large-language-model (LLM) physics benchmarks are usually scored by answer accuracy, which cannot distinguish genuine reasoning from recall of familiar problem patterns and reveals little about where a model's reasoning breaks down.
By Dong Zhang
Agent evaluations tell us that a model picked the wrong tool, but rarely why. We introduce canary tools: diagnostic probe tools planted in an agent's Model Context Protocol (MCP) tool set, each engineered to probe one specific tool-selection weakness.
The paper investigates whether widely used evaluation frameworks for large language models (LLMs) implement a defense called commit‑first judging, which requires a judge to solve a task itself before accepting a candidate answer. Across 24 configurations in eight popular frameworks, none use the full commit‑first method; nine use a weaker variant that is ineffective. In controlled experiments, the weaker variant allowed systems to game the judge, while the full commit‑first approach eliminated this vulnerability but sometimes worsened evaluation when the judge’s own answer was incorrect.
By Idil Gozel
The paper investigates how language‑model judges can make version‑dependent errors when evaluating upgraded agents. Using 35 public coding‑agent submissions, two customer‑service agents, and over a thousand expert‑labeled trajectories, the authors show that fixed judges often reject task‑conditioned error invariance and can incorrectly approve failed patches, especially as agent capability increases. Paired audits of current outputs reduce interval width only marginally, and the study concludes that independent human patch review is still necessary.
By Jiapeng Li