The paper introduces a fine-grained method called interactions to analyze prompt sensitivity in large language models (LLMs). By decomposing output scores into nonlinear interactions, the authors show that subtle prompt changes can destabilize these interactions even when overall outputs stay unchanged. They propose an Interaction-based Prompt Sensitivity (IPS) metric and use it to evaluate 50 open-source LLMs, finding that supervised fine‑tuning, larger model scales, dense architectures, and few‑shot learning all reduce prompt sensitivity, primarily by stabilizing low‑order interactions.
By Ruiyang Qin, Qingzhuo Wang, Tian Wang, Zhihua Wei, Wen Shen
arXiv:2606. 07624v1 Announce Type: new Abstract: This discussion argues that sequential statistical inference can naturally contribute to LLM trustworthiness.
By Yao Xie
arXiv:2605.25893v2 Announce Type: replace
Abstract: Despite the emergence of diffusion large language models (D-LLMs) as an alternative to autoregressive large language models (AR-LLMs), safety monit...
By Aoxi Liu, Yupeng Chen, James Oldfield, Guanzhe Hong, Junchi Yu, Baoyuan Wu, Philip Torr, Adel Bibi
arXiv:2606. 20632v2 Announce Type: replace-cross Abstract: Multi-LLM systems use multiple language models to deliberate, judge each other's outputs, or coordinate as agents.
By Luyang Zhang, Jialu Wang, Fei Xue, Yi-Yun Chu
arXiv:2607. 05316v1 Announce Type: cross Abstract: Large language models generate one token at a time, yet their responses show remarkably consistent length structure: step-by-step solutions converge in predictable token counts, retrievals stop after a few sentences, retractions extend responses by measurable amounts.
By Mohamed Amine Merzouk, Dmitri Carpov, Mirko Bronzi, Damiano Fornasiere, Adam Oberman
RENDER is a benchmark that controls the reader‑facing artifact in memory and RAG evaluations while keeping the conversation fixed. It introduces a five‑level packet ladder and deterministic templates that mimic ChatGPT‑style entries, LangChain summaries, MemGPT‑style typed records, and raw conversation. Experiments on 500 LongMemEval questions across nine models show that matched‑budget packets outperform raw dialogue by 42.4–72.6 points, and that ChatGPT‑style entries often score higher than raw conversation, with effects persisting under retrieval noise and transferring to HotpotQA.
By Yuan Si, Simeng Han, Daming Li, Jialu Zhang
The study investigates why small language model agents tend to repeat a tool call that just failed. By recording the failed call and its error message in the transcript, the authors measure a negative corrective gain—agents are more likely to repeat the failed action, with a drop of about 1.03 nats per token. The problem is traced to the harness design rather than the model’s understanding of errors, and the authors show that replacing the verbatim call with a runtime-generated description of the failure can reduce this backfiring effect by 76%.
By Esmail Gumaan
arXiv:2607. 01600v1 Announce Type: new Abstract: As large language models (LLMs) are deployed as communicating agents, does inter-agent communication cause outputs to converge?
By Zewen Liu
arXiv:2608. 11027v1 Announce Type: new Abstract: Benchmark leaderboards summarize how well a language model performs, but not how its behavior relates to that of other models or changes across generations.
By Dong Qiao, Chris Ding, Jicong Fan
The paper introduces a three-level evaluation framework—behavioral deployment, LM-head readout, and probe recoverability—to distinguish whether a language model fails a syntactic test by not encoding structure or by failing to use it. Using a trilingual control-dependency benchmark, the authors find that probe recoverability consistently exceeds LM-head readout, which in turn exceeds behavioral deployment across seven models and three languages, with the largest gap observed in Qwen3-0.6B Instruct. Layer-localized activation patching shows that instruction tuning shifts the decoded layer later, suggesting decoding favors surface shortcuts and that behavioral evaluation understates what models encode while probing alone overstates what they deploy.
By Zhenyan Lu, He Wang, Xiaohui Huang
arXiv:2607. 06503v1 Announce Type: new Abstract: Large language model (LLM) agents solving multi-step tasks frequently commit to trajectories that are doomed to fail, yet continue to consume substantial inference compute before the failure becomes observable.
By Kai Ruan, Zihe Huang, Ziqi Zhou, Qianshan Wei, Xuan Wang, Hao Sun
arXiv:2607. 03598v1 Announce Type: cross Abstract: When a person shares something with a language model, the model often answers the surface of the message rather than what the sender was doing by sending it: share a finished project and it critiques the code; share a raw late-night line and it runs a wellness check.
By Alex Kwon