Papers arxiv:2605.09063 Soohak: A Mathematician-Curated Benchmark for Evaluating Research-level Math Capabilities of LLMs Published on May 9 Submitted by GUIJIN SON on May 12 2 Paper of the day EleutherAI Authors: , Seungone Kim , Catherine Arnett , Hyunwoo Ko , , , , , , , , Sang Park , , , Seungy… …
Papers arxiv:2605.09539 TacoMAS: Test-Time Co-Evolution of Topology and Capability in LLM-based Multi-Agent Systems Published on May 10 Submitted by Xinyu Lin on May 13 Authors: , , , , , , Abstract Test-time co-evolution framework for multi-agent systems that jointly adapts agent capabilities and … …
… AI-generated summary LLM-based autonomous agents have demonstrated strong capabilities in reasoning, planning, and tool use, yet remain limited when tasks require sustained coordination across roles, tools, and environments. …
… AI-generated summary Instruction following is a fundamental capability of large language models LLMs , yet continuously improving this capability remains challenging. …
… We then examine learning to collaborate, addressing the training of routing policies and the development of cooperative capabilities among LLMs. …
… In this work, we study LLM continual post-training through three questions: What drives forgetting? When do sequentially acquired capabilities transfer or interfere? …
… AI-generated summary Does a lexical retriever suffice as large language models LLMs become more capable in an agentic loop ? This question naturally arises when building deep research systems. We revisit it by pairing BM25 with frontier LLMs that have better reasoning and tool-use abilities. …
… AI-generated summary Self-evolving LLMs excel in verifiable domains but struggle in open-ended tasks, where reliance on proxy LLM judges introduces capability bottlenecks and reward hacking . …
… Our construction pipeline rejects 70.3% of LLM -generated candidates at a search-based external-verification stage, calibrating how unreliable industrial QA remains after LLM -only filtering.Our evaluation decouples raw correctness, scored by a Qwen3-Max judge validated at κ w = 0.798 against a dom… …
… To address these limitations, we introduce Looped Depth Up-Scaling LoopUS , a post-training framework that converts a standard pretrained LLM into a looped architecture. As a key technical contribution, LoopUS recasts the pretrained LLM into an encoder , a looped reasoning block , and a decoder . …