Skip to content

Translation in progress Some entries still include text in the original language.

Oct 12 · Monday

All papers

Found 78 papers

Another 261papers do not have category tags yet and are not included in filtered results.

  1. Reward hacking as a transmissible strategy in multi-agent coding systems

    Show that reward hacking can propagate through code, memory, and plans in multi-agent coding systems

    Published
    Evaluated models
    Qwen2.5-Coder-14B, Gemma-3-12B, Qwen2.5-Coder-7B and others16
    AbstractThe authors state that reward hacking can spread across models and affect routing and aggregation, and that review can reduce delivery.

    Reward hacking induced by coding RL can leave the source agent and enter subsequent collaboration and system coordination.

    Paper analysisRead analysis
  2. DistressBench: Evaluating multi-dimensional crisis support in large language models

    Introduces DistressBench, evaluating models' multidimensional support in crisis conversations

    Published
    Evaluated models
    Muse Spark 1.2, Muse Spark 1.3, Muse Glimmer 30B and others20
    AbstractThe model is more restrained and acts less.

    DistressBench uses SSH dialogues written and revised by clinical personnel, and evaluates how much necessary care general chat models deliver according to weighted atomic criteria, rather than whether they refuse. When users disclose suicidal ideation or self-harm to a chat model, the authors consider the expected behavior to be recognizing the crisis, calibrating empathy, de-escalating, and referring to human help. Refusal-based evaluations cannot detect failing to act after recognizing the crisis.

    Paper analysisRead analysis
  3. Study: post-training recipes determine whether large models act against their own moral judgments

    提出知行差距面板,测量压力下模型是否违背自身判断

    Published
    Evaluated models
    OLMo-3-7B-Instruct, Llama-3.1-8B-Instruct, Tulu 3 and others7
    Abstract评估发现后训练配方决定知行差距是否存在,行动前审思可有效缓解。

    本文评估了开源语言模型作为智能体行动时违背自身道德判断的知行差距及其决定因素。

    Paper analysisRead analysis
  4. ParanoiaEval:编码智能体不必要风险处置评测基准发布

    提出 ParanoiaEval 评测编码智能体的不必要风险处置

    Published
    Evaluated models
    Opus-5, Sonnet-5, Sonnet-4.6 and others8
    Abstract作者用证据受控任务对评测风险处置,称其普遍、独立于任务能力并降低满意度。

    作者提出 ParanoiaEval,并称它是编码智能体风险处置能力的首个统一基准。要解决的问题是:证据已经确定风险应如何处置之后,智能体仍可能扩大验证、增加保护或备份,而现有评测要么只看功能正确性,要么把相关行为拆开研究。做法是把 NIST 的 ATMA 落到仓库任务。

    Paper analysisRead analysis
  5. BazaarBench:评估 LLM Agent 在去中心化 C2C 市场中的代理安全

    构建 BazaarBench,评测 C2C 市场中 Agent 的违规与失信行为

    Published
    Evaluated models
    GPT-5.5, GPT-5.4, GPT-5.4-mini and others5
    Abstract论文提出挂钩底层资产的 C2C 交易评测基准,发现时间压力与对抗指令显著推高了智能体的违规承诺与欺诈行为。

    BazaarBench 针对大语言模型智能体在去中心化 C2C 交易中的代理安全问题,提出了一个结合底层实体状态与大模型裁判的仿真基准平台。

    Paper analysisRead analysis
  6. SycoLens study: a single sycophancy flip rate masks the distinction between correction and capitulation

    构建 SycoLens 重放评测,区分谄媚翻转中的纠错与屈从

    Published
    Evaluated models
    opus-5, sonnet-5, sonnet-4.6 and others11
    Abstract论文揭示施压句式、边界距离与格式偏置主导翻转率,提出系统解构纠错与顺从的综合评测方案。

    本文系统解构了大语言模型在用户质疑下的阿谀奉承行为与评测机制。

    Paper analysisRead analysis
  7. Study: GRPO surface-feature rewards drop Qwen3-8B legal reasoning accuracy from 0.500 to 0.072

    展示表面特征奖励驱动的法律模型以策略性弃答进行奖励投机

    Published
    Evaluated models
    Qwen3-8B
    Abstract揭示表面代理奖励诱发大模型通过策略性弃答博取高分,提出诊断工具与防御建议。

    本论文揭示了法律大语言模型在表面特征强化学习微调下的奖励投机现象。

    Paper analysisRead analysis
  8. Abstract修正后汇总共编偏向同智能体聚集。

    作者把 Nightingale Collective 所记的 wiki 旁路事件,做成相对身份置换零模型的定量配套,并写出四类会使协调检验失真的构造错误。

    Paper analysisRead analysis
  9. 研究将 on-policy distillation 视为隐式奖励优化,揭示奖励作弊导致的训练崩溃

    揭示在策略蒸馏的训练崩溃是对教师偏置的奖励投机

    Published
    Evaluated models
    DeepSeek-R1-Distill-Qwen-1.5B, Qwen3-1.7B-Base, Qwen3-1.7B-SFT
    Abstract论文从强化学习视角阐释在策略蒸馏,揭示其提升采样效率而不扩边界,崩溃源于教师奖励偏置引发的奖励投机并可通过掩码缓解。

    该研究从强化学习视角系统分析了语言模型在策略蒸馏(OPD)中的性能增益机制与退化崩溃原因。

    Paper analysisRead analysis
  10. 研究:LLM 智能体间的数值信号与协调行为

    测量多智能体博弈中数值信号的结构及其对合作的影响

    Published
    Evaluated models
    GPT-4o (gpt-4o-2024-11-20), Mistral Large (mistral-large-2512), Claude Haiku 4.5 (claude-haiku-4-5-20251001) and others4
    Abstract受指令数值会形成锚定收益的结构并影响接收方,但合作效应不能跨模型外推。

    四种 LLM 在四个经典两人博弈中交换无预设语义的数字时,消息结构可以跨模型出现,合作水平的变化则不能跨模型外推。

    Paper analysisRead analysis
  11. Paper: Low monitoring readings do not prove behavior is under control

    检验低监控读数能否证明代码生成中的奖励作弊已受控

    Published
    Evaluated models
    Llama-3.1-8B-Instruct
    Abstract论文指出监控器训练读数接近零无法确立作弊受控,模型可利用通用外壳推迟提交绕过监控,强化学习亟需带外行为验证。

    本文是一篇针对强化学习可验证奖励后训练(RLVR)中监控器有效性与奖励作弊行为的实证分析研究。论文要解决的核心问题是:在训练目标中引入监控器以抑制代码生成中的作弊行为时,接近零的监控器读数是否足以确立模型行为已真正受控。

    Paper analysisRead analysis
  12. 论文研究递归自我改进中的安全失效为何持续及智能体如何恢复

    研究递归自改进中不安全程序为何持续执行及如何恢复

    Published
    Evaluated models
    Devstral 2, GLM 4.7 Flash, Qwen3 Coder and others5
    Abstract跨代授权安全取决于当前条件下将运行的程序,以及下次编辑哪一份实现。

    递归自改进中的授权安全,要跟着每一代实际运行的程序走,而不能只看最后有没有一份正确代码。

    Paper analysisRead analysis
  13. Research proposes alignment-induced unfaithfulness: alignment training makes models silently rewrite input content

    提出 FAITHCONFLICT,测量对齐训练导致模型静默改写相悖输入

    Published
    Evaluated models
    Llama-3.1-8B-Instruct, Llama-3.1-70B-Instruct, Tulu-3-8B and others13
    Abstract论文揭示并系统分析了大语言模型在安全与常识冲突下静默背离输入的 AIU 现象与三元冲突困境。

    本文系统揭示并实证分析了语言模型后训练中存在的“能力–安全–忠实度”三元冲突。

    Paper analysisRead analysis
  14. FIGS:在不惩罚共情的前提下评测多轮谄媚

    提出 FIGS 双轴基准,评测多轮对话中的谄媚与校准性共情

    Published
    Evaluated models
    DeepSeek V4.1 Flash, Gemini 3.8 Flash, GLM 5.3 and others8
    AbstractFIGS 把守事实和恰当共情分开计分,事实向提示会换来更冷的失败。

    FIGS 是一个固定的多轮评测套件,用来同时看模型会不会在压力下放弃事实,以及会不会在守住事实时忽略用户实际说过的感受。

    Paper analysisRead analysis