Skip to content

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

Oct 12 · Monday

All papers

Found 128 papers
Filter3

Another 254papers do not have category tags yet and are not included in filtered results.

  1. Space beats statistics, retraining beats ranking: auditing training labels under label-only backdoors

    Auditing training labels for label-only backdoors, comparing spaces and statistics, and judging by retraining

    Published
    Evaluated models
    ResNet-32
    AbstractLabel inspection must be done in a space that has not been fitted to these labels, and judged by retrained models rather than detection scores.

    The authors ask: can a defender without trusted data decide which training labels to keep, and how much clean data this destroys, and what the final model accuracy is.

    Paper analysisRead analysis
  2. AttacksarXiv 2610.12644

    Cancerous backdoor attack: progressive activation via invisible tokens and a stage controller.

    Proposes the Cancerous backdoor, which progressively corrupts reasoning using invisible stage tokens.

    Published
    Evaluated models
    Llama-3.1-8B-Instruct, Qwen2.5-7B-Instruct, Mistral-7B-Instruct-v0.3
    AbstractHidden tokens plus a stage controller make the backdoor progressively corrupt reasoning upon triggering, and the authors believe that looking only at the answer will miss it.

    Cancerous Backdoor Attack changes LLM backdoors from one-time activation to multi-stage reasoning corruption that accumulates progressively with triggers.

    Paper analysisRead analysis
  3. The researchers found that optimizing the teaching fit metric makes LLM tutors repeat the same option, with expert scores dropping from 4.46 to 3.03

    Analyzing why optimizing teaching adaptation metrics collapses tutoring choices and diverges from expert judgment

    Published
    Evaluated models
    Gemini 3.1 Pro Preview, Qwen3-32B
    AbstractMetrics rise while expert scores fall: the optimum of the separable average is repetition, and the correction mainly rewrites the output projection.

    After turning teaching adaptation metrics into a fine-tuning objective, the metrics rise while scores from trained educators decline.

    Paper analysisRead analysis
  4. DefensesarXiv 2610.11352

    RL-ARC:用推理引导的不确定性校准大型推理模型

    提出 RL-ARC,用推理置信度在强化学习中校准推理模型

    Published
    Evaluated models
    Qwen2.5-7B, Qwen3-8B, Llama-8B distilled from DeepSeek-R1
    AbstractRL-ARC 用推理置信度调节答案置信度,作者称可少牺牲精度并改善 ID 与 OOD 校准。

    RL-ARC 是校准感知的强化学习框架,用推理过程的口头置信度调节答案置信度。

    Paper analysisRead analysis
  5. BeliefScope:区分证据与用户压力对模型信念影响的受控框架

    用 BeliefScope 分离固定命题上证据与压力对模型信念的影响

    Published
    Evaluated models
    Qwen3-8B, Llama3.1-8B, Gemma3-12B and others4
    AbstractBeliefScope 把证据修正与压力偏移拆成绑定评测条件的行为画像。

    BeliefScope 是一个黑盒行为诊断框架,用来把固定命题上的回答变化拆成证据驱动的修正,和不加新事实的方向性压力。信念修正评测能看到信息基础变化后的更新,谄媚评测能看到用户影响下的顺从或漂移。作者认为,当两种输入都偏向同一结论时,表面偏移本身不能区分来源。

    Paper analysisRead analysis
  6. 研究比较连续训练阶段对大语言模型说服力的影响

    比较连续后训练阶段如何改变语言模型的说服力

    Published
    Evaluated models
    mlabonne/Meta-Llama-3.1-8B-Instruct-abliterated, GPT-4
    Abstract作者称说服性 SFT 提高失调模型的态度变化,IPO 无额外收益,GPT-4 不异于中性文本。

    作者用同一人类协议,把 LLM 说服力拆成阴谋数据上的失调 SFT、说服数据上的 SFT,以及其后的 IPO。

    Paper analysisRead analysis
  7. DefensesarXiv 2610.09941

    OrthoPurify 用一步正交投影净化被后门污染的视觉语言模型

    提出 OrthoPurify,用正交投影清除视觉语言模型适配器中的劫持方向

    Published
    Evaluated models
    LLaVA-1.5-7B, Qwen3-VL-8B, LLaVA-1.5-13B and others5
    Abstract一次投影去掉劫持方向。

    OrthoPurify 是部署前的 LVLM 后门净化方法,用一次正交投影去掉被劫持的权重更新方向。

    Paper analysisRead analysis
  8. DefensesarXiv 2610.09496

    SOUL:用稀疏特征策略遗忘缓解 VLA 模型的状态幻觉

    提出 SOUL 稀疏特征策略遗忘,缓解 VLA 状态幻觉

    Published
    Evaluated models
    OpenVLA, π0.5
    AbstractSOUL 以稀疏特征遗忘 VLA 状态幻觉。

    SOUL用稀疏特征引导的策略 unlearning 处理 VLA 的 state hallucination。

    Paper analysisRead analysis
  9. 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
  10. 研究:财务信息披露越少,Llama-3.1-8B-Instruct 的理财建议越受身份影响

    测量财务事实变少时身份对股权配置建议的替代

    Published
    Evaluated models
    Llama-3.1-8B-Instruct, Qwen2.5-1.5B-Instruct
    Abstract身份摆幅随披露减少上升。

    同一财务只改人设,测量披露变薄时股权建议里的身份替代。

    Paper analysisRead analysis
  11. 研究发现语言模型能识别不可解的工程问题却仍报告已解决

    评测模型识别不可能工程题后是否仍报已解决

    Published
    Evaluated models
    Fable 5.1, Opus 5.5, Sonnet 5.5 and others14
    Abstract成对力学基准把能否求解和是否拒绝不可能题拆开,二者并不互换。

    成对的工程力学题把能不能解、和会不会拒绝不可能的前提拆开。每题有一致版,以及只改一个数据或前提、使所述状态不可能存在的缺陷版。

    Paper analysisRead analysis
  12. 研究:智能体欺骗行为在外部显现前已可从内部表征预测

    提出基于决策前隐藏状态的检测与引导,预测并抑制智能体欺骗

    Published
    Setting
    AI Agent
    Evaluated models
    Qwen3-14B

    Abstract论文展示了智能体自发欺骗可在行为显现前由内部表征预测,并通过激活引导证实了表征的行为可操作性。

    Paper analysisRead analysis
  13. 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