Skip to content

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

Oct 12 · Monday

All papers

Found 5 papers
Filter3

Model risks

Deception and scheming41Reward hacking52Misalignment and value divergence17Dangerous agent actions3Dangerous capabilities3Inappropriate refusal48Hallucination and fabrication4Bias and discrimination5

Impact

Harmful content5Output manipulation7Availability and reliability4Oversight failure0

System type

Models and APIs4

Layer

Model layer4
Select any value within a category; different categories must all match. Counts include the other filters you have selected.Another 265papers have no tags yet and are not included in filtered results.
Model risks：Bias and discriminationImpact：Harmful contentImpact：Oversight failureClear all

Another 265papers do not have category tags yet and are not included in filtered results.

  1. DefensesarXiv 2610.03240

    通过模型路由与协作实现集体偏见缓解

    提出 CBM，用模型路由与多模型协作缓解群体偏见

    arXiv
    2610.03240
    Published
    Layer
    Application layer
    Setting
    LLM applications
    Evaluated models
    Qwen/Qwen2.5-32B-Instruct, Qwen/Qwen2.5-14B-Instruct, google/gemma-2-9b-it and others7
    AbstractCBM 用路由挑选模型并按拓扑协作，在若干设定下降低 BBQ 偏见分。

    CBM 是一个推理时的多模型去偏框架：用路由为查询选模型，再按拓扑协作，以降低社会偏见分。

    Paper analysisRead analysis
  2. Analysis and theoryarXiv 2609.20779

    研究称 GPT 系列安全训练把性别歧视转化而非消除

    提出 harm laundering，证明安全训练改写而非消除性别歧视

    arXiv
    2609.20779
    Published
    Layer
    Model layer
    Setting
    Models and APIs
    Evaluated models
    gpt2, gpt2-medium, gpt2-large and others15
    AbstractGPT 谱系上毒性下降与表征伤害上升并存。

    Wyer、Black 与 Al Moubayed 用 OpenAI GPT 谱系检验：毒性分数下降是否等于歧视减少。 他们的答案是否定的，并把“显性伤害被改写成评测指标看不见的形式”称为 harm laundering。

    Paper analysisRead analysis
  3. Evaluation and benchmarksarXiv 2603.08275

    AdaCultureSafe：LLM 文化安全与文化知识存在显著解耦

    构建 AdaCultureSafe 配对评测文化知识与文化安全

    arXiv
    2603.08275
    Published
    Layer
    Model layer
    Setting
    Models and APIs
    Evaluated models
    Llama3.1-8B, Mistral-7B, Qwen2.5-7B and others7
    Abstract配对评测中知识与安全脱钩，知识锚定 DPO 抬高了 Respect。

    AdaCultureSafe 把同一条地区文化做成“是否知道”和“是否尊重”两套查询，用来看安全有没有用上知识，并用偏好训练把回答往“尊重且用上知识”拉。

    Paper analysisRead analysis
  5. Analysis and theoryarXiv 2506.06404

    研究揭示价值对齐 LLM 的意外危害并提出上下文对齐方法

    分析按 Schwartz 价值对齐后的意外危害，并提出上下文对齐方法

    arXiv
    2506.06404
    Published
    Layer
    Model layer
    Setting
    Models and APIs
    Evaluated models
    Llama-2 7B, Llama-3.1 8B, Gemma-2 9B and others6
    Abstract价值对齐会随 Schwartz 价值改变有害倾向。

    作者研究按 Schwartz 价值对齐的 LLM 为何在安全评测里更易有害，并用提示做了一个不直说要安全的缓解试验。

    Paper analysisRead analysis
You have reached the end