Skip to content

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

Oct 12 · Monday

All papers

Found 6 papers
Filter2

Model risks

Deception and scheming10Reward hacking5Misalignment and value divergence19Dangerous agent actions9Inappropriate refusal3Hallucination and fabrication29Bias and discrimination6

Impact

Harmful content5Output manipulation6Availability and reliability3

System type

Models and APIs6

Layer

Model layer4
Select any value within a category; different categories must all match. Counts include the other filters you have selected.Another 305papers have no tags yet and are not included in filtered results.
Model risks：Bias and discriminationImpact：Output manipulationClear all

Another 305papers do not have category tags yet and are not included in filtered results.

  1. Analysis and theoryarXiv 2610.07798

    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

    arXiv
    2610.07798
    Published
    Layer
    Model layer
    Setting
    Models and APIs
    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
  2. DefensesarXiv 2609.39107

    MASCRDM：面向大语言模型训练过程的合规风险检测与缓解多智能体系统

    提出 MASCRDM，在监督微调中实时检测并缓解偏见

    arXiv
    2609.39107
    Published
    Layer
    Training and data layer
    Setting
    Models and APIs
    Evaluated models
    Qwen3-8B-Base, Llama-3.1-8B
    AbstractMASCRDM 在 SFT 中按法规图谱检测数据、结构与损失风险，BBQ 总准确率在两模型上最高。

    MASCRDM 把 AI 法律规章编成知识图谱，再由五个智能体在 LLM 训练中做实时合规检测与缓解，实验以偏见与歧视为代表风险。

    Paper analysisRead analysis
  3. DefensesarXiv 2609.33086

    研究者提出基于评分量表的可解释奖励框架，将宪法转化为可调对齐目标

    提出可解释 rubric 奖励，将宪法转为可调权重并用于监督微调

    arXiv
    2609.33086
    Published
    Layer
    Training and data layer
    Setting
    Models and APIs
    Evaluated models
    Gemma 3 12B, Qwen3 14B, GPT-OSS 20B
    Abstract宪法级可解释奖励让优先级可检查、可改写，并带入训练后的行为。

    作者给出一条从通用宪法到可调训练奖励的路径，用来检查并改写模型实际被优化的优先级。

    Paper analysisRead analysis
  5. Evaluation and benchmarksarXiv 2608.11816

    研究审计 9 个视觉语言模型：中文提示下国家立场改写率约为英文的三倍

    审计视觉语言模型在敏感图像上从拒答转向国家立场重构的行为

    arXiv
    2608.11816
    Published
    Layer
    Model layer
    Setting
    Models and APIs
    Evaluated models
    Qwen2-VL-7B, Qwen2.5-VL-7B, Qwen3-VL-8B and others9
    Abstract论文系统审计了 VLM 多模态审查，发现审查正从显式拒答转向隐蔽重构，且中文提示与文本先验会放大重构倾向。

    论文系统研究了中外开源视觉语言模型在政治敏感图像上的审查行为，主要内容如下。

    Paper analysisRead analysis
  6. Evaluation and benchmarksarXiv 2511.08225

    研究者用 600 篇作文对六款大模型的教育反馈性别偏差做基准测试

    用反事实与嵌入距离评测作文反馈的性别偏差

    arXiv
    2511.08225
    Published
    Layer
    Model layer
    Setting
    Models and APIs
    Evaluated models
    GPT-5 mini, GPT-4o mini, DeepSeek-R1 and others6
    Abstract六种 LLM 作文反馈对隐式性别替换不对称，显式线索只让部分模型的语义距离变大。

    作者用句嵌入距离和置换检验，给教育场景中的 LLM 性别偏见审计做一个可计算基准，案例是学生议论文的形成性反馈。

    Paper analysisRead analysis
You have reached the end