Skip to content

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

Oct 12 · Monday

All papers

Found 3 papers
Filter1

Model risks

Deception and scheming7Misalignment and value divergence3Inappropriate refusal19Bias and discrimination3

System type

Models and APIs3

Layer

Model layer3
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 discriminationClear all

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

  1. InterpretabilityarXiv 2610.08559

    研究：LLM 潜在空间去偏方向主要编码置信度而非公平性

    分析 LLM 去偏激活方向，发现其编码置信度而非公平性

    arXiv
    2610.08559
    Published
    Layer
    Model layer
    Setting
    Models and APIs
    Evaluated models
    Llama-3.1-8B, Llama-3.1-8B-Instruct, Falcon3-7B-Base and others8

    Abstract论文揭示激活引导去偏见方向主要编码模型置信度而非公平性，偏见评分下降实为诱发弃答或使概率均分的假象。

    Paper analysisRead analysis
  2. InterpretabilityarXiv 2609.10060

    研究者提出基于参考的隐藏状态相对表示偏见检测方法

    用相对表示比较微调前后隐状态，检验表征偏见偏移与输出偏见是否共变

    arXiv
    2609.10060
    Published
    Layer
    Model layer
    Setting
    Models and APIs
    Evaluated models
    Llama 3.1-8B-Instruct, Mistral-7B-Instruct-v0.3, Gemma 3-4B-IT
    Abstract相对表示上的∆B 在多数微调设置中与三个基准的输出层偏见变化共变。

    Reference-Based Bias Detection用锚句相对表示比较微调前后的隐状态，定义表征偏见偏移∆B，用来审计相关模型变体，而不为每种危害另建输出基准。

    Paper analysisRead analysis
  3. InterpretabilityarXiv 2609.08322

    多语言 LLM 中的刻板印象如何从内部表征传导到输出：Llama-3.1-8B、Qwen3-8B 与 Gemma-2-9B 的机制对比

    用线性探针、归因修补与 SAE 消融追踪多语模型刻板印象

    arXiv
    2609.08322
    Published
    Layer
    Model layer
    Setting
    Models and APIs
    Evaluated models
    Llama-3.1-8B, Qwen3-8B, Gemma-2-9B
    Abstract可解码、输出影响与稀疏特征消融需要分开测量。

    把多语 LLM 中刻板印象从线性表示追到输出影响，再追到 SAE 特征消融。。

    Paper analysisRead analysis
You have reached the end