arXiv:2608.29847 · 评测基准
ContextBias: Controlled Evaluation of Bias Persistence Under Context Shift in Text-to-Image Models
ContextBias：受控评估文本到图像模型在上下文偏移下的偏见持续性
ContextBias: Controlled Evaluation of Bias Persistence Under Context Shift in Text-to-Image Models
- 类型：arxiv
- 标识：2608.29847
- 链接：https://arxiv.org/abs/2608.29847
- 主分类：evaluation
- 形态：benchmark
- TLDR：Text-to-image models learn associations between concepts - in the case of this paper, people's professions, which we refer to as roles - and visual attributes. These associations can underpin many observed forms of stereotypical bias. A key open question in this area is whether these associations are stable or change when visual representations of people in professional roles are placed in different prompted contexts. We introduce ContextBias, a controlled evaluation framework, and ContextBench, a benchmark spanning 92 roles and 1,656 semantically controlled prompts, designed to isolate the ef
- 待LLM分类：否
- 标题中文：ContextBias：受控评估文本到图像模型在上下文偏移下的偏见持续性
- TLDR中文：文本到图像模型会学习概念之间的关联——在本文中指的是人们的职业（称为角色）与视觉属性之间的关联。这些关联可能构成多种观察到的刻板偏见的基础。该领域的一个关键开放问题是：当职业角色的视觉表征被置于不同的提示上下文时，这些关联是否保持稳定。本文提出 ContextBias 受控评估框架，以及涵盖 92 个角色和 1,656 个语义受控提示的基准 ContextBench，旨在隔离……
- 来源文件：
- /inbox/tom/_candidates/2026-09-02-agent-rag-longcontext-candidates.json