大模型生成自闭症沟通时会扭曲真实表达，安全对齐反而强化神经正常化偏见。
Algorithmic Fragility and Persona Bias in LLM-Generated Autistic Communication
- 用双人格重写框架让大模型模拟自闭症与非自闭症话语，对比生成差异。
- 自闭症人格生成的词汇和情感表达差异大，但语义相似度高，多数模型却混淆二者。
- 模型生成崩溃由对齐策略决定，人类内行标注可揭示模型分类系统性错误。
安全对齐虽减少有害输出，却无意中编码了被净化的、符合神经正常标准的边缘群体沟通模式。我们采用双人格重写范式，让十款大语言模型（LLMs）将自然发生的自闭症话语分别以自闭症或神经典型人格重写。结果显示，尽管语义相似度相当，自闭症人格重写在词汇形式和情感基调上显著偏离；而多数模型将跨人格生成坍缩为近乎相同的输出。为揭示生成失效机制，我们提出多智能体定性分析框架。结果表明，系统性输出抹除、刻板幻觉与任务回避型元评论是普遍失败模式，且按对齐策略聚类而非参数规模。最后，与自闭症人群内行人标注者的针对性对比显示，社区内知能引发相对于模型分类的系统性标签反转。研究证实，当前对齐训练导致仅通过定性分析可见的人格特异性生成崩溃，确认了提示工程无法解决的深层表征鸿沟。
原文摘要 · Abstract (English)
Safety alignment reduces explicitly harmful outputs but inadvertently encodes a sanitized, neuronormative representation of marginalized communication. We investigate this encoding using a dual-persona rewrite paradigm, prompting ten large language models (LLMs) to rewrite naturally occurring autistic discourse from either an autistic or neurotypical persona. We uncover autistic-persona rewrites diverge significantly more in lexical form and affective register than neurotypical rewrites, despite equivalent semantic similarity. Furthermore, most models collapse cross-persona generations into near-identical outputs. To uncover the mechanisms behind this generative breakdown, we introduce a multi-agent qualitative analysis framework. Our results reveal systemic output erasure, stereotyped hallucination, and task-evasive meta-commentary are pervasive failure modes for this task that cluster by alignment strategy rather than parameter scale. Finally, our targeted comparison with autistic human annotators demonstrates that community-insider knowledge produces systematic label reversals relative to LLM classifications. Our findings indicate that current alignment training causes persona-specific generative breakdown visible only through qualitative analysis, confirming a deep representational gap that prompt engineering cannot resolve.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。