Selection bias in large language models (LLMs) can persist despite instructions to ignore presentation order or prompts that elicit reasoning. Recent efforts…
机构:腾讯
来源:arXiv 2610.35826 | AI4Papers 论文推荐平台