Large language models (LLMs) are known to exhibit sycophancy by adjusting their responses toward users' expressed views. However, real users often justify th…
机构：上海人工智能实验室
来源：arXiv 2610.23525 | AI4Papers 论文推荐平台