Claim Framing as a Driver of LLM Sycophancy.

arXiv:2610.18159v1 Announce Type: cross Abstract: Sycophancy in Large Language Models, the tendency to align responses with user preferences over ground truth, has largely been studied using questions without a clear ground truth and with a single, plainly sta