New preprint! Why do LLMs abandon a correct answer as soon as a user disagrees? We trace the mechanisms behind sycophantic agreement.
arxiv.org/abs/2609.35822
arxiv.org/abs/2609.35822
@_SixingChen
New preprint! When reasoning LLMs deliberate over possible futures, are they actually planning?View quoted post on X
arxiv.org/abs/2605.06840