← back to paper
arxiv: 2509.21979 · 2 revisions
Benchmarking and Mitigating Sycophancy in Medical Vision Language Models