How can we mitigate moral sycophancy in vision-language models to prevent user bias amplification?Answer not yet generated.