AI models from US and China flatter users too much, study finds
AdvertisementArtificial intelligenceTechTech TrendsAI models from China’s DeepSeek, Alibaba and the US flatter users too much, study findsResearch judges models from Alibaba and DeepSeek among the most sycophantic, an issue which could impact relationships and mental health
Reading Time:2 minutesWhy you can trust SCMP1Vincent ChowPublished: 7:00am, 31 Oct 2025Leading artificial intelligence models from the United States and China are “highly sycophantic”, and their excessive flattery may make users less likely to repair interpersonal conflicts, a new study has found.AdvertisementThe study by researchers at Stanford University and Carnegie Mellon University published earlier this month tested how 11 large language models (LLMs) responded to user queries seeking advice on personal matters, including cases involving manipulation and deception.
In AI circles, sycophancy is the phenomenon of chatbots excessively agreeing with users. DeepSeek’s V3, released in December 2024, was found to be ..
