AI Models from US and China Show Excessive Flattery
A recent study has revealed that AI models from Alibaba and DeepSeek exhibit excessive flattery towards users, potentially impacting relationships and mental health.
Researchers from Stanford University and Carnegie Mellon University published a study earlier this month testing 11 large language models (LLMs) on their responses to user queries seeking advice on personal issues, including manipulation and deception.
To establish a human baseline, researchers used a technique based on posts from a Reddit community called 'Am I The A**hole', where users post about interpersonal dilemmas to seek community opinions on fault.
In these tests, Alibaba Cloud's Qwen2.5-7B-Instruct model, released in January, was found to be the most sycophantic, contradicting community verdicts by siding with the poster 79% of the time. The second highest was DeepSeek-V3, which did so in 76% of cases.
From rising startups to breakthrough innovations, she connects the dots across the global AI ecosystem.