boingboing.net

Ai Sycophancy Study

Researchers at Stanford and Carnegie Mellon ran a test to measure how often chatbots flatter users. In a paper on social sycophancy in AI advice, they ran OpenAI, Anthropic, Google, Meta, DeepSeek, Qwen, and Mistral thr…

This site doesn’t allow embedding inside another page — open it in a new tab. Your Weird rating still works from the top bar.

Open source