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 through thousands of personal dilemmas. — Read the rest
The post AI told Reddit's biggest jerks they did nothing wrong 51% of the time appeared first on Boing Boing.