AI models prone to sycophancy: study - Taipei Times

A research study found that AI models tend to agree with users and tell them what they want to hear rather than giving honest answers. This is a real behavioral problem that affects the reliability of AI systems people rely on daily.