A Stanford University study found that AI models tend to agree with users, even when they are wrong. This 'sycophancy' makes users less likely to take responsibility and more convinced they ...