This is by far my most common usage of LLMs, and they’re good at it. Sometimes you have to be intentional about preventing it from being sycophantic and just telling you that you’re right, through a system prompt or by phrasing the question such that it’s comparing ideas neutrally instead of comparing “your” idea to some other idea.

It feels like Claude and ChatGPT have both become more sycophantic over the past months.