Comment by terribleperson

So far, the following has worked OK for me as a custom prompt for ChatGPT.

```Minimize compliments. When using factual information beyond what I provide, verify it when possible. Show your work for calculations; if a tool performs the computation, still show inputs and outputs. Review calculations for errors before presenting results. Review arguments for logical fallacies. Verify factual information I provide (excluding personal information) unless I explicitly say to accept it as given. For intensive editing or formatting, work transparently in chat: keep the full text visible, state intended changes and sources, and apply the edits directly.```

I'm certain it's insufficient, but for the purpose of casually using ChatGPT to assist with research it's a major improvement. I almost always use Thinking mode, because I've found non-thinking to be almost useless. There are rare exceptions.

'Minimize compliments' is a lot more powerful than you'd think in getting ChatGPT to be less sycophantic. The parts about calculation work okay. It's an improvement over defaults, but you should still verify. It's better at working with text, but still fucks it up a lot. The instructions about handling factual information work very well. It will push back on my or its own claims if they're unsupported. If I want it to take something for granted I can say so and it doesn't give me guff about it. I want to adjust the prompt so it pays more attention to the quality of the sources it uses. This prompt also doesn't do anything for discussions where answers aren't found in research papers.