One problem I encounter when using ChatGPT to troubleshoot coding issues is that it seems very heavily biased towards positive responses to questions. If I ask it "could the problem be X" or "might Y be a solution", it will usually say yes even when it's not the case. At most I'll get a "yes, but". Rarely will it flat out tell me I am way off. Maybe ChatGPT genuinely "believes" what it's confirming but I doubt it and I find I get more constructive answers with less leading questions.