How to prompt Copilot or ChatGPT to accurately troubleshoot your PC - without the AI overconfidence

I got tired of frustrating, error-filled troubleshooting sessions with AI chatbots, so I asked Copilot for help.
Scientific American

AI scores a ‘C–’ on its hardest math test yet

The second batch of “First Proof” problems is meant to evaluate AI’s usefulness for research-level math. The best model got six or seven of the ten questions right.