My few tests on chat.openai have been mixed as well. For some things, it helps condense a bunch of information down to a really useful procedure. However, you never know when it is wrong as it seems very confident. For instance, recently I was trying to get help with a Caddy2 config issue and it kept giving me syntax for Caddy1 config even though I asked for v2.
This sounds a lot like the advice from: CoPilot Review: My Thoughts After 6 Months