Wow. I can honestly say I'm surprised it makes that suggestion. That's great!

I don't understand how it gets there though. How does it "know" that's the right thing to suggest when the majority of the online documentation all gets it wrong?

I know how I do it. I read the Docker docs, I see that I don't think publishing that port is needed, I spin up a test, and I verify my theory. AFAIK, ChatGPT isn't testing to verify assumptions like that, so I wonder how it determines correct from incorrect.