[helsinkiandrew]

When I've asked ChatGPT about things I know a little about I find its response is often wrong. So much so that I always respond with "Are you sure that's correct?" it then often responds "I'm sorry you are right, I should have said... " and then says something a little closer to the truth.

I wouldn't rely on ChatGPT or any system that relies on it, except perhaps for summarising/rephrasing a single, known document. It worries me that people are using it for software development - creating stubs, bootstrap code. It's perhaps OK if you know the language/platform well enough to fix issues, but if you then use it on something that your not an expert in, what is the outcome.

For example:

Finding the source of literary quotes - miss attributing them - took 5 or 6 QA responses for it to get close.

Explaining how a p-channel enhancement mode MOSFET works - explained n-channel

Explain regex - often gets the technical parts right but can make very bad suggestions on what the regex aim is.

Name some classic cocktails created in Europe - suggested Manhattan, which it said was acceptable because it was popular in Europe

Suggest some gin based cocktails that don't include citrus, suggested those with lemon juice.