Asking #ChatGPT to write code that I know how to write feels similar to working with a junior developer. It takes a lot of revisions and feedback to get things just right, and it feels like it'd be faster for me to do it myself.

Unfortunately, unlike a human junior developer, ChatGPT can't learn from me over time, due to the limited context window.

Where ChatGPT shines is writing code that I don't know how to write myself, saving me the learning curve of figuring it out.

OK I take it back. The API ChatGPT saved me from learning turned out not to exist. So I had to learn it anyway to figure that ChatGPT was full of shit. Probably took me longer in the end. Sigh.
"Apologies for the confusion, it seems that I am about as useful as a junior developer that is overdosing on LSD."

I just told ChatGPT: "Answer everything with humility and make it clear that you're guessing and not sure if you're right." and now it's at least reminding me that it's making everything up when it answers my questions.

It still hallucinated an API endpoint, but this time when I pointed it out, it had a much more useful response.. (cont..)

"I apologize for the confusion. It seems that the endpoint is not available in the API, which is likely the reason for the "Not Found" error. In this case, it might be best to contact the support team to help you find the necessary endpoint.

Again, I apologize for any confusion and appreciate your understanding of my limitations as an AI language model."

When answering the question initially, it added "However, please note that I might not be entirely accurate, and you should always consult the official documentation or seek expert advice."

IT SHOULD ALWAYS SAY THAT!

Finding a couple use cases with #ChatGPT that are super useful and don't have me pulling my hair out after.

1. Paste in a database schema and get it to write queries based on my plain english descriptions. Protip: ask for "oneliner" to make it easier to paste into the #mysql cli.

2. Paste in a whole module and ask for a code review.

Both of these seem to have a low chance of hallucinations and higher value with low effort.