* * *

I thought the same thing about the compiler, but the threads I read said GPT would just make up imaginary crates to use in some cases.

> GPT would just make up imaginary crates to use in some cases

I think this is where trial and error comes into play. If GPT makes up a crate for, say, an HTTP server, and the API is very close to a real HTTP server, then it's not a big deal. You can just swap in a crate you know works.

If GPT is making up a crate and also mangling the use of such a crate, then it's probably not worth the human effort to fix all of that.

The same is true with any language: you'd have to try it a few times to decide if it's even worth the time saved.

Copilot is probably a better product for coding regardless, although I think it's deeply unethical and hope it gets sued out of existence.

It generates pretty good Typescript, if it doesn't select a deprecated library.

What have you tried to prevent it from selecting a deprecated library?

Yeah, I've noticed that too. Its code is generally alright, but it has a habit of picking outdated / deprecated / poorly maintained libraries.

I wonder if that's because it's not weighting by recency in it's training or something.

The original was mostly trained on Python[1], but JS seems a close second. So I assume python has the highest quality. If you don't specify a language, it seems pick either of these.

[1] https://openai.com/blog/openai-codex

Chatgpt help me in a very nice way. though sometimes my needs not available. There's a trick how to proper ask chatgpt.

I think it works best in Python, but I don’t have numbers to bring. It’s just how I feel. Other languages have more wrong answers.

I have tested it with Python doing function working with pandas dataframes and the code was perfect.

I’ve been using it to learn SQL - it’s been helpful asking it to explain queries or joins that I want to perform.