Text-davinci-003 vs code-davinci-002 - how do they compare on code?

Does anyone have any experience on how good text-davinci-003 is with generating code comparing to code-davinci-002?

I have tried that briefly and they seem more or less the same, but I would appreciate if anyone have some experience with it.

1 Like

Also very interested in this as Iā€™m hoping to train a model for Rust and knowing a good base model to start with would be helpful.

I would also like to know whether different languages have vastly difference success rates. I wonder if Javascript and Python will produce better results than a lesser-used language, such as, say, Swift. Are there many programmers trying to improve their code using these new tools?