I also played a little bit with OpenGPT and Nim these days. I tried to see how good could it be on creating bindings:
<https://gpt.best/I666BU01> Which won't work (there are too many failures), but it doesn't look too alien either. I was wondering how good would it be this sort of approach for bindings. There are some open source options: GPT-J, [GPT-Neo](https://github.com/EleutherAI/gpt-neo), ... that could be finetuned for Nim. But the training process looks scary to me. I am amazed about how these things will look in 5-10years.
