Ask HN: Stable Diffusion for Programming?

I know little about the inner workings of AI and ML, so I've been wondering about the possibilities of this relatively new stable diffusion technology being adapted for creating working programs via input, ie: "Write a program that generates a fibonacci sequence"

Assuming there is plenty of source material and targeting a single language (or if you think it could be language agnostic?). Perhaps you decompile working programs down to assembly/machine-code and let that be the input/output.

I know about GitHub Copilot, which is very cool, but is it the same tech?