It feels to me like properly instrumented, these diffusion models are going to be really powerful coding tools. Imagine a “smart” model carving out a certain number of tokens in a response for each category of response output, then diffusing the categories.

Why do you think so? I've played with the Diffusion model a bit and it makes a lot of mistakes

It's super fast to iterate with? Curious what mistakes are you seeing and how were you using it?

I use the instant edit feature with directions on exact code changes.