@lain 6-8 months ago i had no success whatsoever with LLMs, they were really outputing garbage code imho. The models have greatly improved lately.
I still find them pretty bad at software architecture though, i made them review some PR at work and there was very obvious problems in these PR on the architecture side (with terrible distribution of responsibilities in between classes) that was itself from an LLM, and even fable was like "oh yeah this looks great".
If only llms write code in the future, that's not a big problem: they can handle that stuff like its nothing. But i'm not sure LLMs will be available that "cheaply" (it's already not cheap for most people) in the future, which means code still needs to be properly architectured for now, so that needs a lot more guidance for the LLM.
Anw yeah, latest models are nice. Have you checked
https://alien.arwalk.net ?