Paul Cantrell on Nostr: That part is anecdotal, and it’s an open question how long lived it is. There’s ...

That part is anecdotal, and it’s an open question how long lived it is. There’s also a family of partly theoretical questions: What are models trained on once people start using models to write code? Do they propagate expertise from experts to newcomers? Do they create a kind of model in breeding and degenerate? Do they •prevent• the creation of new tools?

I have serious doubts about the sustainability of even the current very mixed success of LLMs for code.