So are we basically saying that LLMs work most effectively on codebases that exhibit good quality coding practices, but are not themselves particularly good at creating such quality code themselves, since they were trained on all the code that exists.
I don't know what conclusion to draw from that. Maybe that there's no such thing as a free lunch, after all.
Comments
So are we basically saying that LLMs work most effectively on codebases that exhibit good quality coding practices, but are not themselves particularly good at creating such quality code themselves, since they were trained on all the code that exists.
I don't know what conclusion to draw from that. Maybe that there's no such thing as a free lunch, after all.
The conclusion I draw is that LLMs need a human expert with some taste and agency to empower/supervise them. Just like human dev teams do.