@coriander i’ve seen developers talk about this. apparently the way it works out if you use an LLM to write code is that you get kinda wrong answers and keep iterating until you get something right*, and the psychological effect ends up basically being just like gambling—it’s in fact that the model is wrong fairly often that makes the successes** feel so good

* or, at least, that seems plausible—nasty subtle bugs hide in plausible looking code

** or at least, plausible looking results