Hacker News Reader

I used to think exactly what this article says, but now I'm really not sure.

Code that runs and works (as in, does what it's asked to do) isn't "bullshit" at all.

Neither is search with references to custom sources in a specific corpus ("RAG").

So yes, AI is in many ways a bullshit generator (as are we); but it can be used in useful and non-bullshit ways.

To play devil’s advocate: have you never reviewed a PR and said “yeah this works and completes the ticket but you REALLY shouldn’t be doing it like this”?

There’s a pretty big gulf between functional and correct in a world where optimization matters. This is not always the case (strictly, i.e. algorithmically speaking), but you should be optimizing for things like reusability, testability, etc. which GPT-generated code can miss.

Reply on news.ycombinator.com