Hacker News new | ask | show | jobs
by pu_pe 39 days ago
> And most importantly, you don't care at all if the tool you vibe-coded is any good. If you write at tool that converts an image to black & white, you are the kind of person who doesn't know or care what KIND of black&white it is. The fact that there are many algorithms to choose from would never cross your mind.

Why do you believe that? What if I care so much about black and white and need a specific algorithm, isn't it much better to do that through a tool I can control rather than any proprietary one?

1 comments

... because there are open source tools that implement these things properly? Take Darktable, if you try to build an alternative because you "just want to do a bit of color grading" you will produce garbage.

My point is that AI bros are forgetting that the LLM is only as good as the person using it. If you don't sufficiently understand the problem you're solving, your solution will be terrible. If you believe that AI will simply take that work away from you and do it itself, you should ask your favorite LLM to explain how LLMs work.

But what if I _only_ need to convert images to black and white, why would I install a full ass Darktable setup for that?
> ... because there are open source tools that implement these things properly? Take Darktable, if you try to build an alternative because you "just want to do a bit of color grading" you will produce garbage.

... only if you don't pay attention. The existence and popularity of Darktable literally means that LLMs have a reference implementation on-hand and probably seen some of it in training too. If for some reason they don't immediately go looking at it, then you can always ask.

> If you don't sufficiently understand the problem you're solving, your solution will be terrible.

Fortunately the LLM is able to help you with first understanding the problem space, too. Just don't expect it today to zero-shot things you don't understand well enough.

I agree with you, and my issue is exactly that seemingly the popular opinion here is that you CAN zero-shot things you don't understand.