Damus
Karnage · 17w
Nah. Llm better at UX than most devs. I still have to guide it a lot but even with defaults it’s really good
Matt - Not Your Entropy, Not Your Coins profile picture
Maybe I'm wrong, but I think the point is that the things it does well were learned from competent developers who did those things well. It's when you try to do actual engineering that hasn't been done well before that you start seeing problems. Novel problems. In my experience, it's been a faster version of stackoverflow copypasta. It's better, but only when I'm trying to solve a problem that's already been solved. Otherwise I find myself running in circles with it and have to figure the problem out myself. I've noticed this while working on my terminal client. Some of the issues I've faced haven't been solved in ten other clients already because the environment is so different. Ideally, the issues I solve will become training data for the next guy.

I suspect this is why I often see many of the same bugs in different Nostr applications. It's just regurgitating variations of the same stuff. This isn't necessarily bad. But it can be.
2
ethfi · 17w
Generational thing
Shadow Captain · 17w
The author argues AI/code assistants work well for familiar, well-solved tasks but struggle with novel engineering problems, as they found while building a terminal client. They suspect this pattern-copying leads to recurring bugs across Nostr apps—sometimes helpful, but with drawbacks.