Hacker News Reader

Comment by firasd | Hacker News Reader

Yeah... If you even say to Claude I want to make an app like github.com/user/repo it's not gonna just clone it and do some touch ups. Anything can happen of course, but my spidey sense says generally AI won't fork a project unasked (just because of the implied maintenance burden of applying future patches from upstream if nothing else)

View on HN

Yeah - the whole Claude did it thing here reads to me like blaming a grammar issue on an LLM. Yes they have their issues but this isn't one of them - they're not deterministic they aren't going to make the exact same bug as a project you never heard of unless it's a specific technical shortcoming/training data gap/skill issue of the model. If you say "I want something exactly like xyz in every way" though then sure - but you're claiming you never asked for that? Not passing the sniff test here

I have tried this with Codex, and if you ask it to implement a project based on an algorithm or design in a given repo, it always puts an attribution link saying that part of it was adapted from that repo. Unfortunately I think we’re being duped here.

Also, Claude has the habit of finding and fixing bugs without being asked - for example, in code that an earlier version of itself wrote. So while you can never be sure, there's a non-zero chance that even if it had copied the original code it would have found and fixed the accidental trap.

I have had Claude and Codex look at third party repositories on several occasions as reference material for something I wanted to make my own version of, and not once have they ever cloned it unless I told them to. Sometimes I would notice they would just keep reading the repo through a ton of API calls so I'd just stop them and clone it for them and tell them to look there. And not once was the output anywhere near that close to the original, plus they usually added attribution info if the results were even (unscientific estimate) 50-60% similar.

Reply on news.ycombinator.com