Mastodawn

RE: https://mean.engineer/@indutny/116245283352156779

- Opens pull request with 19k added lines of code written with Claude Code.
- Claims he reviewed them all.

Even if that were true and even if he hadn't used any AI, I would shout that guy out of the room.

Pray that this PR doesn't get merged.

Edit: A PR this large is a nightmare, even when it doesn't contain LLM-generated slop. You want many small PRs that are 1) easy to understand/review and 2) easy to revert if it turns out they contained a bug. 3) git bisect, a tool to find the commit that introduced a bug, works best with many atomic commits. Git bisect becomes useless with a commit that big, since it can't narrow down the origin of the bug. The people who approved this atrocity are bad software engineers. Even calling them bad "engineers" is an insult to actual engineers.

#JavaScript #NodeJS #LLM

Show thread

Pelle Wessman

@davidculley @indutny Hmm, I’ve spent the day doing the opposite with Claude Code – having it split apart a gigantic PR that arose from commit early commit often with a lack of focus – and it’s performing quite well doing that

It’s not necessarily the tool that’s the issue, it’s more how one uses it I think

Show thread

David Culley Mar 18

@voxpelli @indutny A kitchen knife can be used both to provide food and to murder someone.

An atomic bomb, by contrast, has no defensible use cases. Unlike a kitchen knife, it's not neutral.

LLMs are like atomic bombs regarding their neutrality. They are too large to be trained by individuals, they can only be trained in corporate-owned data centers. They are a non-neutral tool designed to concentrate even more power in the hands of oligarchs.

The tool is the issue in the case of LLMs.