Re: AI assisted code already committed?

On Wed, 01 Apr 2026 20:41:43 +0200,
"requiem." <rqm@0xde501a7a.net> wrote:
>
> Hello misc@
>
> Given the recent discussions regarding ext4fs and AI commits, can I ask what the status of these two commits are?
>
> https://github.com/openbsd/src/commit/9c2b8e445a0bdfafdd6148b1760f00aa5429627b
>
> https://github.com/openbsd/src/commit/e9af5eb5a61d189327b553b24d0d31f19c64b63f
>
> ( Originally discovered via https://mastodon.social/@mrmasterkeyboard/116329909911804392 )
>
> Are these going to stay or go?
>

Shall code which is produced by autocomplete like treesitter be banned?

I don't think so.

Shall such code be read, reviewed, and understood by a person who creates it
and reviews it?

Definitely, isn't it?

If you treat an LLM generator as autocomplete on hyper steroids, you'll
understand that as long as you read and understand and review the code which
it produces, and actually heavily rewrite it so it looks nice and not like
nonsense, it is good to be committed.

The issue isn't the technology itself.

Issue is the noise that is created by people who generate dozens of
thousands of lines of code which somehow works and try to commit it.

Thus, OpenAI restricts access to opening new Pull Requests to their Codex on
GitHub to accounts which already committed something to it, which explains the
scale of that weird behaviour BTW.

(Here I ignore copyright issue which is different story)

--
wbr, Kirill