Mastodawn

RE: https://social.treehouse.systems/@swags/116235865719541791

Today was the SuperTux 0.7.0 release (yay!), but some of us were alarmed by the "Claude contributed to this repo" message on GitHub (oh no!)

Well, I've learned a few things from the discussion about this topic.

A single PR "co-authored by Claude" has slipped through. The code was written by a human (according to the commit), then properly reviewed and eventually merged by a maintainer. The Claude contribution is attributed solely to the code comments added through a mid-PR-flow commit update.

And now the commit history will show Claude as a contributor. It might be impossible to rewrite it. Rewriting kind of goes against the idea of having the git history, too. Even if that was possible, it's an extra effort which not all maintainers can prioritise.

So: seeing "Claude was here" in a GitHub repo is a "red flag", but it shouldn't be "therefore guilty of slop and LLM proliferation by default"; please use your head/best judgement.

I guess it is a cautionary tale for the maintainers, too

Show thread

Sharp Cheddar Goblin 2d ago

@nina_kali_nina Problem is, once that red flag flies, even if you check the commits and find out it's innocuous you have lost your ability to determine if something less innocuous appears. You have to check for new Claude commits continuously.

For many it may be easier just to move on to something else that has no red flags rather than commit to constant vigilance.

Show thread

LisPi

@SharpCheddarGoblin @nina_kali_nina That is pretty much inherently inevitable from using tools controlled by the opposition.

Github is adversarial here. It's purely incidental the tool is available at all, nevermind remotely reliable in any sense.

Better Free Software tools that are not based on SaaSS are absolutely necessary going forward.