Free software has two cases for free: free as in 'free drink', and free as in 'free to change'.

The first use of free is definitely improved by AI, it's the second use of free that bothers the author. There's too much AI slop and PRs, and it's impossible for the author to identify 'good PRs' from bad ones. There's too much volume of change to sift through, and so the ability to change software is stifled.

What can you do then? Either close the repo to the public, or limit who can send you feedback: as it is now, there's too much volume for reviewers to keep. It's too difficult to validate 'AI Slop' from quality code written by someone who 'has put in the time' and done their due diligence. Someone who has done this wastes less time for the reviewer, and probably improves the quality of the 'free beer' product.

Are you suggesting, AI hasn't improved the first case of free, 'free drink'? It's certainly an issue for the second case, 'free to change', but it seems like there are things that can be done to help with those problems: but it adds more churn and probably requires more time (unless I'm misunderstanding things here: which is very possible).