They could make two kinds of pull requests and add much more strict criteria for public contributions. For example, they could say that the PR has to be smaller in size and well-documented for human review, otherwise it's closed by an automation.

And then if someone wants to do a larger contribution, they could have a process like making an issue, discussing the approach and then collaborating with a maintainer to get it in.

Blocking public contributions means that they want to have complete control of the project and AI is likely a good excuse to do that.

That doesn't solve the volume or quality problem. LLMs can split one giant PR into 50 smaller PRs just as easily and "well-documented" isn't something you can determine automatically.

Why is it so hard to just accept that AI PRs suck and create an enormous amount of toil?