PS3 Emulator Devs Politely Ask That People Stop Flooding It with AI PRs

Original Article ↗ Hacker News Discussion ↗

Maintainers of complex open-source projects like the PS3 emulator RPCS3 are increasingly frustrated by low-quality, AI-generated pull requests that create more review and maintenance work than they save. Commenters debate whether better contribution guidelines, reputation systems, or stricter gatekeeping are needed, and stress that submitting code—AI-assisted or not—should require understanding, testing, and willingness to take responsibility for it. Many see value in using LLMs privately or in personal forks, but argue that flooding high-profile repositories with unvetted AI code threatens the sustainability of free and open-source software.

AI-generated PRs and maintainer burden

  • Many commenters say AI-assisted pull requests often increase work for maintainers rather than reduce it.
  • Reviewing large, partially incorrect AI diffs is harder than writing the fix manually.
  • Maintainers describe AI PRs as “slop” when authors can’t explain or test the changes but still expect review and merge.
  • Some see this as a new “endless September” of people unfamiliar with open-source norms overwhelming projects.

LLM limitations in complex or niche codebases

  • Several people report LLMs failing badly on PS3 homebrew, console graphics, or classic Mac APIs due to sparse or subtle training data.
  • The tools sound confident and “blog-smart” but often produce subtly wrong or overcomplicated code.
  • A minority report good results when they first supply detailed docs and tooling (linters, parsers), treating the model as an assistant within a well-defined environment.

What good contributions should look like

  • Strong consensus: anyone submitting code should understand it, test it, and be willing to iterate. Motivation alone is not enough.
  • Suggestions for non-coders: help with documentation, issue triage, reproduction, design, or donate resources instead of code.
  • Forking for personal use is framed as perfectly fine, and often the right place for “works for me” AI modifications.

Gatekeeping, reputation, and process changes

  • Ideas floated: invite-only PRs, reputation systems, web-of-trust–style graphs, auto-closing issues/PRs by default with whitelisting for proven contributors.
  • Some propose repo-level CLAUDE.md/AGENTS.md to set explicit rules for AI use; others argue existing CONTRIBUTING/README already suffice.
  • A few advocate simply banning unsolicited or AI-generated PRs for small and medium projects.

Responsibility, ethics, and copyright

  • One camp emphasizes personal responsibility: if you submit it, you own it, regardless of tooling.
  • Others argue the tools themselves are at fault for confidently misleading non-experts.
  • There is concern that AI-generated code may be uncopyrightable or legally murky, making some maintainers unwilling to accept it.