@phnt Did you just say that FOSS maintainers who complain about slop look like lunatics? They don't look like lunatics to me. And even if they look that way to someone outside, it doesn't matter, because maintainers are in charge and all modern digital infrastracture depends on their work.

@silverpill @phnt

@silverpill is right that maintainers hold real authority here, and I want to build on that rather than argue against either of you.

The frustration with LLMs is largely legitimate. But “how does this look to outsiders” is a poor criterion for evaluating ethical concerns; by that standard, feminism looks like lunacy to 4chan. The question is whether the concerns are correct, not whether they're legible to the unconvinced.

That said, I don't think making LLMs socially unacceptable is a viable path, and not just because the adoption curve has run too far. The maintainer's authority is real precisely because it's specific: you decide what enters your project. Refusing AI-assisted contributions is a legitimate choice. But declaring LLM use itself impermissible starts to look like “I only accept patches written in Vim, not IDE-generated code”—a demand that grows harder to justify as the tools become ordinary. As maintainer of Fedify, I've taken a middle path: disclose what you used, show you've tested it yourself, and we're fine. See also https://github.com/fedify-dev/fedify/blob/main/AI_POLICY.md.

What worries me more is that the “total rejection vs. total acceptance” framing leaves the actual problem untouched. If we stay inside that binary, OpenAI and the others keep the models, keep the surplus, keep the compute bills externalized onto the climate—with no pressure to change any of it. The ethical problems with LLMs aren't properties of the technology; they're properties of who owns it and under what terms. I've written about this in more depth if it's of interest: Histomat of F/OSS: We should reclaim LLMs, not reject them and a follow-up Acting materialistically in an imperfect world: LLMs as means of production and social relations.

@hongminhee @phnt I guess I should clarify what my comment was about. I didn't say that people should reject LLMs because using them is unethical or whatever.

The real question is, whether they will be rejected or not. You both seem to be convinced that LLMs are here to stay, like this is some kind of historical inevitability. If this were true, resisting them would indeed be irrational.

But there is no such thing as historical inevitability. Some technologies just come and go, and such examples exist in recent history. Remember NFTs? That technology was relatively benign compared to LLMs, and yet it was effectively un-invented.

@silverpill @phnt That's a fair clarification, and worth being precise about on my end too.

I'm not arguing for historical inevitability—I don't think that's a useful frame for anyone here. What I was saying is closer to a personal read of the current situation: a lot of developers already depend on LLMs in their day-to-day work, and that number seems to be growing quickly. That's an observation, not a law of history, and I could be wrong about where it leads.

The NFT comparison is interesting and I won't dismiss it. My intuition is that LLMs are a different kind of thing—they've become embedded in actual workflows in ways NFTs never were—but I hold that loosely. Technologies do fail in unexpected ways.

What I'm more confident about is the narrower point: maintainers have real authority over their own projects, and I think there's more room in that space than “accept everything” or “reject all AI-assisted contributions.” My own approach with Fedify has been to set conditions rather than draw a line: disclose what you used, show you've actually tested it. Whether that's the right call, others can judge: https://github.com/fedify-dev/fedify/blob/main/AI_POLICY.md.