@glyph Typos and misspellings used to be great starter bugs for new contributors. The places where the bug was trivial but learning the social and technical parts of the contribution process was the real hurdle. Letting an AI solve those problems is a tactical win and a strategic disaster.

@glyph If I ask a program to check my spelling and it finds and alerts me to errors, that's one thing. That's "centaur".

If someone else's program non-consensually checks my spelling and brings every apparent typo to my attention, that's something else. That's "reverse-centaur" and "someone else" can go fuck themselves.

@glyph Yeah, I think fixing the spelling at this point in time, even in a separate, handmade PR, will only encourage them.

@glyph close them. If the mistakes are actually causing confusion, someone real will presumably fix them. Using codespell and languagetool yourself to proactively fix as much detected stuff as possible in one go is also an option. Neither of them are LLM as far as I could tell.

@glyph close the PR, block the bot, and if the spelling error bothers you, fix it yourself. if a spam bot accidentally identifies a legitimate issue, and the issue is still legitimate after you independently confirm it, then the issue is still legitimate. however, that does not stop the spam bot from being a spam bot. it's still an automated process attempting to manufacture a fake history of contributing to "high value" projects to be sold for money for god knows what purpose down the way

@glyph Personally I'd block the contribution... and in this case probably the account sending it, though, that's mostly because they're using an incel meme as an avatar, and that along with not respecting the LLM policy screams "nazi" to me.
If anyone cares, they can go fix the typos the old-fashioned way, but, I think the higher priority is maintaining the policy and not giving ground on it just because it's only a "small" concession. b/c of the "give an inch, they take a mile" problem.

@glyph I'm on team "leave 'em in". Definitely just close the bot PR. If the spelling errors do not matter to meaning, don't bother. If they do matter to meaning, use the opportunity to make the writing better, not just fix the spelling issue. Maybe learn new synonyms!

(I did not read the PR, I've had enough psychic damage for today)

@glyph if the devil pointed out my shoe was untied and offered to tie it, I’d turn them down, but I’d still tie it.

@glyph would you like me to hand craft a pr that adds https://github.com/crate-ci/typos to ci

@glyph Interesting indeed

Honestly, I tend to think "using AI" is not the most useful concept because of cases like this. I prefer to see when maintainers are more precise about how they do or don't want LLMs to be used in connection with their project.

That being said... if you care about collecting random people's opinions (🤷), I'd say that seeing an LLM-generated PR, closing it, and making a new PR with the same diff is in roughly the same class of "using AI" as having an LLM make some changes and reviewing/refactoring/building on them locally before manually submitting them as a PR. E.g. I don't think it would pass muster in a project whose LLM policy is that LLMs can't be used to aid development in any way.

@glyph This is an issue I've been talking about for a while and why I *don't* think the solution is to say "any amount of AI use is banned, regardless of how small," despite being a huge hater of generative AI.

If someone reads an article they don't realize was written by AI to learn about a popular design pattern, are they *never* allowed to use that pattern because it was technically suggested by an AI?

@glyph this doesn't really seem like a complicated ethics problem. Someone spent tokens, found something real, and trivial, with their spend. If you don't want to encourage the behavior don't merge their change, go make a similar one yourself so they don't get credit, removing the incentive to continue this behavior. Or just don't do anything, reporsitory is a funny word that hurts no one.

@davidism yeah this is the solution I am angling towards. definitely don’t want to reward/credit the reporter

@tartley so just … leave the errors in? this feels weird.

@tartley not necessarily *wrong*, mind you. just weird

@glyph @tartley I heard worse was better ;)

@mhoye at the level of the system as a whole, you are 100% correct. but it doesn’t quite say what to do when we are non-consensually exposed to this from a thousand directions at once. particularly when some of the people doing this are, well, people, manually driving the AI as a tool to try to help them get involved, and others are just weapons-free rules of engagement openclaw machine guns who are not available to be reasoned with

@mhoye adi has cut the gordian knot https://github.com/twisted/towncrier/pull/748