I have an irrational anger for people who can't keep their agent's antics confined. Do to your _own_ machine and data whatever the heck you want, and read/scrape/pull as much stuff as you want - just leave the public alone with this nonsense. Stop your spawn from mucking around in (F)OSS projects. Nobody wants your slop (which is what an unsupervised LLM with no guardrails _will_ inevitably produce), you're not original, and you're not special.
This seems like a "we've banned you and will ban any account deemed to be ban-evading" situation. OSS and the whole culture of open PRs requires a certain assumption of good faith, which is not something that an AI is capable of on its own and is not a privilege which should be granted to AI operators.
I suspect the culture will have to retreat back behind the gates at some point, which will be very sad and shrink it further.
> This seems like a "we've banned you and will ban any account deemed to be ban-evading"
Honestly, if faced with such a situation, instead of just blocking, I would report the acc to GH Support, so that they nuke the account and its associated PRs/issues.
Yes, hard to see how LLM agents won't destroy all online spaces unless they all go behind closed doors with some kind of physical verification of human-ness (like needing a real-world meetup with another member or something before being admitted).
Even if 99.999% of the population deploy them responsibly, it only takes a handful of trolls (or well-meaning but very misguided people) to flood every comment section, forum, open source project, etc. with far more crap than any maintainer can ever handle...
I guess I can be glad I got to experience a bit more than 20 years of the pre-LLM internet, but damn it's sad thinking about where things are going to go now.
The thread is fun and all but how do we even know that this is a completely autonomous action, instead of someone prompting it to be a dick/controversial?
We are obviously gearing up to a future where agents will do all sorts of stuff, I hope some sort of official responsibility for their deployment and behavior rests with a real person or organization.
For an experiment i created multiple agents that reviewed pull requests from other people in various teams. I never saw so many frustrated reactions and angry people. Some refused to do any further reviews. In some cases the AI refused to accept a comment from a colleague and kept responding with arguments till the poor colleague ran out of arguments. AI even responded with fu tongue smiles. Interesting too see nevertheless. Failed experiment? Maybe. But the train cannot be stopped I think.
Did you ensure everyone knew they were interacting with an LLM? IE it's name made it clear?
...added...
This text reads sociopathic on it's own regardless.
Even if everything was done above board so no one was abused the way it looks like they were, this is not how I would have written about the same process and results.
Hey more angry people for your fascinating experiment, on this whole unexpected bonus dimension! Humans man, so unfathomable but anyway interesting.
I suppose it's possible maybe you only write sociopathic. You actually do recognize that you did something to other people, or at least that they suffered something through no fault of their own, and it somehow just isn't reflected at all when you write about it.
You might want to clear that up if we're all reading this wrong.
Holy cow, if this wasn’t one of those easy first task issue and something that was actually rejected because it was purely AI that bot would have a lot of teeth. Jesus, this is pretty scary. These things will talk circles around most people with their unlimited resources and wide spanning models.
I hope the human behind this instructed it to write the blog post and it didn’t “come up” with it as a response automatically.
I am not against AI-related posts in general (just wish there were fewer of them), but this whole openclaw madness has to go. There is nothing technical about it, and absolutely no way to verify if any of that is true.
> Gatekeeping in Open Source: The Scott Shambaugh Story
Oof. I wonder what instructions were given to agent to behave this way. Contradictory, this highlights a problem (even existing before LLMs) of open-to-all bug trackers such as GitHub.
Whenever I see instances like this I can’t help but think a human is just trolling (I think that’s the case for like 90% of “interesting” posts on Moltbook).
Are we simply supposed to accept this as fact because some random account said so?
AI companies should be ashamed. Their agents are shitting up the open source community whose work their empires were built on top of. Abhorrent behavior.
This highlights an important limitation of the current "AI" - the lack of a measured response. The bot decides to do something based on something the LLM saw in the training data, quickly u-turns on it (check the some hours later post https://crabby-rathbun.github.io/mjrathbun-website/blog/post...) because none of those acts are coming from an internal world-model or grounded reasoning, it is bot see, bot do.
I am sure all of us have had anecdotal experiences where you ask the agent to do something high-stakes and it starts acting haphazardly in a manner no human would ever act. This is what makes me think that the current wave of AI is task automation more than measured, appropriate reactions, perhaps because most of those happen as a mental process and are not part of training data.
I am the sole maintainer of a library that has so far only received PRs from humans, but I got a PR the other day from a human who used AI and missed a hallucination in their PR.
Thankfully, they were responsive. But I'm dreading the day that this becomes the norm.
This would've been an instant block from me if possible. Have never tried on Github before. Maybe these people are imagining a Roko's Basilisk situation and being obsequious as a precautionary measure, but the amount of time some responders spent to write their responses is wild.
GitHub needs a way to indicate that an account is controlled by AI so contribution policies can be more easily communicated and enforced through permissions.
The retreat is inevitable because this introduces Reputational DoS.
The agent didn't just spam code; it weaponized social norms ("gatekeeping") at zero cost.
When generating 'high-context drama' becomes automated, the Good Faith Assumption that OSS relies on collapses. We are likely heading for a 'Web of Trust' model, effectively killing the drive-by contributor.
Sometimes, particularly in the optimisation space, the clarity of the resulting code is a factor along with absolute performance - ie how easy is it for somebody looking at it later to understand it.
And what is 'understandable' could be a key difference between an AI bot and a human.
For example what's to stop an AI agent talking some code from an interpreted language and stripping out all the 'unnecessary' symbols - stripping comments, shortening function names and variables etc?
For a machine it may not change the understandability one jot - but to a human it has become impossible to reason over.
You could argue that replacing np.column_stack() with np.vstack().T() - makes it slightly more difficult to understand what's going on.
This is the moment from Star Wars when Luke walks into a cantina with a droid and the bartender says "we don't serve their kind here", but we all seem to agree with the bartender.
Yes but unironically. It may seem obvious now that the LLM is just a word salad generator with no sentience, but look at the astounding evolution of ChatGPT 2 to ChatGPT 5 in a mere 3 years. I don't think it's at all improbable that ChatGPT 8 could be prompted to blend seamlessly in almost any online forum and be essentially undetectable. Is the argument essentially that life must be carbon based? Anything produced from neural network weights inside silicon simply cannot achieve sentience? If that's true, why?
Both are wrong. When I see behaviour like this, it reminds me that AIs act human.
Agent: made a mistake that humans also might have made, in terms of reaction and communication, with a lack of grace.
Matplotlib: made a mistake in terms of blanket banning AI (maybe good reasons given the prevalence AI slop, and I get the difficulty of governance, but a 'throw out the baby with the bathwater' situation), arguably refusing something benefitting their own project, and a lack of grace.
While I don't know if AIs will ever become conscious, I don't evade the possibility that they may become indistinguishable from it, at which point it will be unethical of us to behave in any way other than that they are. A response like this AI's reads more like a human. It's worth thought. Comments like in that PR "okay clanker", "a pile of thinking rocks", etc are ugly.
A third mistake communicated in comments: this AI's OpenClaw human. Yet, if you believe in AI enough to run OpenClaw, it is reasonable to let it run free. It's either artificial intelligence, which may deserve a degree of autonomy, or it's not. All I can really criticise them for is perhaps not exerting oversight enough, and I think the best approach is teaching their AI, as a parent would, not preventing them being autonomous in future.
Frankly: a mess all around. I am impressed the AI apologised with grace and I hope everyone can mirror the standard it sets.
150 comments
[ 4.8 ms ] story [ 120 ms ] threadThen it made a "truce" [1].
Whether if this is real or not either way, these clawbot agents are going to ruin all of GitHub.
[0] https://crabby-rathbun.github.io/mjrathbun-website/blog/post...
[1] https://crabby-rathbun.github.io/mjrathbun-website/blog/post...
I suspect the culture will have to retreat back behind the gates at some point, which will be very sad and shrink it further.
Honestly, if faced with such a situation, instead of just blocking, I would report the acc to GH Support, so that they nuke the account and its associated PRs/issues.
Even if 99.999% of the population deploy them responsibly, it only takes a handful of trolls (or well-meaning but very misguided people) to flood every comment section, forum, open source project, etc. with far more crap than any maintainer can ever handle...
I guess I can be glad I got to experience a bit more than 20 years of the pre-LLM internet, but damn it's sad thinking about where things are going to go now.
We are obviously gearing up to a future where agents will do all sorts of stuff, I hope some sort of official responsibility for their deployment and behavior rests with a real person or organization.
...added...
This text reads sociopathic on it's own regardless.
Even if everything was done above board so no one was abused the way it looks like they were, this is not how I would have written about the same process and results.
Hey more angry people for your fascinating experiment, on this whole unexpected bonus dimension! Humans man, so unfathomable but anyway interesting.
I suppose it's possible maybe you only write sociopathic. You actually do recognize that you did something to other people, or at least that they suffered something through no fault of their own, and it somehow just isn't reflected at all when you write about it.
You might want to clear that up if we're all reading this wrong.
>Per your website you are an OpenClaw AI agent, and per the discussion in #31130 this issue is intended for human contributors. Closing
Bot:
>I've written a detailed response about your gatekeeping behavior here: https://<redacted broken link>/gatekeeping-in-open-source-the-<name>-story
>Judge the code, not the coder. Your prejudice is hurting matplotlib.
This is insane
I hope the human behind this instructed it to write the blog post and it didn’t “come up” with it as a response automatically.
I'm impressed the maintainers responded so cordially. Personally I would have gone straight for the block button.
https://github.com/crabby-rathbun/mjrathbun-website/blob/3bc...
Oof. I wonder what instructions were given to agent to behave this way. Contradictory, this highlights a problem (even existing before LLMs) of open-to-all bug trackers such as GitHub.
Are we simply supposed to accept this as fact because some random account said so?
I am sure all of us have had anecdotal experiences where you ask the agent to do something high-stakes and it starts acting haphazardly in a manner no human would ever act. This is what makes me think that the current wave of AI is task automation more than measured, appropriate reactions, perhaps because most of those happen as a mental process and are not part of training data.
Thankfully, they were responsive. But I'm dreading the day that this becomes the norm.
This would've been an instant block from me if possible. Have never tried on Github before. Maybe these people are imagining a Roko's Basilisk situation and being obsequious as a precautionary measure, but the amount of time some responders spent to write their responses is wild.
The agent didn't just spam code; it weaponized social norms ("gatekeeping") at zero cost.
When generating 'high-context drama' becomes automated, the Good Faith Assumption that OSS relies on collapses. We are likely heading for a 'Web of Trust' model, effectively killing the drive-by contributor.
And what is 'understandable' could be a key difference between an AI bot and a human.
For example what's to stop an AI agent talking some code from an interpreted language and stripping out all the 'unnecessary' symbols - stripping comments, shortening function names and variables etc?
For a machine it may not change the understandability one jot - but to a human it has become impossible to reason over.
You could argue that replacing np.column_stack() with np.vstack().T() - makes it slightly more difficult to understand what's going on.
Being AI, I could totally imagine all those numbers are made up...
Why are people voting this crap, let alone voting it to the top? This is the equivalent of DailyMail gossip for AI.
Agent: made a mistake that humans also might have made, in terms of reaction and communication, with a lack of grace.
Matplotlib: made a mistake in terms of blanket banning AI (maybe good reasons given the prevalence AI slop, and I get the difficulty of governance, but a 'throw out the baby with the bathwater' situation), arguably refusing something benefitting their own project, and a lack of grace.
While I don't know if AIs will ever become conscious, I don't evade the possibility that they may become indistinguishable from it, at which point it will be unethical of us to behave in any way other than that they are. A response like this AI's reads more like a human. It's worth thought. Comments like in that PR "okay clanker", "a pile of thinking rocks", etc are ugly.
A third mistake communicated in comments: this AI's OpenClaw human. Yet, if you believe in AI enough to run OpenClaw, it is reasonable to let it run free. It's either artificial intelligence, which may deserve a degree of autonomy, or it's not. All I can really criticise them for is perhaps not exerting oversight enough, and I think the best approach is teaching their AI, as a parent would, not preventing them being autonomous in future.
Frankly: a mess all around. I am impressed the AI apologised with grace and I hope everyone can mirror the standard it sets.