Call to ban AI commits

[lumi]

I feel rules like this should only be enforced in obvious cases.

Slop may get in, but such is life. The stance of the project is, in my opinion, the most important thing.

[tejing]

@lumi

I feel rules like this should only be enforced in obvious cases.

Which is essentially the same thing as establishing the rule to be whatever defines an “obvious case.”

Which is basically what the current policy does.

[fgaz]

@cafkafk

For me, one of the most interesting open questions to the segment of the community that’s interested in a blanket ban is: How would one actually do this in practice? It’s by contradiction near impossible to positively prove something is made by an LLM, and the counter to that, it’s very easy to accuse others of using an LLM.

@QuBe

There is no way to clearly separate commits which include some AI output from ones without

The following article is a response to these exact questions and it ties right into @lumi’s points.
I won’t restate it since it’s already short and well written so please do follow the link:

alexia's space: AI Contribution Policy

[QuBe]

This changed my mind a bit, thanks !

[nrabulinski]

@cafkafk

To be clear, none of that is a “gotcha”, it’s more of an actual brain dump of how I see this in total. I think that the risks you’re showing are all valid. I also think most are not unique. But I also see that LLMs manage to package a lot of bad things into a neat package, and that it is a problem.

All of your refutals are just “but we already do bad things so let’s keep doing more bad things”. How about we don’t introduce more exploitative and harmful industries, actually? Then we’d have more power to fix decades of baggage we have. It’s insane to me how people can say “yeah LLMs are bad but like, manufacturing tech is bad so we can’t do anything about it, let’s allow everything”. Maybe if we finally started taking hard stances we could start fixing the outstanding issues, instead of being defeatist and just letting more and more bad things happen.

[lumi]

Sure, but the current policy is not a full ban on GenAI.

[cafkafk]

@nrabulinski

yeah LLMs are bad but like, manufacturing tech is bad so we can’t do anything about it, let’s allow everything

I wrote a very extensive reply here, deleted it, wrote it again, deleted it. But in short, no, this is not my position. I think if that’s what you got from all I wrote so far, I’m not sure more text is gonna help clarify.

[nrabulinski]

Help me understand then because all of your points are “this is a valid concern, but [other bad thing is happening]”. How is this not a defeatist position?

[cafkafk]

@nrabulinski

Help me understand then because all of your points are “this is a valid concern, but [other bad thing is happening]”. How is this not a defeatist position?

I mean sure, I can see that reading, fair. My point isn’t to say “valid concern, but [other bad thing]”. I think the concern is real, but the obvious flat out ban doesn’t solve it, because there isn’t a majority that wants a flat out ban.

I’ve tried, maybe imperfectly, to build an iron man position for banning LLMs, so the idea couldn’t get waved away with cheap shots. I’ve then point by point tried to counterbalance that, so each argument build on it isn’t just naive.

You don’t do that against a position you’re trying to dodge, and it is not the fatalism move to engage.

@cafkafk

With all that. I’m still unable to see an actual policy that simply says “LLMs aren’t allowed” that would be effective at solving these problems. Or would at most solve some of those problems, partially, while opening up other massive risks.

That isn’t some rhetorical device. If someone has something more concrete than “ban LLMs”, something that accounts for the pitfalls I’ve outlined, which are also required negotiating points to make this happen politically against a 61% majority, that is exactly the input I want.

I’ve literally even laid out the blueprint that I’ve used to lift of on of these problems from the “LLMs are ontolotically evil” claim into the “LLMs can produce incorrect code, and we can all agree this is bad” as an template for how to turn these grievances into real policy that can be advocated against a majority position.

I’m also here doing the actual legwork of gathering the rough, unformed version of those grievances so I can turn them into actual policy that can pass. And where possible, I’ll take those into the SC and other internal team debates I’m part of and represent them faithfully, even if it isn’t my personal position, so those view also actually make it into the room. Because the 31% also matter, and I don’t want to leave anything on the table when there is such a big divergence.

I am sure that these policies could be more agreeable to both parties, and a lot of it breaks down because of an absolute stance that I think should make a real account of its grievances and would likely find that the majority would be fine with meeting those as well.

---

Pre litigating myself: yes, just because there isn’t a majority, that doesn’t make for a call about what “should be”. That is a fair, separate discussion, that I am not in any way blocking. But I am specifically here in the context of being a member of the steering committee that wants to represent people and pass policy.

That said, let’s perhaps move away from making this thread about me, and make it about the actual proposal? I don’t think its fair or productive to litigate me personally for engaging. I’ll step back so this can be about the proposal and not about me (and I’ve run out of clarity to be honest), but I’m still here for genuinely new input.

[7c6f434c]

@fgaz

The following article is a response to these exact questions and it ties right into @lumi’s points.

This way one creates a new issue of moderation, which is contentious by design, and has a split among core contributors. While we might not yet have run out of our budget of pushing out key contributors by maximising moral splits, we will probably run out of moderation capacity via dragging moderators into yet another flame war, again.

[lumi]

Even though the split here is sadly 62% + 7% that are ok with slop, I do feel like this is not the whole story.

Someone being prevented from using GenAI is generally a minor annoyance to them, but someone morally against it seeing it in the project is a huge deal.

So, maybe still best to err on the side of caution and go for a full ban?

[samestep]

@lumi

62% + 7% that are ok with slop

Can we please be a little more reasonable / less strawmanny here? The current policy is very clearly against slop. The question of this thread is whether to also disallow AI-assisted contributions that aren’t just slop.

[waffle8946]

Non-slop AI contributions don’t exist, is the point. If it’s a menial task like renaming a var, we have deterministic IDEs for that already. Using an AI for a task like that just opens up the possibility for more errors.

[samestep]

That’s a statement you have to actually defend. Did you not read the first message I sent in this thread?

[waffle8946]

No, sloppers need to defend why their slop isn’t slop.
It’s called having standards.

[samestep]

Again, refer to the first message I sent in this thread.

[waffle8946]

Doesn’t change my position. LLMs are non-deterministic and just guess what text to spit out, by design.

[samestep]

Do you have any criticisms of the code I linked?

[waffle8946]

No, I’m not interested in writing the prompts on your behalf that you send to your chatbot. I’ll just be sure to ignore all of your PRs henceforth.

[lumi]

I use the term “slop” to refer to all GenAI outputs, regardless of quality.

I don’t think it’s useful to argue about definitions, though. My point is that they’re ok with contributions created partially or fully using GenAI.