The Rust Evangelism Task Force has declared "ethics" to be out of scope. And that's going as well as you might guess:
This document establishes a policy for how LLMs can be used when contributing to rust-lang/rust. [...] No comment on this PR may mention the following topics:
• Long-term social or economic impact of LLMs
• The environmental impact of LLMs
• Anything to do with the copyright status of LLM output
• Moral judgements about people who use LLMs
@jwz the excerpt left me with an impression that things are worse than they are
to add a few more quotes to balance that take:
> We still consider these topics to be important, we simply do not believe this is the right place to discuss them.
> This intentionally does not address the moral, social, and environmental impacts of LLMs. These topics have been extensively discussed on Zulip without reaching consensus, but this policy is relevant regardless of the outcome of these discussions.
@jwz and most importantly, the policy itself
> The policy's guidelines are roughly as follows:
> It's fine to use LLMs to answer questions, analyze, distill, refine, check, suggest, review. But not to **create**.
@flpvsk @jwz agreed - it's largely just saying "we need a policy either way. constructive comments welcome, broader discussion belongs elsewhere" and that seems... fine? Github is hardly an ideal (or even good) place for heavily threading discussions. And they're correct that they need a policy, as many treat "no comment" as permission.
That said, the Zulip they link to is not publicly visible, which is rather concerning. Private discussions are fine, but they're not evidence, and they don't provide a place to go to contribute.
@groxx @flpvsk No, it's saying, "This is where we're going to decide what our policy should be, and oh by the way, the primary and most fundamental objections that many people have to using LLMs are out of bounds for this discussion."
That's not just putting your thumb on the scale, that's kicking the legs out from under the table.
@jwz @flpvsk with a large group of people, how you do get a policy written down and agreed on when both sides feel very strongly?
I would much prefer they ban LLMs entirely, but many clearly disagree and you still need enough of them to sign off on it for it to be adopted. How do you reach that point when both are brigading heavily?
@jwz @flpvsk the conversations have been happening and will continue to happen, yes? Or is there a sign that it has been stopped everywhere?
And on the ethical side, it really does seem to me that it's largely a brick wall between the two, and few cross over. The kind of unproductive fights that leads to are obvious, and happening all over. So you're kinda left with: A) fight and go nowhere (which we agree is not what they need), B) fork and the associated costs (either you leave or you kick them out), or C) moderate to try to make progress. I'm not really seeing any other options.
(I'm not deeply active in the community, maybe there are signs it's just being shut down everywhere? If there are, then I entirely agree with you)
@flpvsk Anyone who says "we want to craft this policy only around technical reasons without any discussion of ethics" is:
A) Putting their thumb on the scale
B) A massive piece of shit
@jwz i largely agree. my *charitable* read on this would be:
1. they are not ready to put together AI guidelines in it's full and final form, bc the discussion is ongoing (not set aside, just not finalized, ongoing on Zulip).
2. At the same time AI-authored PRs keep coming in, so they need something in the policy to point to to reject those.
that's the impression I got at least
@flpvsk if all they needed was a clear rule that allowed the rejection of slop PRs, then a straightforward "no slop" policy would work. "Setting aside the ethics" weaselage is a clear indication that they want to allow slop, while framing the debate in such a way that nobody is allowed to make them feel bad for doing so.
jwz
flpvsk
groxx
Jared White (ResistanceNet ✊)
Monokeros
Dźwiedziu
Matt Palmer
The Penguin of Evil
Sam Coulter
ShadSterling