request: CoC should explicitly forbid use of generated content (LLM, AI, etc.)

[ctcpip]

not my request, but I am not opposed to the idea. while arguably covered by the existing tenets of the CoC, it has been suggested that the CoC should explicitly prohibit the use of LLMs / AI for contributions and participation in TC39 spaces

[ljharb]

I'm not convinced it's actually a conduct violation any more than filing a silly question is, or asking in the wrong place. This is just the societal cost of LLMs being so accessible.

We should definitely indicate in contribution guidelines that it's not tolerated/accepted, and we should definitely discourage it.

[ctcpip]

agree, a blanket ban would be overly broad. I'm not sure we could nail down a qualified prohibition either. open to ideas

[bakkot]

I mostly just want to require up-front disclosure when submitting anything (including comments) written by an LLM.

[ljharb]

That seems like a perfectly fine thing to add to contributing guidelines - it can be done org-wide, and then also added to individual repos that have their own file.

[bakkot]

This came up again. Seems like there's broad support for forbidding LLM-authored content, not to include translation. Let's do it.

[ljharb]

I agree we should do it. I strongly disagree that it's a CoC issue.

[tabatkins]

It is a CoC issue, in exactly the same way that sending spam or irrelevant commentary is. To the extent that the CoC captures some expected rules of how we interact with each other, this belongs there.

I'm fine with capturing it in the contributing guidelines, and having CoC point to the contributing guidelines as something else that needs to be followed (in addition to the less technical requirements the CoC places on communication)

[ljharb]

There's overlap, sure - I guess I'll wait and see what verbiage people propose for the CoC part vs the contributing part, because I can't see how we could word those to exclude the bad usages and not punish the reasonable ones.

[bakkot]

Proposal:
[edit: tweaked wording per below]

Author your own comments

Any contributions or comments must be your own writing, not the product of LLMs or other tools. Do not prompt an LLM to expand on or explain your idea and then post the output; just provide the idea itself.

Machine translation is permissible, including translation by an LLM, but your use of translation should not introduce any new content.

[tabatkins]

s/should/must/, and maybe make "comments" more generic - "comments and contributions"?. Otherwise lgtm

[michaelficarra]

- Do not prompt an LLM to expand on your idea and then post the output; just provide the idea itself.
+ Do not prompt an LLM to expand on your idea or explain your idea for you and then post the output; just provide the idea itself.

[ctcpip]

I would think an acceptable use policy must forbid any technical contributions where an LLM is involved. Strict machine translation is probably fine.

[MylesBorins]

I fail to see why this is necessary.

Spam is spam is spam.

If an LLM is used to spam that feels irrelevant to the fact that it is spam. If an LLM could, for example, help a non English speaker engage in good faith would that not be a net benefit?

Similarly, why should contributions leveraging an LLM be banned? I could train a model entirely on historical spec text and leverage that to help me make decisions about changes to the spec. Why should that be banned?

[bakkot]

@ctcpip "any technical contributions where an LLM is involved" is definitely too strong. Many people use some form of LLM-based autocomplete without even thinking about it and I don't think we can reasonably ask them not to. Even less insidiously, many people use them as rubber ducks, and I think it would be pretty absurd to ban contributions that you've talked over with an LLM. The property I want to preserve is that you're meaningfully the author of whatever it is you're submitting.

@MylesBorins It's necessary because we've seen a number of occasions when people have submitted issues or comments which were (it eventually became clear) authored by an LLM, and not a particularly coherent one. (See most recently here.) When people respond they feed that into the LLM and then post the LLM's response. These aren't "spam" in the sense of being self-promotion or off-topic or something like that, and there is nothing in the current CoC which clearly prohibits doing this. I want to prohibit this because reading these threads, much less responding, is a massive waste of my time.

I don't have a problem with you leveraging LLMs to make decisions, if you want to. I do have a problem with you submitting things which an LLM wrote. I've tried to narrowly scope the prohibition to the latter case in my suggestion above.

[michaelficarra]

@MylesBorins The thing we're tying to ban is foisting the burden of filtering out garbage onto TC39 delegates who have to triage these inputs. If you use an LLM for ideas and then write your own proposal, that's fine. If you feed an idea to an LLM and have it write a proposal, then post whatever comes out of it to one of our forums, that is not fine.