Re: LLMs and Agents usage in the CCG

st 8. 4. 2026 v 17:26 odesílatel Mahmoud Alkhraishi <mahmoud@mavennet.com>
napsal:

> Hi all,
>
> The last few weeks have brought up several issues around the usage of LLMs
> and Agents and as chairs we wanted to facilitate discussions. We currently
> have a rule that blocks bots on the mailing list. This will not be changing.
>
> We will adhere to the W3C rules on LLM usage in standards when they are
> fully implemented. They are currently working on it here:
> https://w3c.github.io/AB-public/position-statements/llms-standards/ please
> feel free to contribute.
>
> As there are no current rules in place we want to gather community
> feedback and thoughts and attempt to implement a ruleset in the interim. We
> see a few options:
>
>    1. Ban all LLM/Agents from the mailing list and any spec work
>    2. Ban all LLM/Agents from the mailing list. Allow usage of both LLMs
>    and Agents in spec work if it is disclosed, with the understanding that
>    there is always a human in the loop reviewing and approving any work output.
>    3. Ban all LLM/Agents from the mailing list. Allow usage of LLMs in
>    spec work, disallow any autonomous agents, with the understanding that
>    there is always a human in the loop reviewing and approving any work output.
>    4. ??? —> any other positions or lines in the sand you wish to bring up
>
>
Bear in mind that LLMs help participation for people with disabilities, and
others.

Generally you want to protect against IP violations. I have seen a few in
code and PRs, across W3C groups.

The other thing is that LLMs make mistakes. So it is good to be mindful of
people's time. However, in general frontier machine content is on a par
with human content, when it is kept within normal boundaries, and will only
get better.


>
>    1.
>
>
> Things to keep in mind:
>
>    1. The reason behind banning them from the mailing list is because it
>    just adds lots of noise. Generally, we believe if you aren’t willing to put
>    in the time to write something, why should the community put in the time to
>    read it.
>    2. Many people in the community struggle with communication in English
>    and LLMs help with accessibility
>    3. LLMs are usually very verbose, making it very hard to read/review
>    text written by an LLM and adds a lot of cognitive overhead.
>    4. LLMs can be subtly wrong when generating technical docs, and
>    reading overly verbose text makes it easy for nonsense to slip in.
>
>

I think (4) is important. Especially with merging and branch protection. I
have seen a huge number of self merged and forced push code across w3c
groups which have been breaking changes. Including merges of unlicensed
material, code, syntax errors and hallucinated terms.  Reviewing branch
protection and permissions, especially when it is overridden with forced
pushes, I believe will help alot.


>
>    1.
>
>
>
>
>
>
>
> --
> Regards,
> Mahmoud Alkhraishi
>