14:41
<Aki>
Ecma is still working out our overarching AI policy, but keep in mind that IP is terribly important to be attentive toward in standards and LLMs… do not support that requirement.
15:04
<James M Snell>
Small side note... Keep in mind that Gen AI technologies *are* being integrated into legally protected assistive devices for people with various disabilities. Any AI policy does need to be sensitive to the fact that use of such is protected in most jurisdictions. Most attempts to apply blanket bans on use of AI to generate contributions aren't adequately taking this into consideration
15:06
<James M Snell>
While such use in open source or standards contributions is likely minimal or even theoretical at the moment, it still needs to be considered
15:06
<James M Snell>
(this coming from someone whose arthritis is starting to make it increasingly more difficult to type for long periods of time)
15:22
<Zb Tenerowicz (ZTZ/naugtur)>
I believe considering LLMs an assisted keyboard is generally helpful in considering the outcomes, especially responsibility for the text produced.
15:27
<Chris de Almeida>
yeah, I've had similar conversations with delegates about our AI policy as currently written, particularly with regard to inclusivity, neurodivergence, and assistive tech. I understand the position of "I don’t want to read AI prose", and I share the same sentiment, particularly with regard to AI slop and/or walls of text. but I think these concerns raise a reason to revisit our policy. the goal being to avoid unintentionally excluding people who use AI or similar tools as assistive technology, including some neurodivergent participants we need to balance concerns about AI generated contributions with inclusivity and reasonable accommodation, not just under the law but also ethically. our policy should target unaccountable or low-effort AI generated contributions without making participation harder for people who need support communicating clearly
16:27
<bakkot>
I'm a little unclear on how that works. All AI tools operate on a model of taking some machine-readable human-generated input and producing some output. I want to be given the human-generated input instead of the output.
16:29
<bakkot>
Though of course there is a great diversity of assistive tech, so it's hard to have a completely blanket policy. I agree that having clearer wording about assistive tech would be good, but I don't want it to extend to allowing LLM-authored prose in general.
17:48
<bakkot>

Thoughts on striking "not the product of large language models (LLMs) or other tools" from https://github.com/tc39/how-we-work/blob/main/AI_POLICY.md? That could be read to suggest we're forbidding all use of LLMs, which is not the intention. So it would read

This policy applies to prose contributions (e.g., ideas, comments, issues, etc.). It does not apply to code contributions, which are evaluated through our review processes.

Prose contributions and comments must be your own writing, not the product of large language models (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.

Similarly, you may use an LLM for proofreading as long as this doesn't add any new content.