Re: Dealing with LLMs in IETF discussions draft

[Stephen Farrell <[e-mail]>]

Hiya,

On 03/08/2026 21:56, Brian E Carpenter wrote:
> +1, but that doesn't stop humans using AI assistance. For example,
> using AI to _shorten_ a draft message before sending it, or for
> summarising a complex thread. 

I think there are dangers inherent in that, e.g. LLMs will
select a meaning, where a person might know to ask what a
sender meant. I see that lack of uncertainty/curiosity as
being one of the significant dangers of LLM use (and one of
the current "tells" that LLMs are being used.)

> Perhaps a wiki about good ways to use
> AI in the IETF would be the most fruitful approach.

Too early to say IMO. I think we will hit some disagreements
about what's ok/not-ok or good/bad. Seems unlikely a wiki
that gets edited back & forth between those would be the right
end point;-)

I don't think we want RFCs to document this stuff, but we may
need to run consensus calls somehow after we've gotten closer
to documenting the different perspectives.

Cheers,
S.