Re: Dealing with LLMs in IETF discussions draft

On 10-Aug-26 10:44, Stephen Farrell wrote:
> 
> Hiya,
> 
> On 08/08/2026 21:30, Brian E Carpenter wrote:
>> On 08-Aug-26 00:37, S Moonesamy wrote:
>>> Hi Chong,
>>>
>>> One of the points in your email was about who makes the technical
>>> claims in an Internet-Draft.  At the other end there is a person who
>>> reviews the technical claims and dispute/disagree with those
>>> claims.  That person (it can be more than one person) previously did
>>> not use a LLM to assess the correctness of the claims or to submit
>>> his/her disagreement.  Is it okay for the person to use a LLM to do that?
>>
>> I think that question is unimportant.
> 
> I very much disagree with Brian. ISTM crucial that readers (and
> there are many of those) are generally confident that the opinions
> expressed in a message are those of the sender, and not opinions
> being mimiced by an LLM. (The fact that there are many readers for
> each sent-thing, IMO imposes a higher burden on the sender.)

I thought that initially, as you can see from earlier versions of
my RSWG draft, but comments by several RSWG participants convinced
me otherwise. My personal preference is that writers should acknowledge
when they've used significant AI assist, but the message that a person
sends under their name is their message, however it was created.

I think the problem that LLMs have created for the IETF is real,
and it's that it allows people to quickly generate floods of plausible
I-Ds. And if we start to see robotic floods of plausible emails,
that will also be a real problem which will demand a technical solution.

     Brian



> 
> That aspect is not at all addressed by the sender naming themselves
> as responsible for the content
> 
> Cheers,
> S.
>