Re: Dealing with LLMs in IETF discussions draft

--0000000000007d80be0657e78852
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

What do you mean by "know how to interpret"?

Do you mean:

1. That you can't understand the point they are making?
2. That you don't believe they are actually aware of the point they are
making?
3. That you don't want to deal with this problem?

3 is not something the IETF can help with imo.
2 is not actually a new problem, but its scale has been massively amplified=
.
1 is also not a new problem, but it is one that AI can help with : )

I have an even more cynical answer for 2: who cares?
If you got the info you needed and your experience was okay, why does it
matter if the person (without access to their tools) making the point
actually understands that point?
I guess the core of this is whether we share a view of the utility of ietf
discussion, which I believe is to build a "better" shared mental model, to
coordinate human efforts.
The problem is that mental models no longer exist solely in human minds; we
can't easily separate the tool user and the tool.

This isn't new either; I don't understand the assembly I am creating when I
write in higher order languages.
If human (non assisted) understanding truly became more important (was
rewarded more) than productivity, things might change... but this would
require dividing people and distinguishing between productivity and
accessibility tools, and I don't like the idea of where that will lead us.

As an aside, I already dislike the ability to "spend more money to think
harder about a problem," but I don't see the IETF as a community that can
do much about that, other than rewarding people for being transparent about
AI use and encouraging them to use models that consume less power and are
scoped to the difficulty of the task at hand.

Anyway, I support guidelines, naive use of AI is generally worse than
thoughtful contextually aware use of it.

Cheers,

OS






On Fri, Jul 31, 2026 at 6:37=E2=80=AFAM Stephen Farrell <stephen.farrell@cs=
.tcd.ie>
wrote:

>
> Hiya,
>
> On 31/07/2026 12:31, Orie wrote:
> > It sounds like the goal is more effective use of AI in IETF
> > discussions.
>
> Yes, that is my co-author Frank's goal. It is not mine. My
> high level goal in this is to know how to interpret messages
> emitted by others in IETF discussions, despite use of AI.
>
> Cheers,
> S.
>
>

--0000000000007d80be0657e78852
Content-Type: text/html; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

<div dir=3D"ltr">What do you mean by &quot;know how to interpret&quot;?<br>=
<br>Do you mean:<br><br>1. That you=C2=A0can&#39;t understand the point the=
y are making?<br>2. That you don&#39;t believe they are actually aware of t=
he point they are making?<br>3. That you don&#39;t want to deal with this p=
roblem?<br><br>3 is not something the IETF can help with imo.<br>2 is not a=
ctually a new problem, but its scale has been massively amplified.<br>1 is =
also not a new problem, but it is one that AI can help with : )<br><br>I ha=
ve an even more cynical answer for 2: who cares?<br>If you got the info you=
 needed and your experience was okay, why does it matter if the person (wit=
hout access to their tools) making=C2=A0the point actually understands that=
 point?<br>I guess the core of this is whether we share a view of the utili=
ty of ietf discussion, which I believe is to build a &quot;better&quot; sha=
red mental=C2=A0model, to coordinate human efforts.<br>The problem is that =
mental models no longer exist solely in human minds; we can&#39;t easily se=
parate=C2=A0the tool user and the tool.<br><br>This isn&#39;t new either; I=
 don&#39;t understand the assembly I am creating when I write in higher ord=
er languages.=C2=A0<div>If human (non assisted) understanding truly became =
more important (was rewarded more) than productivity, things might change..=
. but this would require dividing people and distinguishing between product=
ivity and accessibility tools, and I don&#39;t like the idea of where that =
will lead us.<br><br>As an aside, I already dislike the ability to &quot;sp=
end more money to think harder about a problem,&quot; but I don&#39;t see t=
he IETF as a community that can do much about that, other than rewarding pe=
ople for being transparent about AI use and encouraging them to use models =
that consume less power and are scoped to the difficulty of the task at han=
d.<br><br>Anyway, I support guidelines, naive=C2=A0use of AI is generally w=
orse than thoughtful contextually aware use of it.<br><br>Cheers,<br><br>OS=
<br>=C2=A0<br><br><br><br><br></div></div><br><div class=3D"gmail_quote gma=
il_quote_container"><div dir=3D"ltr" class=3D"gmail_attr">On Fri, Jul 31, 2=
026 at 6:37=E2=80=AFAM Stephen Farrell &lt;<a href=3D"mailto:stephen.farrel=
[email protected]">[email protected]</a>&gt; wrote:<br></div><blockquote =
class=3D"gmail_quote" style=3D"margin:0px 0px 0px 0.8ex;border-left:1px sol=
id rgb(204,204,204);padding-left:1ex"><br>
Hiya,<br>
<br>
On 31/07/2026 12:31, Orie wrote:<br>
&gt; It sounds like the goal is more effective use of AI in IETF<br>
&gt; discussions.<br>
<br>
Yes, that is my co-author Frank&#39;s goal. It is not mine. My<br>
high level goal in this is to know how to interpret messages<br>
emitted by others in IETF discussions, despite use of AI.<br>
<br>
Cheers,<br>
S.<br>
<br>
</blockquote></div>

--0000000000007d80be0657e78852--