RE: IETF AI-Pref Wiki

7 April 2026

 

Dear Laurent,

 

I have not followed the discussion closely but would like to make an
observation about the wiki:

 

(i)               There is a distinction between “training” and “use” – is
“Training” a subcategory of “use” in your nomenclature?

(ii)              The first item (1.) under training is affirmative –
“prefer that a blog entry BE used to train AI models”, whereas the first
item under “use” (1.) is phrased in the negative – “prefer tha blog entry
NOT be used. What is the reason for going affirmative of training and
negative on use?

(iii)            Whilst “Training” and “Use” are categories, all items under
“Training” still refer to “use” as a verb, e.g. prefer that a blog entry be
“used” to train AI models. Would it not make sense to avoid the word “use”
to avoid confusion, e.g. “prefer that a blog entry BE an in-put for training
AI models.”

(iv)            All items appear to stick to plural “AI models”, yet under
“Training” item four (4.) uses singular and the term “general AI model”.
Question: why plural and singular – perhaps all singular or all in plural?
Also, is AI model a wider category encompassing “general AI model” or is it
alternative, meaning not “general” AI model, or what is it? It seems
ambiguous to me. 

(v)              Under “Use” – the term “AI” is omitted, and the word
“model” is used – is it AI model, why singular, does it include “general AI
model”?

(vi)            Under “Presentation” the term “content” is deployed which
probably means a portion of a text or other human expression, as opposed to
the whole, e.g. “content from an article” could mean a portion or an
illustration or image associated with an article.  Seems imprecise. The term
content is not deployed in a very clear use - how about using “excerpt” or
“portion”.  “Content” may be appropriate if you wish to denote also human
expression that is no longer in-copyright, or non-original human expression,
e.g. a database protected under the sui generis right in the EU and some
other countries, e.g. South Korea, Mexico etc. 

(vii)           Lastly, the expression of “preferences” may have a special
meaning. Does this mean in the case of “blog entry” that the person
expressing the preference has no right to grant permission (a third party
blog entry hosted by the person expressing the preference?) or what is the
scenario under which a “preference” rather than a “permission” or
“prohibition” is desirable? What is the legal or practical implication, if
any?

 

Apologies if some of my observations are duplicative or have been discussed
already.

 

I am curious about your answers, however and thank you for pursuing this
initiative!

 

Warm wishes

Carlo Scollo Lavizzari, Avocat
Gysin Avocats, Hirzbodenweg 95, 4052 Bâle, Suisse

 

From: Laurent Le Meur <laurent.lemeur@edrlab.org> 
Sent: 07 April 2026 17:37
To: public-tdmrep@w3.org
Cc: Róisín Trelfa <roisin@spurcoalition.org>; Nikolas Moschakis
<nikolas.moschakis@epceurope.eu>; info@spurcoalition.org
Subject: IETF AI-Pref Wiki

 

Hello, 

 

Here is the link to the IEFT AI-PRef Wiki; this page describes vocabulary
use-cases. 

https://github.com/ietf-wg-aipref/drafts/wiki/Vocabulary-Use-Cases

 

It drafts both an opt-in and opt-out mechanism (for preferences like
train=no, search=yes).

 

The terms "Use" and "Input of a model" are used, rather than RAG or
knowledge retrieval. This sounds simpler; it is precise enough? 

 

Opting out of "AI Overviews" is replaced by a set of Presentation use cases.
AI Overviews may be framed as a mix of content from different sources,
generating an answer. I added Presentation/4 to clarify the use case. Don't
laugh about the recipe example, I'm French after all.   

 

Don't forget that the whole work is presented as an expression of
"preferences", not strict constraints. 

 

Your feedback is welcome 

 

Best regards

Laurent Le Meur