Re: Improve support of LLMs?

Dear Colleagues, 

At what time is the meeting today and what is the link, please ? Thanks in advance. 
In the framework of the [ https://www.e-vita.coach/ | https://www.e-vita.coach/ ] project, we have organized a workshop 
on KGs&LLMs : [ https://www.e-vita.coach/wp-content/uploads/2024/02/2024-03-Evry-LLMKG_v4.pdf | https://www.e-vita.coach/wp-content/uploads/2024/02/2024-03-Evry-LLMKG_v4.pdf ] 
Talk to you later,... 

Kind regards, Gérard 
​​_____________________________________________________________________ 
Gérard CHOLLET,PhD,DR,CNRS-SAMOVAR,gerard.chollet@TELECOM-SudParis.eu 
Institut Polytechnique de Paris, IMT-TSP, 9 rue Charles Fourier, 91011 Evry 
Tel: +33 1 75319628 , http://perso.telecom-paristech.fr/~chollet/ 
_____________________________________________________________________ 



De: "Dirk Schnelle-Walka" <dirk@switch-consulting.de> 
À: "Noreen Whysel" <nwhysel@gmail.com>, "Gerard CHOLLET" <gerard.chollet@telecom-sudparis.eu> 
Cc: "public-voiceinteraction" <public-voiceinteraction@w3.org>, "leonardo" <leonardo@chiariglione.org>, 
"SANSEN Hugues" <hugues.sansen@telecom-sudparis.eu>, "JokinenKristiina" <kristiina.jokinen@aist.go.jp>, 
"graham.wilcock" <graham.wilcock@helsinki.fi>, "Alya Yacoubi" <ayacoubi@zaion.ai>, "Franz FODÉRÉ" <ff@zaion.ai>, 
"Nigel Cannings" <nigel.cannings@intelligentvoice.com>, "Mark Seligman" <Mark@speechmorphing.com>, 
"ffyassa" <ffyassa@gmail.com>, "Jerome BOUDY" <jerome.boudy@telecom-sudparis.eu> 
Envoyé: Mercredi 3 Avril 2024 09:52:14 
Objet: Re: Improve support of LLMs? 



Thank you so much Gerard. 

Knowledge graphs are already part of what we described so far. 

See [ https://w3c.github.io/voiceinteraction/voice%20interaction%20drafts/paArchitecture/paArchitecture-1-3.htm#finalarchitecture | https://w3c.github.io/voiceinteraction/voice%20interaction%20drafts/paArchitecture/paArchitecture-1-3.htm#finalarchitecture ] 



As the current description is still vague, we may want to have a closer look at it, today? 

Gerard, I think you had another meeting in parallel. Would it be possible for you to join today, to report about some of your efforts? 

Dirk 
On 30/03/2024 21:20, Noreen Whysel wrote: 


This is amazing. Thanks, Gerard! 
Noreen 


BQ_BEGIN
On Mar 30, 2024, at 1:04 PM, Gerard CHOLLET [ mailto:gerard.chollet@telecom-sudparis.eu | <gerard.chollet@telecom-sudparis.eu> ] wrote: 





BQ_BEGIN

Dear Colleagues, 

We have been keeping track of publications in this domain. Please, find attached a list of links to references 
which could possibly be useful,... 

All the best, Gérard 
​​_____________________________________________________________________ 
Gérard CHOLLET,PhD,DR,CNRS-SAMOVAR, [ mailto:gerard.chollet@TELECOM-SudParis.eu | gerard.chollet@TELECOM-SudParis.eu ] 
Institut Polytechnique de Paris, IMT-TSP, 9 rue Charles Fourier, 91011 Evry 
Tel: +33 1 75319628 , [ http://perso.telecom-paristech.fr/~chollet/ | http://perso.telecom-paristech.fr/~chollet/ ] 
_____________________________________________________________________ 



De: "Noreen Whysel" [ mailto:nwhysel@gmail.com | <nwhysel@gmail.com> ] 
À: "Dirk Schnelle-Walka" [ mailto:dirk@switch-consulting.de | <dirk@switch-consulting.de> ] 
Cc: "public-voiceinteraction" [ mailto:public-voiceinteraction@w3.org | <public-voiceinteraction@w3.org> ] 
Envoyé: Samedi 30 Mars 2024 14:29:41 
Objet: Re: Improve support of LLMs? 

Perhaps a way to integrate knowledge graphs specific to a domain. Integration is key since each application would have its own knowledge graph 
and the standard can’t (and and shouldn’t have to) anticipate all concept models. I don’t personally know how to do that, but I’ve been reading 
a lot about how concept model integration is the next phase from Gen-AI to something that would actually be useful with a knowledge domain. 
And I do know some people who work in the knowledge graph space. 

Noreen 


BQ_BEGIN
On Mar 29, 2024, at 5:08 AM, Dirk Schnelle-Walka [ mailto:dirk@switch-consulting.de | <dirk@switch-consulting.de> ] wrote: 


BQ_END


BQ_BEGIN



Dear all, 

I have the impression that the currently followed architecture of is still too closely bound to traditional NLP based voice interaction concepts. While it aimed at including LLM with speech, LLM with multimodality, ... it is potentially not visible enough. We may want to also improve here. 

Any thoughts on this? 

Dirk 

BQ_END


<KGs and LLMs_v35.docx> 

BQ_END

BQ_END