00:00Je pense que la chose la plus importante que je vais dire à mes enfants
00:04c'est que vous allez à l'école pour apprendre comment apprendre.
00:10Il ne faut pas que ce soit pour ce que vous allez à l'école,
00:13ce que vous apprenez, tant que c'est difficile.
00:15Et ça vous enseigne comment apprendre à faire des choses difficiles.
00:20On est ravis au point d'accueillir Aydan Gomez, le créateur de Coher,
00:25une entreprise canadienne d'intelligence artificielle
00:28qui a une vision très différente de beaucoup d'acteurs d'IA
00:32et c'est pour cette raison qu'on lui pose des questions ici dans le studio Dupont.
00:36Aydan, merci beaucoup d'être ici.
00:38Vous co-crééziez Coher.
00:40A quel point les valeurs de Coher sont différentes de l'industrie ?
00:45Pourquoi est-ce important pour vous avoir une AI différente
00:49de celle qui est développée aujourd'hui par les USA ou par les Chinois ?
00:54Et la question aussi sur les démocraties, vous m'avez dit qu'il est important
00:59qu'il y a des autres démocraties aparte de l'U.S.
01:04Oui, je pense que c'est très important que le démocraties en général
01:08continue à lead sur le global stage.
01:11Et donc, nous sommes en train d'une période de large-scale géopolitique change.
01:18Et ça peut être une bonne chose, c'est qu'il peut changer pour le plus.
01:22Nous pouvons voir que les démocraties sont plus fortes,
01:25et les démocraties sont plus fortes, plus fortes, plus fortes,
01:28et plus fortes, plus fortes, plus fortes, plus fortes, plus fortes.
01:32Et donc, je pense que c'est en l'intérêt de tout le monde qui croit,
01:36c'est une démocratie qui croit, c'est une démocratie qui croit,
01:41qui croit, qui croit, qui croit, qui croit, qui croit,
01:44et qui croit, qui croit, qui croit, qui croit, qui croit.
01:46Si la démocratie croit, le projet croit.
01:50Donc, là-bas, je pense que le challenge pour Canada, pour Germany, pour France,
01:56et pour tous qui croit, qui croit, qui croit, qui croit,
01:59qui croit, qui croit, qui croit, qui croit, qui croit,
02:00c'est pour ensure que les capacités se déterminées
02:03et que nous sommes tous capables,
02:07et que nous avons tout ce que nous avons à contribuer à la globalité.
02:10En AI, nous avons le risque de la possibilité de la une-player race,
02:17et donc nous avons vraiment besoin d'assurer que nous sommes un de ces players.
02:22Nous sommes un des peuples qui connaissent commenter cette technologie,
02:25que peut shapeer ça dans le futur,
02:27et peut faire sure que ça protège les valeurs et les interests
02:29de notre peuple et de notre façon de vivre.
02:33C'est true pour France, right?
02:34France should have also more important digital champions,
02:38et maybe very quickly, how could France go into that goal?
02:42Of course, France, you know, one thing that I would say
02:47is that sovereignty is not isolation.
02:51And so it's not the case that what I mean is every single country
02:55needs to build everything, the entire AI supply chain.
02:58We all have to do chips, we all have to do models,
03:00we all have to do applications.
03:01That would be A, impossible, and B, colossally inefficient.
03:07And so that's not what I'm proposing.
03:09What I do think is that there needs to be a diverse supply chain.
03:13You need to have multiple providers,
03:15and countries need to come together,
03:17middle powers need to come together
03:18to cultivate a pool of resources
03:21and develop and invest in these capabilities.
03:26That way, no one holds too much power over you.
03:30There isn't this single point of failure.
03:32Instead, if you have two people supplying you something,
03:35if one goes down or one disappears,
03:38you can move over to the other.
03:39That's what resilience and sovereignty comes from,
03:42is that really there's no one else who has that much power over you.
03:45And so you have optionality,
03:47and you can make your own decisions
03:48without the influence of anyone else.
03:52So that's what needs to be protected.
03:54That's what needs to be pushed forward.
03:55It's not that every country needs to do everything.
03:59Very good.
04:00In these times of acceleration,
04:02there is one question everyone is asking to itself.
04:05What should we learn when we are 15, 20, 50 years old?
04:10And you had a unique childhood, right?
04:14In the forest, with not good access to internet.
04:17Did it help?
04:18And to what extent?
04:19What should we teach our kids?
04:21What should make them read, for example?
04:23I think the most important thing
04:26that I'm going to tell my children
04:28is that you go to school to learn how to learn.
04:33It doesn't matter that much what you go to school for,
04:36what you learn, so long as it's difficult,
04:38and it teaches you how to learn to do difficult things.
04:44So my advice to my brother and my sister-in-law
04:48has always been do the hardest thing you possibly can.
04:52Because regardless of if you're going to do that
04:54for the rest of your life or not,
04:55it will teach you how to do incredibly hard things.
04:58and that skill can be applied anywhere.
05:01And then you can do whatever you want,
05:03whatever the world needs from you.
05:06Which was art for you, for example?
05:08I think I pursued something pretty hard.
05:12AI and computer science is a pretty technical field.
05:16But yeah, I guess mathematics has always been
05:19the thing that's fascinated me.
05:21I've always thought it's been beautiful, creative,
05:25and something that is very, very challenging.
05:29Very good.
05:30Maybe one of two favorite books you have?
05:34I would say
05:36The Emperor's New Mind by Penrose,
05:39which I read very early on,
05:41I think like in my first year of undergrad.
05:44And I understood very little of it, I would say.
05:47But it was something that,
05:50it was like a physicist's exploration
05:54of consciousness and intelligence.
05:57And I found that from like way before
06:01computing was like a mainstream
06:04in your pocket type thing.
06:06I found it so fascinating and eye-opening.
06:10The other one that I liked was Cryptonomicon.
06:12And it was this historical story
06:15set at three different points in history.
06:18One of which was like the Second World War.
06:20The next was like the Vietnam War.
06:22And then more present day.
06:27And I think, you know,
06:28your listeners should read it
06:31because it's such a powerful story
06:33about technology,
06:35its evolution over time,
06:36how it gets applied in different settings.
06:40and I think it's,
06:42it shows different periods of change
06:45in the world,
06:46different periods of technological change
06:48and also cultural and geopolitical change.
06:51Very good.
06:52So I found it super fascinating.
06:53Thank you.
06:54Very quickly,
06:56you mentioned the mathematics you love
06:58and we've seen AI doing
07:00so much progress in mathematics.
07:03Can we imagine in maybe 30 years or 50 years,
07:07the AI being able to do mathematics alone
07:10and leaving the human being
07:13not able to do some progress,
07:15not able to think by itself,
07:17you know,
07:18we're all afraid about the cognitive apocalypse.
07:21I just wanted to hear about you about this.
07:24I think it's already happening.
07:26Like certainly in mathematics,
07:28we're starting to see theorems
07:30that have been open for like 200 years,
07:32100 years get cracked by AI.
07:36That will only accelerate.
07:40I still think it will be humans
07:42who choose what problems to work on.
07:44What is most relevant?
07:46What's most important to work on?
07:48But in terms of the,
07:50the labor or the effort
07:51of actually solving those mathematics problems,
07:55much more will rest on the backs
07:57of large language models.
08:01Ultimately,
08:02I think that's a good thing.
08:04I think it will lead to faster breakthroughs.
08:07We'll be able to solve these problems
08:08faster and more effectively.
08:11But certainly it's bittersweet, right?
08:13Like it,
08:13it means that
08:16less of that work
08:17will be done by humans.
08:20Humans will work on higher level problems
08:22rather than these problems.
08:26but that is what I grew up
08:27loving about mathematics.
08:28The thing that got me so excited about it
08:30was proving theorems.
08:32It's such a fun problem solving.
08:34And can the world models,
08:36you know,
08:37researchers like Ian Lequin
08:38or Fei-Fei Li in Stanford
08:40are working on,
08:41can it help to build
08:42the future of AI?
08:44Because they say that LLMs
08:46are less clever than,
08:48you know,
08:49five years old child,
08:51et cetera.
08:51the fact that
08:52can have a representation
08:53of the world around them.
08:55Do you think that's going to
08:56create a different AI?
08:58World models are fascinating,
08:59but I expect world models
09:01will just be a part of LLMs.
09:02Like I think that process,
09:04that objective
09:05will be solved using LLMs
09:06rather than
09:08being some sort of
09:09entirely independent
09:12track.
09:13So I do think world models
09:14matter a lot,
09:15but I don't think
09:16they're as distinct
09:17from LLMs
09:18as some people think.
09:19But what could be
09:19the next step of AI?
09:24I mean,
09:24for me,
09:25probably continual learning.
09:28So being able
09:29to learn from experience,
09:30grow over time
09:33and improve over time
09:34by interacting
09:35with the world.
09:37Thank you so much.
09:38Cool.
09:39Okay.
09:39Thank you so much.
09:40Thank you.
09:41Thank you.