- il y a 1 jour
AI: To Infinity and Beyond
Catégorie
🤖
TechnologieTranscription
00:00Greetings to all and welcome to the closing session of the day.
00:05And for many of you, the conference, you survived.
00:07Give yourselves a hand.
00:13Appropriately, we're going to talk about AI.
00:16I know it hasn't been discussed much at this conference,
00:18so we're only going to get a chance to tackle this big subject.
00:23And we're going to talk about specifically where it's going.
00:26We have a panel of amazing luminaries.
00:30And actually, I think this session is pretty well populated
00:33because you're eager to hear from these people.
00:36Before we get to the future, I'm going to ask a few questions on the now to our panelists.
00:42And I'll start with Thomas Wolfe of Hugging Face.
00:45You know, I find it fascinating.
00:47I cover a lot of the very big AI powers.
00:53You know, OpenAI and Anthropic and Google and the rest.
00:58But I keep hearing about Hugging Face.
01:00You know, you're staying in there.
01:04Tell me how you see Hugging Face in the AI ecosystem
01:08and where you see, you know, that growing from there.
01:13And why I keep hearing about you.
01:16Yeah, I think it's an interesting question
01:19because we have a different pedigree than many of the large labs.
01:24I mean, first, we're small.
01:26We're not U.S.
01:28Originally, the three founders.
01:29Even though it's a U.S. company, the three founders are French.
01:32And kind of we manage in a way to stay independent among these giants.
01:37Google, Meta, I count OpenAI also in the giants now, obviously, Anthropic.
01:44And sometimes people call us the Switzerland of AI
01:47because we also manage to collaborate with all of them.
01:52We even manage sometimes to bring them together on some projects.
01:55So we did four years ago, we did a project
01:56where actually some people from Google and Meta
01:58work together on training a model,
02:00which is very surprising in the intense way.
02:03I think the way that all of this work is because the goal
02:06and what we are building at HuggingFace
02:08is this open source platform
02:10where everyone can bring their models
02:12and you can find the models of everyone.
02:15Every time you don't really want to go to a specific provider,
02:19but you actually want to own the model,
02:20you want to download and use it on your own data center,
02:22you will go to HuggingFace
02:24and here you will find all the models, basically.
02:27So it can be seen, even though it's a startup
02:30and we have a business model,
02:32and we are actually a cash flow policy.
02:34It can be seen a little bit like a public good, almost, right?
02:37And I think it's something that's often seen,
02:40it's often kind of a trademark of open source,
02:43which is people accept to give something,
02:45they accept to give a model,
02:48and in exchange, they get access to the model of everyone.
02:51So just this year, we passed 1.5 million models
02:56that are publicly available on the platform.
02:58This is the public one,
02:59and you have about twice this size as private.
03:01and when we started this, I remember four years ago,
03:05I thought maybe if we got like 10 models,
03:08it's going to be quite a lot
03:09and it should cover everything about AI.
03:12So, yeah, in a way, we managed to grow this thing
03:14to get both Google to open source their model,
03:17it's meta to open source their model at HuggingFace,
03:19and soon open AI as well to share some open models.
03:23I think everyone just kind of feel like
03:27a lot of positive vibes for this community,
03:30which is kind of strange in this very businessy world,
03:33but I think that's also what brings the big company
03:36to keep sharing their model in a way of community.
03:39Great, well, thank you.
03:40I want to go to Peter.
03:43You know, you had a, you know, you're a professor,
03:46you started a very nice AI company
03:49and then sold it to AMD, right,
03:52which is a chip company.
03:58Yeah, you're not the first one asking that question.
04:06No, so, yeah, I left the professorship
04:10with my research group to set up Silo
04:12as a private AI lab in 2017,
04:15and we bootstrapped through to a team
04:18of around 300 AI scientists,
04:20eventually built out a model-as-a-service offering
04:23and open-source models.
04:26As part of that,
04:28we were one of the first companies
04:30to prove throughput scaling on AMD GPUs,
04:33so we trained our models on AMD GPUs,
04:35and that's obviously one of the reasons
04:37why eventually we joined AMD,
04:40and that's actually what we are doing today.
04:42So, I mean, many here also might believe
04:45that there's only one option.
04:47There are actually alternatives,
04:49and most of the big ones are building models
04:53on AMD GPUs today,
04:55and that's what we're doing.
04:56So, we're basically helping companies
04:58across the world basically train models,
05:01but also run inference on AMD GPUs.
05:05Does AMD want to be NVIDIA?
05:08Sorry?
05:09Does AMD want to be NVIDIA?
05:11You know, well, AMD has done a tremendous journey.
05:16Since Lisa Su joined as a CEO more than 10 years ago,
05:21they first took over the CPU market,
05:23and yes, I mean, AMD is a challenger,
05:27but yes, AMD is targeting the GPU market.
05:31Okay, well, watch that.
05:34Tao, your company, you know,
05:37just only a few years old, right?
05:39Yeah, this company is just like two years old,
05:42but Manus is just three months old, you know?
05:44Yeah, wow.
05:45And, you know, it's already making a splash,
05:48and, you know, this is supposedly
05:50the year of the agent, right?
05:52Yeah.
05:52But you've got agents running now,
05:54and you've made, you know,
05:56a stake in the ground of saying
05:58that we want to do general purpose agents,
06:02and that sort of runs against
06:05what the conventional wisdom is,
06:07that if you're going to get things running now,
06:09you know, it really has got to know a domain.
06:11It's got to be specialized.
06:14Otherwise, it's going to screw up
06:15and cause all kinds of problems there.
06:19General interest sounds really harder.
06:22Why are you doing that?
06:23Yeah, that's a good question.
06:25You know, a lot of, like,
06:26investors also ask this question to us
06:29for the past remuners.
06:31You know, this company started, like, two years ago,
06:33and Menace is our second product.
06:36Our first product is a Chrome browser extension
06:38called Modica.im.
06:40And, you know, we are the most successful
06:44Chrome browser extension for AI
06:47in that category.
06:49But for being a browser extension,
06:52the ceiling is kind of, like, low for us.
06:54So we are considering to build
06:57the next product last year.
06:58The first thing came to our mind
07:00is we should build some AI browser.
07:03Yeah.
07:03So from last March to last October,
07:05we spent, like, seven months
07:07on the AI browser project,
07:08but we never released it.
07:10Yeah.
07:10And during that time,
07:12something very big happens,
07:13which is last July,
07:15Cursor.
07:16You know, Cursor was super viral.
07:17And we learn a lot of,
07:20we get a lot of inspirations from Cursor,
07:22which is not from the coders like us,
07:25but the friends and the colleagues
07:26who are long coders,
07:28they are starting to use Cursor
07:30to solve their daily tasks
07:32like data visualization
07:34or convert a video file
07:36to an audio file.
07:38And when those long coders,
07:40when they are using Cursor,
07:42you will find the most interesting part
07:44is they never care about the left side,
07:47which is the code editor side,
07:49because they are long coders.
07:51They can't evaluate the code.
07:52The only thing they can do
07:54is keep press the accept,
07:56accept, accept the button
07:57in the right.
07:58So that's the moment
07:59we have this idea
08:00that maybe we should build
08:02the right panel of Cursor,
08:04but let it run in the cloud
08:06so all these average users,
08:09all these normal users,
08:10can leverage the power of coding
08:12to solve their daily tasks.
08:14And, you know, right now,
08:16there are so many awesome models
08:18on Hugging Face
08:18and awesome projects on GitHub,
08:20but for most of the users in the world,
08:25they don't know how to code it.
08:27So they can't leverage the power
08:29of GitHub and Hugging Face.
08:30But with Manus, actually, you know,
08:32all the average users
08:34can use those power
08:36to solve different type of tasks.
08:38I think that's why
08:39we choose the general agent way, yeah.
08:42Are you a little concerned
08:44that, you know,
08:45it's such a big focus
08:46of GitHub itself
08:48to create agents for coding?
08:50Oh, yeah.
08:51Yeah.
08:52Of course, yeah.
08:53I talked to Thomas,
08:54that Thomas,
08:55yeah, GitHub Thomas,
08:56yeah,
08:56like maybe just two days ago.
08:59Yeah, so,
09:01and you're based in,
09:02you started in China.
09:04Yeah.
09:04And you're based now in...
09:05Singapore.
09:06Singapore.
09:07Yeah.
09:07We're, you know,
09:08the first Chinese company
09:10to go there.
09:11But do you feel that,
09:13you know,
09:13your origins
09:15are in any way
09:17an impediment
09:17to being brought up?
09:20Or do you feel that
09:21in some other ways
09:22it's an advantage?
09:24Oh.
09:26I think...
09:28We don't just think about
09:30where we came from.
09:33It's always about...
09:34I think...
09:35I've been in Paris
09:37for like three days now.
09:39and I talked to a lot
09:40of like investors
09:42and startup founders here.
09:44And I think for this week,
09:46the topic is always about
09:48how like French startups
09:50can do business,
09:52can do AI in Europe.
09:54But instead of thinking that,
09:56like for us,
09:57we always think
09:58AI just removes
09:59the barriers
10:00of language for us.
10:02So from day one,
10:04this company
10:04just focuses on
10:05the global market
10:07powered by AI.
10:09so we never thought about
10:10where we came from.
10:11We thought about
10:12where we should go to.
10:15Well, I mean,
10:16it's funny
10:16because people,
10:17after DeepSeek,
10:18people feel
10:20there's like a novelty
10:22to companies
10:23of a Chinese origin.
10:25And, you know,
10:25I think you're
10:26entering that conversation there.
10:28Oh, yeah.
10:29Of course.
10:29Yeah.
10:30Yeah.
10:30So in a way,
10:31I think,
10:32I don't know,
10:32have you found
10:33that brings more interest
10:34than people are keener
10:35to learn about
10:35what you're up to.
10:37Yeah.
10:38So I think,
10:39you know,
10:41the talents in China,
10:43actually,
10:44you know,
10:44there are so many
10:46engineers
10:46and working so hard.
10:48So it's very easy
10:49for us to, like,
10:49hire smart people there.
10:51Yeah.
10:52But it also, like,
10:53have some challenges
10:54for us
10:54when we go into
10:56the global market,
10:57you know.
10:57Yeah.
10:58So I want to now
11:00shift to the,
11:01you know,
11:02looking ahead
11:02because you folks
11:03all have such a,
11:05you know,
11:06a grip on the subject
11:07being in there.
11:10I,
11:11in talking to the people
11:12who run
11:13these big AI companies,
11:15they seem pretty much
11:17in agreement
11:18that we're going
11:20to reach AGI
11:21in two years,
11:24in three years,
11:25some of them say
11:26five years,
11:27last time I talked
11:28to Demis Asabas,
11:30he said,
11:30oh, yeah,
11:31well,
11:31a 50% chance
11:32in five years.
11:34So he's, like,
11:34being more cautious.
11:36but I'm curious
11:38what you folks think.
11:40Do you feel
11:41that that's
11:43a realistic expectation?
11:45You've got your hands
11:45dirty on this.
11:48And maybe,
11:49you know,
11:51you could feel
11:52if we're using
11:53the term AGI loosely,
11:54say what that means.
11:56Generally,
11:57when I ask,
11:58it means
11:59AI can do anything
12:01cognitively
12:02that a human
12:02could do
12:03but better,
12:04right?
12:05You know,
12:05and when that moment
12:07comes,
12:08it's AGI.
12:10Thomas,
12:11you've written
12:12about that.
12:12Soon?
12:14So,
12:15I'm one of these
12:16persons who don't think
12:17we'll have one
12:18specific threshold
12:19that will pass
12:20that will be named
12:21and AGI.
12:23I disagree with
12:24Dario on many things
12:25but I like that
12:26he don't like
12:27to use the word
12:27AGI
12:28and he calls that
12:29powerful machine
12:31which gives rooms
12:32to improvement
12:32of performance.
12:33So,
12:34the way I see the future
12:35is we'll keep having
12:36this model that will
12:36just keep getting better,
12:38keep getting better.
12:39We'll be able
12:40to really,
12:40you know,
12:41train them
12:42on more and more
12:43tasks
12:43and they will
12:44kind of keep
12:44getting like this task
12:46more and more
12:46like aiming
12:49to perfection
12:50with always
12:51like a little bit
12:51of failure
12:52and probably
12:52there's still
12:53a little bit
12:53too much failure
12:54for Tau sometimes
12:56but I think
12:57we can all agree
12:58it's already
12:58very impressive
12:59if you use
13:00one of these
13:00reasoning tool models
13:01like it has
13:03this feeling
13:03that you're really
13:05using a machine
13:06that was unbelievable
13:07a couple of years ago.
13:08So,
13:09this will keep
13:09better,
13:10better,
13:10but then I think
13:11there is still
13:11also really a ceiling
13:13to what we can do
13:14with our current
13:15projects,
13:15our current
13:16like way
13:16of training AI
13:17and this ceiling
13:18we were talking
13:19a little bit
13:19about that
13:20recently
13:20is around creativity
13:21and wild creativity
13:23and I think
13:24these tools
13:24are really
13:25more and more
13:26becoming extremely
13:26useful
13:27and very obedient
13:29students
13:30that can basically
13:31do very well
13:31a task
13:32that you've trained
13:32them to do
13:33and we expect
13:34OpenAI
13:34and Close
13:35and the open source
13:36company
13:36to train them
13:37on every task
13:37which will be fine
13:38but sometimes
13:39you also need
13:40to learn
13:40to not just
13:41solve a problem
13:42but to ask
13:43the right problem
13:43and I think
13:44that's where
13:44basically
13:45we won't be able
13:46with the current
13:47tool to ask
13:48some deeply
13:50creative
13:51and complex
13:51question
13:52and to give
13:53an example
13:54I think
13:54a typical
13:54question
13:55is something
13:56that's actually
13:59controversial
14:00for instance
14:00or something
14:01that's counter
14:02to the mainstream
14:03thinking
14:04and these models
14:05are very bad
14:05at that
14:06if you ask
14:06chat GPT
14:07to tell you
14:08something
14:08controversial
14:09or to say
14:10one theory
14:10that's possibly
14:11wrong
14:12it's very bad
14:13at that
14:13but a lot
14:14of discovery
14:15and I have
14:16this blog post
14:17around science
14:18and I think
14:18a lot of
14:18major scientific
14:20discovery
14:20and typically
14:21the one
14:21that make
14:22really like
14:22Nobel Prize
14:23winners
14:23there's not
14:24a lot of
14:24them
14:25right
14:25there's like
14:25one per year
14:26for Nobel Prize
14:27but if you
14:28look at
14:28the scientific
14:29progress
14:29over 20 years
14:30these are
14:30the discoveries
14:31that really
14:32drive
14:32scientific
14:33innovation
14:33and a lot
14:35of the discovery
14:35they consist
14:36in taking
14:37something
14:37that existed
14:38before
14:38and saying
14:39maybe that's
14:40all wrong
14:40or maybe
14:41we missed
14:42something
14:42very important
14:43here
14:43and LLM
14:44will always
14:45be very bad
14:46for that
14:46and it's
14:47not only
14:47for science
14:48so for science
14:48it's big
14:49but if you
14:50think about
14:51starting a company
14:52like we're
14:53all kind of
14:53entrepreneurs here
14:54you need to be
14:55a little bit
14:56contrarian
14:57right
14:57if you do
14:58everything right
14:59for a startup
14:59you're sure
15:00to die
15:0090% of startups
15:02die
15:02and all
15:03entrepreneurs
15:03they try
15:04to do
15:04everything
15:04very right
15:05so you need
15:06to do
15:06something
15:06that's not
15:07only
15:07what's
15:08expected
15:08but you
15:09need to
15:09do something
15:10that's
15:10unexpected
15:11something
15:11that's
15:11actually
15:12like
15:12people are
15:12like
15:13well
15:13why do
15:13you do
15:13that
15:14and often
15:15the best
15:15startup idea
15:16at first
15:16you're like
15:16what's this
15:17idea
15:17and then
15:18you think
15:18a little bit
15:18about that
15:19and you're
15:19like
15:20that's
15:20actually
15:20a great
15:21idea
15:21and so
15:22LLM
15:22will have
15:22these limits
15:23I think
15:23so I don't
15:24know if
15:24it's AGI
15:25or not
15:25but at least
15:26it's something
15:26at some
15:27point
15:27if we want
15:28to have
15:28like
15:28machine
15:29that actually
15:30can create
15:31wonderful things
15:32we'll need
15:33to tackle
15:33this
15:36I think
15:37I'm fully
15:38aligned
15:38with
15:38and the
15:40way I
15:40would put
15:41it
15:41is that
15:41I think
15:43we need
15:44a paradigm
15:45shift
15:47if we
15:47actually want
15:48to do
15:48some of
15:49these things
15:49that Thomas
15:50is highlighting
15:51here
15:52I mean
15:52to me
15:53coming
15:54into
15:56setting up
15:56a startup
15:57from
15:58an academic
15:59position
16:01I think
16:01sort of
16:02it's been
16:03linear progress
16:04from a research
16:05perspective
16:06we're still
16:07doing
16:07in a sense
16:08A to B
16:09mapping
16:10supervised
16:10learning
16:11to a large
16:11extent
16:13and
16:13you know
16:14all that
16:15said
16:15I do
16:18think
16:18then eventually
16:18we end up
16:19in this
16:19definition
16:20questions
16:21on
16:22have we
16:22moved
16:23the goal
16:23post
16:23left
16:24right
16:24forward
16:25backward
16:26I think
16:26we didn't
16:26even
16:26agree
16:27on
16:27whether
16:27we
16:27are
16:27moving
16:28the
16:28goal
16:28post
16:28forward
16:29or
16:29backward
16:29at
16:29the
16:30moment
16:30in
16:31defining
16:31AGI
16:31or
16:32whether
16:32we
16:32are
16:33when
16:33we
16:33are
16:33reaching
16:33AGI
16:34so
16:34I
16:35think
16:35that
16:35discussion
16:35doesn't
16:36really
16:36make
16:37sense
16:37but
16:37the
16:38reality
16:38is
16:38that
16:38I
16:39think
16:39we
16:39have
16:40technology
16:40today
16:40already
16:41that
16:42is
16:42creating
16:42significant
16:42value
16:43and
16:43will
16:43continue
16:44to
16:44create
16:45even
16:45more
16:45value
16:46going
16:46forward
16:47and
16:48in
16:48that
16:48sense
16:49I
16:49don't
16:49think
16:56that
16:56yeah
16:57like
16:58you
16:59know
16:59as
16:59we
16:59just
16:59discussed
17:00off
17:01the
17:01stage
17:01when
17:03we
17:03reach
17:04the
17:04previous
17:05AGI
17:05there
17:06will
17:06always
17:06be
17:06a
17:07like
17:07standard
17:07for
17:08it
17:08so
17:09actually
17:09to
17:09be
17:10an
17:11application
17:11company
17:11like
17:12us
17:12we
17:13only
17:13care
17:13about
17:14the
17:15current
17:15day's
17:16scenarios
17:17it's
17:17like
17:18just
17:19two
17:19hours
17:19ago
17:19I
17:20received
17:20a
17:20message
17:20from
17:21one
17:21of
17:21our
17:21users
17:22and
17:22he
17:22said
17:23that
17:23okay
17:24Manus
17:24just
17:25saved
17:25his
17:25thousand
17:26hours
17:26of
17:27time
17:27because
17:27right
17:27now
17:28he
17:28can
17:28use
17:28Manus
17:29to
17:29create
17:30customized
17:31slides
17:31for
17:32each
17:32of
17:32his
17:33customers
17:33in
17:34just
17:34minutes
17:34you
17:36know
17:36even
17:36maybe
17:37like
17:37two
17:37or
17:37three
17:38years
17:38ago
17:38people
17:39can't
17:39imagine
17:39that
17:40yeah
17:40you
17:40can
17:41just
17:41have
17:41different
17:42slides
17:42for
17:43even
17:43just
17:43a
17:44very
17:44small
17:44customer
17:45for
17:46each
17:46of
17:46them
17:46so
17:47I
17:47think
17:47it's
17:47always
17:48like
17:48to
17:49focus
17:50in
17:51the
17:53focus
17:53on
17:54now
17:54not
17:55always
17:55look
17:56into
17:57the
17:57future
17:57because
17:58when you
17:58look into
17:59the future
17:59there are
18:00always
18:00new standards
18:02for you
18:02to reach out
18:03but with
18:05the powers
18:06at hand
18:06like for us
18:07we are using
18:09and with
18:12the power
18:12at hand
18:12we are
18:15at
18:16a time
18:17that can
18:18create
18:19magic
18:19now
18:20yeah
18:20we don't
18:21have to
18:21wait for
18:21the future
18:21yeah
18:22it's kind
18:23of funny
18:23for years
18:24people always
18:25said
18:25that AI
18:27is always
18:27like what
18:28computers
18:29can't do
18:30now
18:31right
18:31right
18:32yeah
18:32at one
18:32point
18:33the big
18:33deal
18:34was
18:34boy
18:34can AI
18:35ever beat
18:36the chess
18:37champion
18:37right
18:38I remember
18:38I covered
18:39that story
18:40for Newsweek
18:40and you know
18:41we put it
18:42on the cover
18:42it was called
18:43The Brain's Last Stand
18:44that was at the
18:45cover line
18:46and you know
18:47and then of course
18:48after AI
18:49beat Gary Kasparov
18:51you know
18:52it was like
18:52well big deal
18:53that's just
18:54you know
18:54doing scenarios
18:56and each step
18:57along the way
18:58you know
18:58is an AI
18:59but I think
19:00what I see
19:01different
19:02is you know
19:03that a lot
19:04of the people
19:05you guys
19:05are in it
19:06you know
19:07aren't settling
19:09for that anymore
19:10they're saying
19:11that you know
19:12actually the goalposts
19:13are fixed now
19:14and we're going
19:16to get there
19:17but what you're
19:18saying is that
19:19hey
19:19in a way
19:20maybe we're
19:20kind of
19:21there now
19:21yeah
19:25in the last few minutes
19:27I do want to talk
19:28you know
19:29a little bit about
19:29since we're here
19:32about Europe
19:33and
19:34what's happening here
19:36some people
19:37I think
19:38maybe a vibe
19:39of this conference
19:41is
19:42you know
19:43especially among
19:44the people
19:45native to Europe
19:47is that
19:48AI is an opening
19:50for
19:51you know
19:52people here
19:54to
19:55catch up
19:57you know
19:57you know
19:58there's a sense
19:59of
20:00unhappiness
20:01and maybe
20:02a little resentment
20:02about
20:03the US's
20:04position
20:05and
20:06there's a feeling
20:07I wouldn't say
20:09maybe it's a little
20:11short of exuberance
20:12but
20:12you know
20:12good feeling
20:13about
20:14what's happening
20:15here
20:16there
20:19do you
20:20sense this
20:21is that
20:21that moment
20:23for
20:24that to happen
20:25for European tech
20:27finally
20:27to flourish
20:28I tell you
20:29because I'm old enough
20:30to real
20:31thought of
20:32like a few times
20:33where people
20:33would say
20:34watch out for Europe
20:35you know
20:36the wave is coming
20:37right
20:37and you know
20:38then you'd sit back
20:39and we'd say it again
20:41but you know
20:42talking about fixed goalposts
20:43is this time
20:44here
20:45now
20:45Peter
20:46you have thoughts
20:47about that
20:49that's a big question
20:52given that we have
20:53a few minutes left
20:54but
20:54let me say
20:55maybe two things
20:56one is that
20:59I think
21:00you know
21:01what's happened
21:02in say
21:03Silicon Valley
21:03and the ecosystem
21:05that eventually
21:06has created
21:06some sort of
21:07tremendous
21:08technology companies
21:09and growth
21:11I think we need
21:12strong ecosystems
21:14and we have
21:16in Europe
21:16talent
21:17we have
21:19compute
21:20and
21:23you know
21:23one thing
21:24that I do not
21:26think we have
21:27is a strong
21:28single market
21:29where to scale
21:30technology companies
21:31and
21:32that's a bigger
21:33question
21:34how we actually
21:35transform that
21:36but what we can do
21:38is to focus
21:38and strengthen
21:39ecosystems
21:40and I think
21:41Paris is a good
21:42example of that
21:43given there is a
21:44strong academic
21:45cluster
21:45there are some
21:47significant R&D
21:48initiatives
21:49ranging from
21:50open AI
21:50to meta
21:51and hugging face
21:53and others
21:53and then we have
21:55seen AI companies
21:56being born here
21:57during the past few
21:58years
21:59now
22:00where I think
22:02the opportunity
22:03lies
22:03given that
22:04Europe has not
22:05really given
22:06birth to
22:07these tech
22:08platforms
22:10and that's also
22:11what we are
22:12doing at the
22:12moment
22:12where we're
22:13seeing large
22:14scale AI
22:14workloads
22:15happening
22:15life sciences
22:16so biopharma
22:18we're seeing
22:19that happen
22:20in automotive
22:20and robotics
22:21we're seeing
22:22that happen
22:23in say
22:23visual media
22:24entertainment
22:24gaming
22:25Europe has
22:26a strong
22:27industry
22:27in automotive
22:29in life sciences
22:31in gaming
22:32and I think
22:33now there is
22:34a possibility
22:35to take
22:36a significant
22:36position
22:38there's also
22:39a risk
22:39that we are
22:40losing
22:41that opportunity
22:42unless we're
22:43doing the big
22:43bets
22:44that needs
22:44to happen
22:45Thomas
22:46your thoughts
22:47yes
22:48I'm very
22:49optimistic
22:49I think
22:50also
22:50I mean
22:51we are
22:51I think
22:52we are
22:52good
22:52demonstration
22:53right
22:53US
22:55European
22:55company
22:55you face
22:56Finnish
22:57one
22:57Chinese
22:58like Singapore
22:59now
22:59you can
23:01create
23:01amazing
23:02AI company
23:03from everywhere
23:03I think
23:04Tao
23:04what you
23:05said
23:05in a way
23:06is that
23:06AI is
23:07kind of
23:07really
23:07reducing
23:08the entry
23:08barrier
23:09both for
23:10your user
23:10to access
23:11your company
23:12but also
23:12for us
23:13to create
23:14company
23:14once you
23:15have all
23:15this LLM
23:16all this
23:16basic
23:17intelligence
23:17machine
23:18available
23:18just your
23:19job is
23:20just to
23:20create
23:20an amazing
23:21user interface
23:22or amazing
23:23application
23:23and that
23:24I think
23:25can be
23:25created
23:25everywhere
23:26so
23:26what we've
23:27seen
23:27in the
23:27past
23:28year and
23:28a half
23:28is
23:29a lot
23:29of very
23:29interesting
23:30startup
23:30emerging
23:31in Europe
23:31emerging
23:32in Asia
23:32emerging
23:33in the
23:33US
23:33so that's
23:34a great
23:35chance
23:35I think
23:35and Tao
23:36I have
23:37to ask
23:37you about
23:38what we
23:39should expect
23:39from China
23:41you know
23:42we have
23:42the deep
23:44sea scare
23:45we have
23:46I guess
23:47you could
23:48talk about
23:49the thriving
23:49startups
23:50but
23:51how do you
23:52see China
23:53fitting into
23:54the global
23:54market
23:55in terms
23:55of AI
23:56yeah
23:57I think
23:58just as
23:58I mentioned
23:59before
24:00I think
24:00it's
24:01always
24:01about
24:02to
24:03to face
24:04the global
24:04market
24:05first
24:05I think
24:06if you
24:06know
24:06where I
24:08came from
24:08in China
24:09it's like
24:09if we only
24:10focus on
24:11the Chinese
24:12market
24:12we only
24:12do business
24:13there
24:13I won't
24:14be sitting
24:15in this
24:15chair
24:15today
24:16you know
24:16I think
24:17why I sit
24:18in this
24:18chair
24:18today
24:18is because
24:19from day
24:19one
24:20we are
24:21focusing
24:22on the
24:22global
24:22market
24:23and that's
24:24what
24:24AOM
24:24just
24:25empowers
24:26us
24:26to do
24:27this
24:27business
24:27I can't
24:28imagine
24:28if I
24:29can do
24:29a global
24:30product
24:30like
24:31Manus
24:31five years
24:32ago
24:32without
24:33the
24:33help
24:34of
24:34AOM
24:34so
24:35I think
24:36because
24:36we are
24:36talking
24:37about
24:37the
24:38AI
24:39Europe
24:39right
24:40so
24:41the
24:41only
24:42advice
24:43I can
24:43give
24:44is that
24:44don't
24:45just
24:45focus
24:46on
24:47Europe
24:47yeah
24:48you have
24:48to
24:49focus
24:49on
24:49the
24:49global
24:49market
24:51and
24:51then
24:51you
24:51get
24:52Europe
24:53yeah
24:53I think
24:54that's
24:54the key
24:55well
24:56we
24:56could
24:56go
24:57on
24:57but
24:58they're
24:59kicking
24:59you
24:59out
25:00so
25:01thank
25:01you
25:02very
25:02much
25:02I
25:03want
25:03to
25:03thank
25:03my
25:03panelists
25:04they
25:04were
25:04wonderful
25:04thank
25:08you
Commentaires