Open Source for AI and Emerging Technologies - UN Open Source Week 2026 (Part 1)

Good morning, everyone.
Please take your seats.
We are ready to begin.
And please do come forward.
There are some seats over here.
We know that not every seat has a set of headphones, so do come closer if you can find an empty seat.
Once again, good morning, everyone, and welcome to the open source for AI and Emerging Technologies Day.
My name is Beth Vananan.
I work for the Office for Digital and Emerging Technologies, and I'm delighted to be the host for you today, so you're stuck with me all day, unfortunately.
This day is really exciting because a lot of you have been coming to the open source activities in the past years, but this is the first time we're doing a fully dedicated day on AI.
It's very exciting to have such an extensive program today.
So if you were here yesterday, that was the takeover day.
And two more, if you're coming back, please do come back.
It will be the DPI Day.
And then on Thursday, it's going to be Opose and on Friday, community led events.
So a very exciting week, so we hope that you will be attending each day.
This week has been months in the making.
I know you come here for maybe five days, but as many of you have been organizing events, they take a lot of work.
So please let's give a round of applause for Medsnene, Moritz, and Omar.
And plenty of other people who've been working on this for months.
Please, let's have the agenda.
So we're going to start off with a keynote and a fireside chat, and then we will be doing a few setting the scene speeches.
After that, there's going to be a high level session on open source AR for development, and then a session on open robots.
Especially you sitting on the right side of the room, sorry, left side of the room, you may have seen there's a robot there, so we will be talking about those later.
Then we will have a two hour lunch break, followed by breakout sessions, and there will be a lot more information on those breakout sessions before lunch.
And then we will end the day with a session on open agents, and then there will be a closing panel.
Before we begin today's program, next slide, please.
There will be a QR code on the slide here, so please do scan that.
This is really an exciting UN initiative, which is the Digital cooperation Portal.
Many of you obviously work in the digital space, so this is your opportunity, if you would like to, to add your projects in this portal to share the work that you're doing in a global platform.
This will be contributing to the follow up of the global digital compact, a very exciting opportunity to showcase your work.
Then to some very boring housekeeping, unfortunately, the microphones are set up in a way that you can't necessarily hear everything if you're listening without the headphone.
If you have a headphone or one of those earpieces on your desk, please feel free to use that.
That is going to help hearing.
Unfortunately, we don't have any at the back, so that's also why it's better to try to find a seat in the front.
And please note that food and drinks are not permitted in this room.
That's why we have a two hour lunch break so you can go and get something nice from outside.
There's no food served, but you can go to the fourth floor cafeteria.
And if you want to quick coffee, there's coffee in the basement.
So just a few floors down, there's another cafe there.
And we have some amazing volunteers, one over here wearing yellow shirts.
They'll be showing you around this building, and I'm sure you all know it's a bit of a maze, so please do ask for help if you're trying to find the cafeteria or anything.
All right.
An important note from security lesson learned from yesterday, please do not leave your belongings unattended.
So if you're leaving this room, take them with you so that they don't take them for you.
So please do carry everything that you have with you at all times.
Last but not least, there's feedback forms.
If you're exiting the room, you'll see there's some QR codes by the doors, so please do scan those and leave us your feedback.
It's so important so that we can continue to make this event engaging next year as well.
With that, I will now hand it over to the Under Secretary-General, Amy Gil, for the opening remarks and fireside chat session.
Thank you so much, Pepe, and welcome to the Open Source Week.
And please remember her suggestion.
These earpieces are a special artifact.
You will see them only in UN buildings, so use them.
So let me welcome you to the second day of the open source week, the first formal day, in a sense.
Yesterday was very exciting, the tech over with hackathon, a maintthon and an ditathon and we are building quite a tradition there, and I want to give a shout out to all those who participated and all those who mentored.
They have taken out time from their jobs to fly over and help out with this.
So thank you very much.
So this opening is special.
It's the first time ever that we are dedicating a full day to open source AI.
And there are several reasons, and I'll get to that.
I want to, of course, start by thanking our co hosts who come from all regions represented at the United Nations, the Dominican Republic, Estonia, Jamaica, Kazakhstan, Lesotho, Tanzania, and Luxemburg.
And thank you also to all the partners who made this week possible.
This is a special week.
There are very few rooms, as the Wag said last year in the UN where those in suits sit next to those in shorts and make things happen, and we deliberately take a back seat in terms of our policy agenda.
So this week is not about, um, the policy aspects of technology as much as it is about those who are actually building the technology.
It's a conversation with those builders, those uh designers.
Now, I mentioned this is the first time we are having a full day dedicated to open source AI, and this subject is not anymore on the margins of the conversation.
Now it's a centerpiece of those discussing AI.
Our expectation is that this will be an agenda shaping day.
We want to be able to leave New York this week with some guidance on how countries should decide which paths to take What choices to make about openness that will shape who gets to participate in shaping our AI future.
Openness matters, firstly because it lowers the barriers to entry.
Letting a researcher, a developer, a ministry, or a startup begin from an existing foundation rather than start everything from scratch.
It also matters because it counters concentration.
It pushes back against the buildup of points of failure, and it counters stagnation as well by bringing in fresh talent, fresh ideas into the innovation space.
But openness alone is not enough.
Excess is not the same as capability, and for open source to deliver on its promise, especially for development, it has to be paired with investment in skills, in compute and local innovation ecosystems.
Capacity is actually the difference between consuming the technology and actually building it.
So it's a bridge that connects the consumers to the architects, takes them across the line.
At the United Nations, this is a very special year for us.
Pepe mentioned the Global Digital Compact 2024, the decisions of that moment have now been translated into practical mechanism functioning mechanisms.
The International Independent Scientific Panel on AI.
Last week, sha Benju and Maria Ressa were here to talk about their inaugural report.
Uh, the first global dialogue on AI governance, which gives every country a seat at the table in shaping the rules for artificial intelligence, that'll take place next month, early next month, six to seven in Geneva and the Secretary-General would be hosting heads of state, heads of government, leaders from companies, civil society, At that historic dialogue.
It'll be the first UN platform where everyone has a seat at the table to look at the opportunities, the risks, the direction that technology takes and start the process of learning from each other.
It's not a top down process, it's a horizontal facilitation of international learning on the governance of Air.
There's a third crucial aspect coming out of the Global Digital compact, which doesn't yet have standing, standing mechanisms behind it, and that's capacity building.
There are ideas on the table, including for a global fund on AI to help the 90 plus countries that are most at risk of getting left behind.
There are some initiatives by independent centers, by member states are coming together in a network of centers for exchange and cooperation on AI capacity building, already more than two dozen, very close to what the open source community is all about.
But this is not enough.
The needs are humongous, and we need to do more on this third pillar.
Cutting through all these threads, science, policy, and capacity is open source as a foundational element, and we are going to be looking this week at everything that can make this connection the most powerful shaper of our AI future.
Ladies and gentlemen, now it's my profound honor to introduce a true title of our field, Turing Award laureate, one of the legendary godfathers of AI, whose convolutional neural networks literally taught machines how to see.
And he was former chief AI scientist at Meta, and now he's the founder and executive chairman of AMI labs and where he's boldly betting on world models to take us beyond the current paradigm and the limits of today's AI, a pioneer, a visionary, and a contrarian.
So please join me in welcoming the one and only Yan Kun.
Thank you very much for the introduction.
Thank you for attending this session.
I think this is a very timely event, and I'm truly honored to be here and to tell you about open source.
So first of all, I came to this building 18 months ago and, argued for AI open source at the UN Security Council.
The message I'm going to deliver today is very similar to the one I delivered 18 months ago, which didn't go very far back then.
I'm hoping since then many things have happened which I think motivate more countries around the world to have AI sovereignty, but also to subscribe and support an open source approach to it.
AI is becoming quickly a platform to some extent, actually has already become a, a platform that a lot of us relies on.
When I'm talking about AI, I'm not necessarily talking about specialized application of AI, but the AI that is built around large language models.
What is happening is that increasingly, AI is mediating all of our interaction with the digital world, with information more generally.
Now, if all of our information is being mediated by AI systems, and by the way, increasingly, this is going to become more prevalent because in addition to smartphones, we're going to be running around with smart devices like the smart glasses I'm wearing at the moment.
I can take a picture of you.
I hope you smiled.
I increasingly, we're going to just ask questions to our AI assistant, which will be with us at all times.
It will be personalized AI assistants that hear what we hear.
Eventually, they will see what we see and they'll become our best digital friend.
They will really become a staff member, if you want.
All of us will be acting like a manager being constantly followed by a staff of helpers, but those helpers will be digital.
Now, it's already happening to a large extent that whenever we have a question, we ask an AI assistant.
We don't go to a primary source.
We certainly don't go to libraries anymore, sadly, and It's a very drastic change behavior which may lead to drastic changes in society.
If the information or information diet is entirely mediated by AI systems, and those AI systems are proprietary systems produced by a handful of companies, otherwise goes to the US and China.
It's very dangerous for culture, linguistic diversity, Diversity in centers of interest in value systems, in political opinions and biases, and for democracy, for human rights.
We cannot afford that all of the information is funneled through systems that are absolutely necessarily biased.
There is no such thing as an unbiased AI system.
So what should we do? Um If we want sovereignty, if we want to preserve cultural diversity, linguistic diversity, if we want people to have access to a wide diversity of AI systems, in my mind, the only way to get to that point is open source AI platforms or open AI platforms, I should say opens AI platform.
I Because most countries around the world cannot necessarily afford or maybe don't have the resources or the talents to actually build their own LLM.
A number of countries can do this.
The LLMs they produce are good, they're not great.
They're not at the top, but they're good.
The talents are there.
Some countries are significant resources in compute, but there is a way that an open source effort that would be collaborative around the world could actually surpass in performance the proprietary systems.
Because in the end, people will just use the best system that's around.
Here is how it would work.
Each country, each region, each academic institution, whatever it is, would digitize their own cultural material.
And will contribute to training a global AI system that would constitute a repository of all human knowledge.
But they would not have to communicate the data.
They could contribute to training a global model by exchanging parameter vectors, which are how AI systems can distill the data down to knowledge.
I was trying to popularize this idea.
I've been trying to popularize this idea for almost three years now, and I talked about it at the UN Security Council, trying to convince also the leadership at META to play an important role in there, but that didn't quite fly.
After I left META, I started with some colleagues at the AI Alliance, which is a nonprofit that is promoting open source AI.
We decided to start a project called Project Tapestry.
You can Google project tapestry and it's a confederation of partners that can contribute to training a global AI model, while preserving sovereignty on the data and only exchanging parameter vectors as often as possible so that collaboratively, each region in the world, each academic group, each private company possibly contributes to training a big model while retaining sovereignty on the data using their own computing infrastructure if they have some.
At the end, we'll get a system that speaks all the languages in the world, understand all the value systems, at least at the basic level, and all the cultural biases, political philosophies, et cetera, and centers of interest.
Now, once we have such an open platform, Anybody can take this open platform and fine tune it for their own purpose, whether it's a commercial enterprise or a government or an academic group or a non profit to serve a particular population.
That way, people will have access to a wide diversity of AI assistant.
We need such a high diversity of AI assistant for the same reason we need a high diversity of the press.
Now there are issues with this.
One issue is to get everybody to work together.
The tapestry project is very much bottom up.
It's people who have expertise in training L&M and other AI models who decide to collaborate.
There's a GitHub repository, you can just sign up.
There's no heavy infrastructure or authorizations to get.
Um the means and the resources are bonaide by the participants themselves, so that can be completely self organized.
But of course, there needs to be political support for it.
If governments tell their academics, their companies and give them an incentive to participate in this project, of course, it will go faster.
Present, Project Tapestry has participant.
There was an inaugurating workshop that took place early May in Paris.
We had participants from several countries in the EU, Switzerland, UK, United Arab Emirates, India, Kazakhstan, Vietnam, Japan, Korea, various groups from academia and industry groups like IBM, NVIDIA, A&D, Intel, all the hardware suppliers, basically the main ones.
There is a groundswell of interests for this project and we see new partners signing up every day, and So I see the history of AI platforms as following the history of the software and hardware platforms of the Internet.
In the late 1990s, if you wanted to start an Internet service of some kind, a website, you will have to buy proprietary hardware from Sun Microsystems, Dell, Hewettt Packard, and other companies, and then use their proprietary operating system software on top of it.
All of these were completely wiped out in the early 2000 when people started using community hardware with open source software stack.
The same thing is going to happen.
A similar thing happened also for the software stack of the mobile communication network.
You cell phone very likely actually runs an open source operating system unless you have an iPhone.
It talks to a cell phone tower that runs an open source software stack based on Linux.
Um, so there is a push by the market.
It's not just government decision.
The market wants open source platforms because it's cheaper, it's more secure.
It's easier to port, uh, to run locally, if you need to, to preserve privacy and everything.
There's tons and tons of advantages, which is why we're having this meeting today.
I mean, this week, I should say.
So, you know, I think that is the direction of history.
It's inevitable.
Government should embrace it and accelerate its, uh, uh, It's progress.
Now there is another discourse around AI, which is essentially opposed to this and it's a discourse that essentially claims that AI technology is intrinsically dangerous and should be regulated.
Its access should be regulated because bad people will do bad things to it, either with cybersecurity or getting a recipe for bioapon or something like this.
I think those dangers are very widely overstated.
I don't think those dangers are nearly as bad as some people have claimed, some people in the industry and academia have claimed.
I think the alternative where if you believe AI is intrinsically dangerous and you should regulate its access and therefore open source AI should be banned, I think is extremely dangerous for democracy and human culture in general, as I pointed out earlier.
But those arguments sometimes are justified by security arguments, which I think are contrary.
This is a big debate, and Well, I disagree with some of my friends on this issue.
But I think to some extent limiting access to AI technology because of security reasons is akin to, in the 15th century limiting the use of the printing press, because, of course, we can control what information will be disseminated through printing.
I think this is akin to medieval obscursm.
I think it's very, very dangerous to limit access to this technology which essentially provides access to all of human knowledge to a wide population.
I'm really happy that this event this week is taking place.
My last point is that there is a feeling in many countries that are neither the US nor China that they've lost the race towards AI and that it's now in the hands of American industry and Chinese industry and there is no way to catch up given the size of the investments that unnecessary.
I don't think that's the case.
There is a race which perhaps many countries have lost or are losing to build LLM based AI systems.
Those systems require enormous amounts of computing power.
To run them, they require enormous amounts of memory.
The reason they require a lot of memory and compute power is because they need to be very large, because they basically are accumulating the clarity knowledge.
Those systems are not very smart.
They're very good at storing declarative knowledge and regurgitating at the right time two questions.
They're only smart in two areas which are very special coding and mathematics.
There they can invent new things a little bit.
But those are very specific domains where the substrate reasoning is actually linked to the language, first of all, and second of all, if a system figure out a new program or a new theorems, it can be automatically verified whether it's correct or not.
Those are very specific domains.
The self improvement of AI system does not apply to any other domain, essentially, or in very limited ways.
So what's going to happen is that we're going to have another revolution in AI.
My new company is based on this hypothesis that we're going to have a new revolution, particularly AI systems that can deal with the real world.
LLMs are really good at dealing with language and sequences of discrete symbols, not so good at handling the real world, which is why we have systems that can pass the bar exam, solve mathematical problems and write code, but we don't have domestic robots.
AI is still completely insufficient to deal with the real world.
That's going to be the next revolution.
It's up for theta.
And we're not going to have human level intelligence over the next two or three years.
This is going to take much longer.
But there are opportunities for international collaborations, both for LLM based AI systems and for this next generation of AI systems.
Thank you very much.
Thank you very much for that opening keynote and I hope we will all get a copy of that picture that you took of the room.
Please, and Kan and USG, you can take seats for the fireside chat.
Thank you.
Thank you.
Thank you, Jan, for setting the stage for this week and you mentioned your most recent endeavor and the work on real world models.
So we are in this paradigm and we want to get to that paradigm.
Can you give us an idea of the roadmap, what is required in terms of capabilities, research to get us to that point? Sure.
I So current technology, LLM, as I said, deals very well with language and other type of data modalities as we call them, that where the information can be formulated in terms of a sequence of discrete symbols.
Human language, computer code, mathematics, these are sequences of discrete symbols.
You can also include in this DNA proteins and other things.
Are things that are very difficult to reduce to a sequence of discrete symbols.
For example, the state of this room, we don't know how to reduce this to a string usefully, at least, to a stream of discrete symbols.
Which is why current AI systems work really well with language, but not so well dealing with the real world.
Are things that we can bolt on top of an LLM that allow them to interpret the content of an image.
Not so much video yet, but it's very primitive in many ways.
How do we get AI systems to deal with the real world? The good news is that we have some ideas and systems that are starting to work.
The bad news is that we cannot use the type of generative architectures that has been so successful in the context of language.
I'll tell you Very quickly, I don't want to be too technical, but I'll tell you why.
If you give an AI system a computer, a sequence of symbols, let's say a bunch of words, a piece of text, and you train it to predict the next word that comes after.
You cannot exactly predict which word will follow a sequence of words, but you can train a machine learning system to do is predict basically what amounts to a probability distribution over all possible words in your dictionary.
Every word in a dictionary now comes with a number that indicates with what probability every word will follow that sequence of words you just observed.
Now, what you can do is pick one word from this distribution, insert it into the input, and now it becomes part of the input and you can predict the second word.
Shift that into the input, predict the third word.
This is called auto regressive prediction, and that's what LLbs are based on.
Works really well for discrete symbols.
If you try to apply this to video, basically show a chunk of video, a few frames to a system or a few seconds and ask it to predict what's going to happen next.
This model does not work at all.
It works, but not well.
The reason is you can't train a system to predict every detail in a video and you can't even train it to produce a probability distribution over all possible futures of a video because there is an infinite number of them and it's a mathematically intractable problem.
If I take a video of this room and I start from this side and I turn the camera slowly and I stop here and ask the system predict what's going to happen next in the video, system will probably predict that the camera is going to keep turning.
There's no way it can predict what all of you look like.
It's just the information is just not there.
It's impossible.
When you train a system to make this prediction, you basically kill it.
It really doesn't work.
You have to use what you can tell at an abstract level is that there are people sitting in seats and I can't tell what they look like, but there's some diversity of origins and gender and everything.
What you can do is build a system that finds this abstract representation description of the world.
It's not in terms of language.
So that you can make those predictions.
This architecture is called pp, joint Embedding predictive architecture, and that would be a replacement for the GPT architecture, the GPT of Chad GPT, which is generative.
Those are non generative architecture.
I apologize if I was too technical.
But basically, it's a new breed of AI system completely.
Now, based on those architectures, you can also build world models.
W models are quickly becoming a buzzword in the AI research, not so much in industry yet, but they will come.
What is a world model? It's a system that would allow an AI system to plan a sequence of actions to accomplish a task.
Given the state of the world that you currently observe and given an action that you imagine taking, the world model would predict the state of the world that would result from taking this action.
If an AI system has such a world model, it can imagine what the effect of a sequence of action would be on the world.
If it's asked to accomplish a particular task, it can search for a sequence of actions that will accomplish this task.
I call this objective driven AI.
This is the stuff that we're building at my new company.
I think it's a new generation of AI systems and we'll see in a few months, a few years whether this works out.
Most of the applications are industry.
Can use this technique to build phenomenological models of complex behavior of a complex system, like say a power plant, a chemical plant, steel mill, jet engine, or a human cell, or a patient.
If you had a complete pg model of a patient, you could plan a sequence of treatments that would bring the patient to a good outcome, for example.
I think there's tons of applications of this in industry which we hope to develop over the next year and this is a revolution that has not happened yet.
Amazing.
We spoke a bit about the direction of potential travel for the tech.
I think another issue you touched upon was the risks.
You've been very clear about existential risk that these are often made up and there are false analogies being brought into the conversation.
That doesn't mean that you're careless, but you focuses more on the short term and the medium term risks.
That's something that concerns us at the United Nations as well.
One of the risks you highlighted was this concentration, of tech power and the need to have a more open approach to building it out.
What would you counsel countries that they do not lose sight of the risks, but at the same time, they don't constrain this open approach to innovation.
What's the sweet spot? A number of different things that need to be considered.
At the moment, AI systems in their current incarnation in the form of LLMs, are not particularly dangerous because they're not particularly smart.
They rarely almost never give you information that is not available from a textbook or a library.
They just accelerate the access to it and widen the access, which is why people are imagining scenarios where badly intentioned people will have access to recipes for bio weapons.
Now, one thing that needs to be considered there is that having a recipe for bioapon is relatively easy.
Building a bio weapon is incredibly complicated, particularly if you don't want it to kill yourself.
In my opinion, this danger is hugely overestimated.
This cybersecurity, people say, Oh, look, my new system is so powerful, we're not going to release it because it can break into lots of secure systems.
Now, if you have a system like this that can detect weaknesses, you can use it to solidify your own cybersecurity system.
There it's a matter of other are the good guys faster, smarter, better, more numerous than the bad guys? The answer is almost certainly yes.
I mean, it's not almost certainly, it's certainly yes.
Having those systems have their own antidotes, if you want, in them.
We shouldn't be scared of this.
This is a first order effect.
Oh, my God, that can penetrate security systems.
Two questions.
First of all, penetrating the system doesn't mean you're going to bring down the electrical grid or anything.
You know, you can't do this at that scale.
That's a good James Bond scenario, but it doesn't happen in reality.
Then second, can I use this to protect myself? The answer is yes.
Just do that.
I mean, every new technology opens the door to, you know, new nefarious effects by better intentioned people.
But generally, we have countermeasures that appear pretty quickly and there's nothing absolutely disastrous that we can expect from those systems.
Don't overestimate the dangers.
Now, of course, a lot of people in counterintelligence are in the business of being professionally paranoid.
Clearly, you have to think about those things, but don't overestimate and kill your sovereignty because of this.
That's one element.
I Second element is, as I was saying in my keynote, That is contrary to a picture of open access to AI systems.
Those two things should be weighed against each other.
That's where I disagree with my dear friend Yersh Benju whom you mentioned earlier.
He's much more worried about bad users, in particular, bad uses by even the companies producing those AI systems because he thinks if it's motivated by profit, bad things will happen, I don't know if it's right about this, I don't think so, but again, the solution to this is open source platforms.
Then there is the sci fi issue of is AI going to take over the world? Are we going to lose control, et cetera One point about LLMs, if you believe that we're going to scale up LLMs towards human level AI, Then you might be entitled to be worried because LLMs are to some extent intrinsically unsafe.
They're not really controllable.
Companies fine tune them so that they don't give you a recipe of a bio weapon, but you can always jaelbr them.
You can always give them a prompt that was not imagined by the designers that will get it to answer a question it should not answer.
Again, as I said, they're not going to invent anything, so it's not going to be information you can't get in other sources.
Um it's a consideration.
Those systems are intrinsically unsafe, but they're not that smart and so they're not that dangerous.
Now, the next generation AI systems May reach at some point, human level intelligence.
It's not going to happen tomorrow.
We have a number of years, perhaps decades to think about it.
But those systems, if they are based on the blueprint that I described, objective driven AI, they're much more controllable because you give them an objective and they can do nothing but fulfill this objective, take actions that fulfill this objective, and you can make them subject to safety guardrails that cannot be broken because it's not fine tuning, it's actually putting an objective function directly in the inference procedure that prevent those systems from producing unsafe behavior.
Assuming their world model is accurate.
I think there is a design for safe systems, which I'm working on with several colleagues and there I think we have a way to make the system secure.
Great.
Let's move now to two quick points about something that matters to us a lot at the UN apart from the risk versus opportunities discussion.
One is the wider access to the enablers of AI.
There is this vectorized number strings, 75%, 15%, um, 10% and then 5%.
So the compute that's available for training AI models in the US, China, Europe, and, you know, 5% is for 5 billion people around the world.
So there is this immense concentration of talent as well in the Bay Area, in some other parts of the world.
So how do countries get going in the light of this? And At the same time, we are getting some food for thought when you listen to the CEO of IBM say that all these trillions of dollars that are being bet on the scaling approach to AGI, you need nearly 1 trillion of returns every year just to be able to repay the loan.
So what do they bet on these countries that are in that 5% category? Well, I agree with him that right now there's a big question as to whether the amount of revenue generated from AI services can justify the investments that are being done at the moment.
It's not clear at all.
There's some number that was recently published.
It may or may not be correct, but I'm going to repeat it anyway.
A typical professional subscription to open AI Anthropic is $200 a month.
The power users, the amount of the cost of serving one of those power users that pays 200 bucks a month is about $15,000.
This cannot go on for very long.
Right now the use of AI is being subsidized by the investors in those companies and in this infrastructure.
At some point, the revenue are going to have to be commensurate with the cost, which means either the price has to go up or the cost of inference has to go down drastically and probably both at the same time.
And in the end, it's not clear that the service rendered by those AI systems would be cheaper than humans.
It's already the case for program generation that if you pay the real cost of code generation, it's probably as expensive as a human.
Some companies are rethinking their strategy about this because of this.
There may be their reckoning over the next year, possibly, which would be similar to the Internet bubble crash in the early 2000 when companies were laying fiber uptakes everywhere, and then a lot of those fibers went dark, and then some companies that Google snap them up because they were cheap.
There might be a similar phenomenon where we're going to have a lot of cold GPUs, and I'm secretly crossing my finger for this to happen because it will make uh, you know, the training of Omar a lot cheaper.
But, you know, I can't make predictions there.
The people running those services, I think, know what they're doing to some extent, at least.
You had a previous point which I forgot.
Well, maybe we can bring that point up in the final question for our chat today, which is what's really useful today? Because apart from this concern about sovereignty, what do I invest in, the question that comes up is what's useful for the development challenge in agriculture in health, in education, small language models, narrow AI in some areas, perhaps LLM.
Where do you see the most of the low hanging fruit, the most value coming out of that kind of low hanging fruit? Certainly, a applications in things like agriculture and things like that are still based on LLMs.
Basically give knowledge information to people.
Do not require the top of the line super expensive models.
You can do this with relatively simple small models that you can run locally in some cases.
I think there is a lot of opportunities here.
I should say also on the hardware side, the kind of hardware you need for training LLMs or training AI systems, and for running them is very different.
There's been some success in hardware companies that have been attempting to break the Invidia monopoly essentially on the side of influence.
If you want to serve the entire population of India, the cost of influence, farmers in India, for example, they could be wearing smug glass.
The experiment was done actually by Ma.
You give smug glasses to farmers in India and they can look at their crop and ask their AI assistant, is this the right time to harvest? What is this disease? I'm seeing on my a crop? What kind of weed is this? Should I get rid of it, things like that.
This is extremely useful to them as long as those systems speak their language.
But the cost of inference must come down by a factor of 20 to 100 for this to become practical to a wide population.
A lot of work to be done in hardware, efficient software.
There's a lot of incentive for existing companies to do this because most of their operational cost is actually power for the data centers, so they have a huge incentive to reduce costs as well.
But that shouldn't stop countries, companies, researchers to participate in this.
Great.
Please join me in thanking Yang Kun for being with us today.
Thank you.
Thank you.
Thank.
Thank you so much.
Thank you so much.
That was very exciting and I'm sure that we're all now very much looking forward to the rest of the day.
Without further ado, we're going to move on with some setting the stage.
The first one is going to be Ser Higago who is the Chief Technology Officer at Cloudera, where he leads the company's technology vision and innovation strategy, aligning cutting edge data and AI capabilities with enterprise needs across hybrid and multi Cloud environments.
Please welcome to the stage.
Thank you very much.
Well, it's a hard act to follow.
I'm going to try to first beat the drum a little bit and then set the stage for today on what is what we need in order to have what resell lamps or world models or robots, which are essentially AI systems with some rbo engines on them, and what is what we need to make sure that that happens to all humanity in equal terms.
Now, I was starting to write these remarks last week and something made it much easier to do it because something happened with the Fable and myths models that made us Europeans a little bit enabled to use them.
And then this made this speech a bit easier to write.
Imagine a future where individual, every company, every country can only hire employees from four companies, five, maybe three.
If you want to hire someone, it has to belong to one of these four companies.
These four companies happen to be in the same place in the world, happen to have the same hive mind.
They've been trained, educated in the same culture.
The companies can dictate how these individuals are, their skills, their salaries, and even when they don't want to serve you anymore.
Now, if you say, sir, that is crazy, that is not that different from the future or present where we are going.
So we often speak as though AI begins with a model, of models, weights, and so on, but it does not.
It begins with data and infrastructure.
And as to that institutions and people.
That's what trains models of any kind.
If those foundations are fully concentrated, opaque, and inaccessible, the AI built on top will reproduce all those biases, only faster and a greater scale.
In other words, let's not start with the weights.
That's the tactical implementation.
Let's wait on the infrastructure.
So I'm going to try to make three points today relatively quickly.
One is that interoperability Is a condition for participation.
Two is that sovereignty is a condition for continuity, and three is that private AI is a condition for having operational responsibility.
So interoperability is usually treated as a technical detail.
It's not a condition for freedom.
A hospital should be able to use one storage system for their data, another analytics engine, a locally adapted AI model without having to copy sensitive data from patients from one place to another or hoarding into the same data platform.
A government should be able to replace a technology provider without having to rebuild an essential service or public service, and a researcher anywhere should be able to use open formats and interfaces and adapt them to local languages, local data and local leads as seamlessly as possible.
Therefore, open source AI cannot mean simply publishing the weights.
It has to mean something else.
An open model running on proprietary data formats, on proprietary orchestration, on proprietary cloud interfaces, or proprietary governance, it's a locked system.
S has an open door painted into it.
So we need openness across the full end to end spectrum and private companies and institutions have to see this in very practical terms.
I represent the private sector.
Open table formats.
Iceberg is one great example that community brings.
Open engines and catalogs like Polaris as well.
Open source compute engines and APIs.
This is the convergence of many of the projects that Apache Foundations and others have been working on for a long time in the data side and ecosystem and converging that into AI.
Having different teams and technologies to work with the same govern data without forcing the necessary copies, moving lots of gigabytes, betabytes, or exabytes from one place to another, and forcing everyone to put everything in the same ecosystem that belongs to one single company.
Interoperability allows an institution to replace a component without having to change everything or without forcing all your data to go into the same vendor.
It prevents today's procurement decisions from becoming tomorrow's permanent dependency.
It's the fundamental element on the ecosystem for the future or I should say, for the present of today.
Now, for years, this is my second point, digital sovereignty was more like a policy preference, especially for enterprises and corporates.
So decidable, but too complicated or expensive for many players in the industry, and that position is no longer credible.
AI is now a fundamental building block for administration, healthcare, education, defense, finance, you name it.
Um, so when this happens, control over the technology becomes a matter of continuity.
The technology itself becomes the critical infrastructure and all the elements that are together with that.
A private provider can change the price of a token.
This happens every week.
It can change the rate limit, retire a model, modify its license, the quality of the output, and it can alter the terms of service or decide a particular capability that will no longer available in a given market.
And a government can impose, of course, an export restriction, a geopolitical dispute that can affect those accesses.
In essence, an institution or a company that believed it was buying technological capabilities suddenly realizes that it only rented permission to use it on a temporary basis.
And we just saw an example of that last week.
So imagine that dependency in a hospital in the tax authority in an electricity grid.
Your national AI strategy or your corporate strategy cannot depend on someone else's terms of service.
So AI sovereignty and private AI together or need to answer the following question.
One, where does your data really reside? Two, who can access it and how and under what conditions? Three, which models can use that data or family of models? Four, can we move the workloads, the pipelines from one place to another seamlessly? Five, can we replace the models instantly? One to another and the systems continue working.
Six, can we audit and inspect the system? S and most importantly, can we continue operating in this if a provider changes it commercial or political position? Don't get me wrong.
Real sovereignty does not mean isolation or autarchy for that matter.
It is the ability to participate in a global ecosystem without surrendering control of the essential capabilities.
The solution for that, Mr.
Lacoon already mentioned that, but is open source, which is central to that answer.
It converts dependence on a single supplier into participation in a served economy.
Now, open source alone does not create sovereignty.
It removes the, the toll booth, but it's up to us to build the roads up to every country.
Private AI means making responsibility operational, making it happen.
It doesn't mean private models.
It means being open to the foundations, but private in the operation, essentially being accountable for the outcomes.
Institutions cannot simply sell all the sensitive data to external black boxes.
They need to bring AI to the data rather than moving data to the AI that comes with infrastructure and digital commons.
The ability to run the models in a public cloud.
Those are good and helpful.
A sovereign cloud, a lot of that going on in Europe, and a private data center or at the edge while applying consistent security and governance and other controls across that.
In a way, interoperability makes replacement possible, sovereignty makes the decision possible and private AI makes control operations possible, being able to have the whole supply chain under wraps.
Open models can be deployed privately, adapted to local knowledge and used with clearly defined boundaries.
An investment in true open source AI, and the key here is what is true open source AI is a key element to realize that AI for humanity is something that we all use.
And by here, I mean LMs but also role models, quantum algorithms, and anything that the future can bring us as well.
Responsible deployment requires identity and access controls, data lineage, evaluation, bias, security testing, red teaming tools, continuous monitoring, transparent incident reporting, audit logs, human oversight, and appealing mechanisms with the ability to shut a system down when required.
All that is part of the digital public goods.
We need to move from innovation to deployment.
Many of the AIU cases get stuck in the experimentation phase, never see production.
This is especially true in the enterprises, in the corporate world, but also at the institutions.
Our industry loves demonstrations, creating we factor.
But humanity does not benefit from those demonstrations.
It benefits from systems that work on Monday and on Friday they are surviving an audit, systems that are HIPAA or GDPR compliant.
That end wine is what we need to provide our citizens and our customers alike.
Deployment discipline, clearly defined public benefit.
Why are we doing this? Legitimate and representative data, security testing, continuous monitoring, and a route for hu manipulation and readiness.
Governments should protect rights and use procurement to require open standards like the ones I mentioned, interoperability, portability, auditability, and credible exit plans.
Researchers, civil society, academia should test those claims, expose failures, localize systems, and ensure that affected communities, big or small, have a meaningful voice on creating them.
And maybe here in this venue, the UN can help a lot with creating a common ground, coordinating standards, sharing safety resources, supporting capacity, and ensuring countries at the edge of AI can also become active participants on the future.
So I'm going to be wrapping up.
If we get this right, and again, for all types of models in AI, Open source will not merely be or make AI cheaper.
Maybe this inference becoming cheaper as Mr.
LeCun was saying, it will make AI more contestable, adaptable, locally relevant, and worthy of trust.
AI for humanity for all of us, means that a teacher can adapt the system to local curriculum, that a public health agency can use sensitive information without surrendering control, that a small country can participate in the AI economy without renting its future to a company that's valued ten times its gross domestic product.
That a citizen can know when AI is being used, for what, and understand the basis of important decisions that are being made.
Don't get me wrong.
We've done this with classical machine learning in the past.
We're not necessarily reinventing the real open source, not just the weights, but the content, the data, the training, the data platforms, and the data lineages, gives us the possibility of building a commons.
Interoperability keeps that commons fully connected and private AI allows institutions to operate responsibly for themselves and for their citizens.
Sovereignty ensures that humanity and not a contract, the price and pays or a distant political decision remains in control.
It is not technological nationalism.
It is not isolation, is the foundation of resilient, democratic, and human centered AI, and it is no longer optional.
Thank you very much.
Thank you very much, Serjo and we're going to hand it over to the next setting the stage speech, which will be given by Thomas umbek who's been having a very long political career in Germany.
And since May 2025, he's been the parliamentary State Secretary to the Federal Minister for Digital Transformation and government modernization.
Over to you.
Ladies and gentlemen, esteemed Secretary Jill.
I'm very pleased to be here and it's very important that UN is tackling the issue of open source as a German government.
We very much appreciate that and we also want to contribute to that.
That's very important because when we speak about open source, that means freedom.
It means freedom for you to choose solutions.
It means for you freedom to understand solutions.
That means freedom for you to create your own use cases on the solution and it means freedom that you can operate it without any hazards and without any interference from third parties or third countries.
So therefore, we very much appreciate all the initiatives here and today is not only the fifth time that this open source week of UN is happening.
It's also the fifth year that the sovereign tech agency that we founded as a German government is working there.
The basic idea is to contribute to the open source ecosystem without any political agenda.
That's really important without any political agenda.
We want to prevent the freedom in this open source ecosystem, and therefore the sovereign tech agency is built independently and can make independent decision and is not under some kind of political control.
We want to scale that.
We want to scale that in Europe.
That's the reason why We are doing an organizational form that brings other countries like our friends from France, for instance, and further countries on the place.
If there are further partners even outside the EU who want to cooperate and to scale that governments in the free world with free democratic values without political agenda, want to scale here, you're highly welcome.
So having said this, today the question is, what do we do on AI? And so the basic constitutional principle in Germany is that we say public money, public code.
And beginning with the sovereign tech agency and all of the projects that are being made there, there is one project that's the Agenic maintainer Support.
I think that's pretty important as maintainers are challenged with all the AI content that's distributed and challenge for their time.
And so I think support here might help.
The second thing is that we built and even though it's not really AI, but it's the base for that, we build a personal ID wallet.
You can install this on your smartphone and for us, it's a basic principle that the whole source code of this central identification system of the government is published as open source so that every citizen can see this, that we have full transparency.
There is no tracking or any kind of things like that, and we make it transparent so that it's proven.
The second thing that we do is that we build an AI platform for citizen services.
We have in Germany very much on the municipal level.
We have thousands of municipalities with a lot of independency, but it's complicated if you want to do any proposals to find the right place in all of these municipal services and federal services and whatever.
We're building an AI platform with an app so that you have a centralized point for all the proposals to the government, even if it's local or on the federal level.
This is also published as open source so that there is full transparency.
You guess that there is no tracking or things like that behind, you can see that.
The third thing and that's really important and the teams also here today is Spark.
This is a platform for large infrastructures.
If you want to build railroads or bridges or highways, in a democracy, it's a pretty complex process and sometimes it takes years and now we built an AI system with a huge budget on that.
That can, for instance, the first step is you need expert reports and you have a lot of documentations, and it takes a pretty long time to check if all of this is complete and consistent.
And the AI system can at first check, is it complete and if something is missing, it can immediately make an email or an announcement, say, okay, there is something missing and it can check if everything is consistent.
We save six months for a typical process here.
The next thing is we have objections, sometimes we have 100,000 objections and all this from citizens and all this can be handled by the system, so it's really accelerating all the processes for building large infrastructure.
The next thing is number four is our Agntic AI hub.
We want to foster startups.
S.
This Spark thing, you can you guessed it, you can completely download it as open source.
The next thing is our agentic AI Hub.
We have all these solutions and municipalities and we want to bring them to the next level.
There's a lot of legacy software, and therefore, we made a competition for startups.
Right now we started with the first batch, 20 municipalities and nine startups.
It's a lot of solutions, you can see there on open source and to explain what they're doing.
For instance, if you make a Social Security claim, you have to deliver 100 pages of documentation, what your earnings is and things like that.
The system also can check if it's complete, it can check if it's consistent, and also it can make an executive summary for the public officer so that they don't have to go through 100 pages.
They see immediately what are the core figures here in this claims.
And the last thing I want to stress is our initiatives around open desk and open code.
When I just mentioned that all of this is published as open source, you can see it on our own open source based platform.
It's open code.
You find it under open code dot d, open code, and there you can download everything I speak here about.
And the thing that's also been developed is Open desk.
You may have heard about that about that issue there at the International Court of Justice in Denar and after they had, let's say, problems with Microsoft, we could help out, and so they're running on our platform and we also would invite you and appreciate if we found further more users not to replace anything, but to have choices, to have an alternative, to not be locked into a single vendor with all the things that are upcoming, and that's not only question of freedom or access, it's also question of pricing these days.
So in the end, I appreciate very much that today also there is this opportunity, I heard in the afternoon to talk about what is really open source AI and what is open weights AI.
A question also for us pretty relevant.
If you earn 1 trillion of numbers, do you really understand what's happening there? So that's open weights.
No, we don't.
And therefore, we very much appreciate more open source AI.
We are building this by ourselves.
These days is announced the German initiative that's called SophiI and they're building a full LLM.
I suppose it's not this frontier LLM kind of thing, but in a hybrid solution.
I think for a lot of workloads, this is attractive and this will be fully open source on this and everything that we need, I think, except of the US is a lack of compute and therefore, the German government also is investing and encouraging especially private businesses to invest more in compute because this is the beginning not only for inference, but also for training and building models.
This is the strategy that we have in a nutshell.
Thanks for listening.
Thanks for the opportunity to talk about all this here.
If you want to join something like this, we are not only open source, we are open government in this case, so don't hesitate to come to us.
Thanks.
Thank you very much.
We are now done with setting the scene and the opening remarks, of course.
I just want to check the room.
Please raise your hand if you feel like the scene has been set.
You're ready for the rest of the discussions.
Pretty good.
All right.
We've also heard feedback that it's quite difficult to hear in the back.
So to all the speakers that are coming up next, please speak as close to the microphone as possible.
I'm practicing what I preach.
And also, if you came a bit late, please note that there are earpieces.
You can use those to hear better.
In the back there aren't any, but I believe that you can connect your own headphones to those machines.
Let's try with that and let us know if you're still having difficulty hearing and we'll look into it.
All right.
Onto the next session.
So let me introduce the moderator.
I actually don't see where Verina is sitting, but please, con, sorry, Chief Executive Officer of Odata dot or, please welcome to the stage to moderate the session on open source for digital development.
We are not seeing Verina.
All right.
I will, in the meanwhile, hand it over to our colleagues to start off the session.
All right.
We have a volunteer.
I have to.
Are we sitting there or sitting? Sitting over there.
Sitting over there.
Do you have the list? I feel like this is what the tech takeover actually was preparing us yesterday.
All right.
Yeah.
Thank you so much.
Let me find a good chair and I have a lot of laptops and then screen and an iPad.
Yeah, that's good.
Let me invite our distinguished guests today for the discussion about open source AI and development.
We'll have a very high level ministerial session today with our distinguished Minister Sal Bai, that I invite her to join the panel.
I hope that she's in the room.
Sierra Leone, Minister Dill, thank you so much.
Please, I'll stand up so I'll give you space to go up on.
These are last minute adjustments.
Thank you so much.
Thank you so much, Minister.
Minister of the responsibility for efficiency Innovation and Del from Jamaica.
Her Excellency Adré Max, please join us.
And Minister delegates to the head of the government for deal transition and administration reform from Morocco.
Her Excellency Amal Falah Sagroshny.
Please join us on the panel.
Do we have Okay.
Thank you.
Please, we will put names for you.
Sorry for a last minute adjustment.
This is very tight space, and it's hot today.
Yeah, that's fine.
I'll give it back to you.
All very much welcome.
How are you doing? Yes, please.
Very nice to see you again.
Thank you.
Welcome.
Hello.
Good morning.
We are very happy today to host to show firstly that the gender problem is not a real problem and we have top leaders in the world coming leading their digital transformation, digital reform, AI strategy in three countries talking today around reality, how they can really do better than men because we have failed.
It was supposed to be a women moderating decision.
Please don't judge me on my physical aspects.
I'm happy to introduce um uh, the three ministers talking mostly about the open source AI and development and the vision that we need to build around that from the top levels, vision, from strategic vision, from policymakers visions, how this could bring this advancement in the AI adoption, into digital reforms, into digital transformation, within government, within public services to reach at the end, with the goal is to reach the citizens, providing better services, providing a lower cost in terms of services, but also ensuring that the countries are upscaling and are not leaving that gap getting bigger and bigger with all these AI advancements.
I'll start with all of you with the same questions.
I would like to get, firstly, how is open source integrated into your national deal transformation or AI strategy? Then if you have any concrete case that have delivered measurable outcomes.
We all know that we are talking about AI outcomes.
It's very hard to measure the AI outcomes.
I want to hear from three ministers today how we do that.
Your Excellency, Mr.
Em, please.
Good morning, ladies and gentlemen.
I'm very happy to join this very important discussion on open source and in particular in the domain of artificial intelligence.
We have been working on open source many decades before, but the rising of AI bring new challenges when it comes to open source.
For example, for many decades, open source was giving the source, the code of the software.
And then everybody was able to improve, to add, to remove, et cetera But when it comes to artificial intelligence sources, the question is quite different because you don't need only the code.
You need many other parameters and features that help you to fine tune the models, to work on the model, to work on the data, to work on the tokenization.
And you can open some aspects.
If you close some others, you cannot really speak about open source.
In Morocco, as Minister of Digital Transition reform Administration, I can say two things.
The first one is when it comes to pure digital, we have the same advancement as any places in the world.
But when it comes to artificial intelligence, I think we come from far.
We started a few years ago working really on artificial intelligence.
We deployed many efforts to develop AI and what we can propose to our developers is two things.
Data factory, we have two concepts.
The concept of data factory and the concept of forge of sources.
This means that people can first learn how to deal with data to be used in their codes, software.
The second is to mutualize somehow the development of algorithms within the country, but also we are open to the world.
We use other gitubs, et cetera, and to take benefit from development in other countries.
Today we are when it comes to AI, we are at the very beginning of the process, to be honest.
But we know exactly what we have to do and how we can do it because you have this experience of developing algorithms for labs, I would say, not to scale for the public administration or the other companies.
But it's going very fast and we do a lot of activities around the open source in AI and I have 2 minutes, I think.
Yes.
I still have time.
Yeah.
Okay.
So if I would like to summarize very quickly the most important obstacles, the first challenge for us is the fragmentation.
For a while, many administration were developing codes, algorithms, et cetera, and if you would like to have something coherent, in particular, When you tackle the question of interoperability between administration, you should have more deep knowledge on the pieces that exist in different places.
The second challenge is data readiness.
For a while also, we had many papers everywhere in the world, not only in Morocco, but I think in Morocco and Africa, it's still a huge challenge to move from papers to have real data processing data, to move from papers to a processing data.
Maybe the third challenge is sustainability.
Because for most people, open source is seen as volatile sources.
There is no maintenance, nobody can provide support when you use the open source, et cetera This make some Introduce some farce in some fear in the process of development of AI and based open source AI.
This is why we put this forge because we need some safeguards to bring to the table this open source to be able to make the added value on these sources, but also to provide some guarantees that it will work and somebody somewhere will take care of the advancement of this source.
Of course, the last but not least challenge we face is the challenge of confidence, the trust.
We think all the time about trust property AI.
When it comes to open source, it's worst.
People have to trust that the source is solid, the source is resilient.
The source is not full of pitfalls, for example, in the development of the source.
We have also to work on the trust toward the use and the integration of open source in a huge development.
Thank you.
Thank you so much, Excellency.
The trust, indeed, it's a big problem when it comes to the open source community.
I remember that meaning as a single developer, it was a big trust issue.
Now, if we take it forward to the level of society, citizens and public services, and then in particular with the rise of the AI.
I'm looking at you, Excellency, Honorable Adré Marx.
I'm asking you about that trust issue that just Mierel was raising.
Um, We know that AI brings a lot of attention and spot the lights on the open source in particular, and then that's why we are organizing evenly that day.
How is the raise of the awareness, the uptake of the considerations of the open source within the government, but also at the societal level is happening? Thank you.
It's on.
Okay.
Okay.
Thank you.
Let me first say good morning to Under Secretary-General Gill.
Excellencies, ladies and gentlemen.
Mr.
Moderator, you have spoiled my opening line was supposed to be Madam moderator.
I was fascinated that we had all female ministerial panel, but we will work with you.
For Jamaica, open source AI is fundamentally a capability issue, not just a technological one.
As we accelerate our digital transformation, we have moved beyond viewing open source merely as a cost saving measure.
AI has fundamentally shifted our perspective.
We now recognize open source as the essential mechanism to inspect, adapt, secure, and govern our technology.
Our vision is clear.
Jamaica must not simply be a consumer of global technology.
We must be active builders.
This require a concerted effort to cultivate data foundations, specialized skills, robust safeguards, and public sector capacity so that Jamaicans can fully participate in the global AI economy.
To date, we have laid a strong groundwork with seven major open source and open data government initiatives alongside nine significant AI policy programs since 2011.
Our progress is accelerating rapidly.
We have established the National AI task force, completed the UNESCO AI readiness assessment, and launched the National AI Lab and what we call a gains program, growing AI Innovation and national skills.
We are driving implementation through mandatory training of our public workers and AI initiatives in health and targeted AI pilots.
Our existing foundation, including the DCN based Open data portal, data for development, EMIS and DHIS to cancer surveillance, demonstrates our commitment to open source.
Moving forward, Jamaica's message is simple.
Open source AI must build enduring capability, not dependency, enabling us to serve our citizens better, protect trust, and create lasting Jamaican owned value.
Thank you, Mr.
Moderator.
Thank you so much.
That's very That's very impressive.
I know about the capacity building issues, the skills issues.
I'm trying to do a lot of that.
It's very good to hear how far have you gone into that.
I'm looking to our last panelists, not least, Your Excellency, Salb Minister from Sierra Leone.
I've been there once.
I've been there once with great colleagues working on AI strategy, helping and trying to understand how we can support at some point.
But I'd like to get a better vision from you.
You are the top leader there on the DLA in AI.
How is AI and open source going in Sierra Leone? Thank you.
Thank you very much for that.
I'm always excited to hear people have been to C Leon.
It's you've crossed the boat.
So you have interesting stories to share, I'm sure.
But really, again, I want to join my colleagues in thanking the undersecretary for putting together this really critical conversation and really everybody that's here.
I really believe when it comes to advancements of digital transformation, really, these are the conversations we need to be able to have a bit more critical I think specifically when we think about open source and open source AI now as we're discussing, I believe the experience of Sierra Leone or specifically our vision when we think about it, I think, could be illustration for a lot of countries to be able to also think through and we're happy to work with other people to see how we can also further advance that.
Because when we think about it, we think about the potential to change the economics of digital transformation.
I think any country that sits down and think about your digital transformation agenda and journey, one of the hugs implications within that is the cost implications of all of that.
And if we think about African countries being on this journey for long working with our partners to transform, as my colleague from Morocco mentioned, in Africa, a lot of our processes are still paper based.
So We know everybody's working to be able to transition that because we want more accountability and transparency in terms of government service delivery, how we operate, and we know technology is critical towards that.
But we look at what's been the biggest challenges towards that is some of it is the cost implications.
We look at building or procuring proprietary based systems when you're vendor locked and not having those finances to keep up with those and licensing based subscriptions and you realize it's almost like you're you're entering a trap and then you keep on going and going, it sometimes feels there's no way out of that.
That's why we believe open source, it's a great opportunity to change that reality that we have.
Then maybe also most fundamentally, one of the When we look at it, especially across the region, we see different African countries building the same solutions, all using the potential limited resources that we have, building the same things at the same time when we don't need to be doing that.
That's why Ce Leon has always seen open source as one of the biggest game changers when we think about digital transformation and specifically for developing countries such as ours.
The idea that we don't have to build the same thing, walk the same road that somebody else has already done, can take what is there and build on top of it, improve on top of it, contextualize it for what we need in Cer Leon.
That has always been critical.
That's why Cer Leon is a founding member of the DPG Alliance.
We're very proud founding member of that alliance, being able to have a seat at that table to be able to contribute to the global advancements of DPG, we think it's critical.
That's why we see Cerion has always been a home to we see some of the biggest open source solutions right now, DHIS two actually was piloted in Cerion when it started.
Open GTP was piloted in Ceron because we wanted to create that home where open source software was welcomed and it could be a place where this could be tested and scaled from because we want to provide that opportunity.
And we're further doubling down on this commitment to open source.
I'm actually recently supported by UNICEF and some of our other partners who've developed the first open source first policy.
And really what this means is to say that in Cevion before we consider any proprietary based solution, we will consider an open source as our first priority as our first go to only if we can't find an open source or that we can contextualize and work on, then we'll look at a proprietary solution.
This has been great for us because issues of fragmentation as everybody mentions, issues of these silo systems.
Issues of sometimes these proprietary systems not really being contextualized for really our needs.
Sometimes we find ourselves with these proprietary systems, we utilize them But there's lots of difficulties within them because they weren't built for us.
So getting them to work for us sometimes is really difficult and also the issues of sustaining it and issues of sovereignty really, I think, if we look at the digital landscape and where we are.
For us, open source has been critical.
I think our commitment and our vision towards this can also then translate to when we talk about open source AI.
I like to say it also provides the opportunity to democratize intelligence.
Because with open source AI means that everybody can contribute and build towards this frontier in terms of what we're going through.
Really in a nutshell, I would say for Slone, open source has been critical.
We're doubling down on our commitment towards it and one of the areas in which we're doing this actually is we're building a DPG pipeline because when we think about the advancement of open source, sometimes within the region, We feel as if it could appear as if it's this thing being done to us and not being done with us.
So we're looking to how do we create the opportunity to how do we converge demand, talent, and resources when we think about open source for the region? Again, supported in partnership with UNICEF.
We're developing this DPG pipeline platform that we believe can provide that convergence, not just for Ce Lon, but we think it could be a tool that could really support the advancement of DPG within the region because Again, one of the things we want to be able to change is sometimes the perception of DPGs are open source, it's free.
Free means cheap, cheap means bad.
We're trying to change that narrative.
We we change that by getting more of our local talent contributing towards this and building towards this.
The platform of itself is open source as well, so invite others to further improve upon it and build upon it and contextualize it for their needs as well.
Within that, we think that our open source framework is building the foundation.
We think within AI as well, the work we're doing, we just developed an AI readiness assessment because Before moving to an AI strategy, we wanted to be able to understand What role does Civilion want to play in this whole AI revolution? What is our comparative advantage? Our comparative advantage might not be building the AI based infrastructure.
It might be looking at the datasets and how we're contributing towards that, we want to be able to define what our role is going to be and have a focus on how we achieve that.
Because for a country such as Civil Leon, that is important.
We don't have unlimited resources.
With the resource we do have, we want to be strategic in how we utilize it and how and how we contribute to the global advancement MOVI.
We're working through that process, and as mentioned, and we look to also provide regional leadership.
As the current chair of ECOS, Sean is the current chair of the ECS, so I also chair the ECS ICT Council.
We just had a ECS ICT Council meeting.
Some of the conversations were around as a regional perspective, how is West Africa, what comparative advantage does West Africa have and how are we able to come together.
As mentioned, our biggest challenge is fragmentation.
We believe within the region, there's opportunities to leverage shared infrastructure rather than every country building their own GPU, every country building their own data censors, which we might not have.
There's an opportunity as a region to come together to see how in the true sense of the word, make infrastructure also open in that sense where every other country then would seek to see how their own competitive advantages could be.
So lots of progress being made, but still lots more to do and looking forward to working with all partners to see how we further advances.
Thank you.
Thank you so much.
That was very comprehensive, covering entirely the value chain of AI from the governance to the data centers at the end.
Thank you so much.
Just to raise a point related to the AI adoptions and open source, we have seen that many strategies are coming to us when we talk about sovereignty, about pushing more of the private sector to take the lead on that because that's a fundamental piece in the thinking about AI.
Um, but when it comes to the AI and open source, how to convince and bring start ups building their autonomy, building also a partial sovereignty of the member states on AI, it's a bit more complex because it's a mix of things that are coming inside and things getting outside.
So with the arrays of these big startups or private economies we've seen in China, we've seen in the US, we've seen in Europe, a lot of investment, state investment into the startups to build that ecosystem, to bring the agility, to bring the innovation capability, but also to bring the ecosystem inside because AI is good, getting benefits from AI is much better.
I know there is a lot of countries running towards the SARub Act, building this and building the ecosystem.
But when it comes to the AI, I'm looking at you, Minister Mel Flare, how Morocco is doing, meaning I know it's a very vibrant ecosystem, but how bringing the AI component inside of that ecosystem of startups? Yes.
Thank you very much for this variable question.
First, I would like to just recall that in Morocco, our digital strategy 2030 has two main pillars.
The first one is digital economy and the second one is reform of administration and both work together.
Because when you work on digital economy, you need laws, you need procedures, many things that rely on the administrative procedures.
We need advances in administration to go ahead with digital economy.
Also we need digital economy to advance the whole stuff in administrative area.
In Morocco, this digital economy is can be articulated on two or three axes.
The first one is offshoring, outsourcing, if you want, and we attract many companies, more than 1,200 companies over the world, big companies that invest in Moroccan talent and develop AI within Morocco.
Just to give you more comprehensive view, today, we have 143 1,000 jobs created by outsourcing in Morocco.
Of course, to do that, we need incentives and we need a very vibrant startups, very vibrant development, et cetera when it comes to developing AI, I think the champions or the leaders of AI in Morocco our startups today.
The startups are, of course, developing new technologies, new solutions and try to connect these solutions to the big stakeholders, including the companies that provide this outsourcing.
We try to have the the whole chain from development to start up to big companies to export digital and one of the biggest incomes of Morocco in digital economy is digital export of this kind of solutions developed by startups and in the context also offshoring.
Have two big programs called related to venture building and venture capital, more than 13 million of Durham, which means $13 billion dedicated to to develop the startups solutions.
Very recently, last Friday, we got the The opportunity to develop also this ecosystem by the World Bank helping us by $250 million, which include 200 million for startups and digital ecosystem.
This means that we have today proved that Moroccan startups can provide a very helpful solution to the rest of the world, and this make World Bank trust in this development within Morocco.
So uh We also we have, I think something very new in the area of ecosystem called Institute Jari which are physical spaces that put together researchers, startups, stakeholders to develop open source solutions.
The Jari institutes are spread on the country, we have 12 they are specialized by region and they provide solutions for regions.
For example, in the South, we have problems with water, so we will specialize the Jas in water.
In other area that will be energy, in other one is industry and so on.
We create this network of Jari Institute with the more than one gigabatt capacity of compute.
Today, we invest in data centers, we invest in compute, we invest in talent, and we invest in startups to build this ecosystem at large scale.
I think what is very important for Morocco and we will have some events during this week, in particular with UNDP, Morocco is today chosen as a hub for digital and artificial intelligence, a hub for AI and data for sustainable development within Africa and Arab state.
We try to provide solutions that can scale at the regional level and not only at Moroccan level.
Our ecosystem is very vibrant.
We have very brilliant minds in mathematics, in algorithm, et cetera, and we are very confident when it comes to talents in Morocco.
Thank you.
Thank you so much.
That's very impressive that you are that precise in the articulation of the impact and job creations.
This is the first time I hear that very precise numbers.
I was wondering how can we do that? The same thing for AI and say, Oh, the AI is bringing that on GDP or on the job creation.
That's another discussion.
Your Excellency, Adré, meaning, it's the same question in different region.
It's the startup, the ecosystem, the market, but also the open source, how this could contribute to faster and build a better startup and innovation ecosystem.
Thank you.
Margaret just a moment.
Yeah.
Thank you.
Thank you.
Open technology is an important enabler of Jamaica's AI startups and wider innovation ecosystem.
The lower barriers to entry, speed up experimentation, and give developers access to advanced tools and knowledge that would otherwise be costly or difficult to obtain, especially for startups.
This allows entrepreneurs to build on existing innovations rather than starting from scratch, improving time to market and competitiveness.
However, the challenge that we have to note and recognize is that while open source reduces license costs, it introduces other considerations such as the need for local technical capability, ongoing maintenance, and in some cases, vendor or community support.
These operational factors must be planned for to ensure sustainability.
In Jamaica's context, the strongest approach is combining open standards, secure APIs, and shared public digital platforms to create a foundation for innovation.
I'm very pleased to share that back in April this year, we were able to launch Jamaica's data exchange platform.
So it's very easy for governments, the different departments and agencies of governments to operate with their own data.
But we are now cutting all the silos and ensuring that we have shared data which will also provide a common area for entrepreneurs to access data.
This enables developers across sectors such as agriculture, tourism, education, and government services to build solutions more efficiently while contributing to more connected and scalable national innovation ecosystem.
Thank you, Mr.
Moto.
Thank you so much, Your Excellency.
Please.
Yeah, feel free.
Probably the data exchange is the hardest part to achieve in particular when it comes to how it's very siloed across a different administration.
I don't know if you are pushing the I'm asking because I'm really curious to know, are you making the data also available for start ups? Yes.
It's secure data, so there will be a process.
We are starting with government departments and agencies, and we have now put in a data protection system.
And our companies, even startups, will have to register and go through a certain compliance to ensure that data is protected, and then there will be access and is especially important for the banking system in Jamaica.
At this point, it's still relatively complicated to get bank accounts.
You have to go and get references and show a number of documents.
We are doing two things.
We are doing the data exchange platform and we are also doing a national identification card.
Every Jamaican will have a national identification card, which will be the one card.
Now, with the national identification card which has already gone through the process of verifying your identity and the data exchange, we are combining both so that to make it easier for you to open a bank account and for startups to be able to identify who they are working with and be able to get going much quicker.
That's setting the stage for tomorrow's discussion on the public infrastructure.
Thank you so much, your excellency.
I'm looking at your excellency, on the startup.
Meaning, I don't know much, to be honest on SAR ecosystem in Sierra Leone.
I'm happy to hear a little bit more.
Then again, how is open source could help building a stronger Su ecosystems in AI at least.
Thank you very much.
I really like conversations around the private sector's role responsibility towards this because I feel it is sometimes within this government government.
I think sometimes we have to central this where we always needs to be in the private sector role.
I think specifically on startup ecosystem in Civil Leon, I think as might be expected, it's a nascent stage, capacity building.
We're seeing the development of VC and funds coming in.
I think it's always a good indication whether your ecosystem is developing is when you see funding coming in because that's always will determine whether the investors see this as a viable ecosystem to invest in.
So we're seeing a lot of that, so we're encouraged by that.
But within this framework, and I really liked the questions or responses around the impact of AI and because this conversation AI is not for AI sake, right? We're looking at how AI can improve healthcare, how AI improves finance, how it improves agriculture, how improves mining, how we really transform and change aspects of our lives and for the better, right? So that's really the gain within all of this and why these conversations are happening.
So But then it's also critical then to discuss what's the role and responsibility of private sector because if we think about the impact and the potential of AI, and I really would like to take it even beyond our local ecosystem where now the world is a global village with technology and these systems now it's really a one part which we're all in, especially when it comes to the impact, social media, AI determines what you'd find out about Cervi Leon today.
If you go on Facebook, the algorithm has decided the most interesting thing about Cer Leon today is this, and all the way from the US, that's what you get to see.
That is influencing your opinion about a specific country region and has real life impacts when it comes to financial.
If you see a great story that portrays the security and advancement of Cer Leon as an investor, you might think great place to go invest.
If you see a story about instability, security risk, Investment medic, that's not where I want to go invest my money.
So these are real life impact healthcare.
AI now is determining global health outcomes and how it runs.
There's real life impact and within that, I believe the private sector has a critical role to play when we talk about lack of representation within it.
If we're really going to see the opportunity for AI to actually reach its full opportunity and potential that it has, we really need to think about where does the obligation when it comes to ensuring representation lie? Is it just the sole responsibility of countries and governments to push and figure this out, or is this a collective responsibility within that? Because AI models that are not fully that doesn't understand the full context of full humanity as a whole, that actually be an intelligent system? Can we claim that to be fully intelligent if it doesn't understand and take into consideration every aspect of this? I believe within this when it comes to impact in the private sector, I do think we should have a more collective conversations around how do we as a collective and and contribute to this because that's the only way AI of itself would really achieve its full potential of what it can do for humanity and for the advancement of our shared agendas as well.
And I'll really end it by saying, specifically when we look at the global digital compact that was signed with private sector partners, big tech private sector partners being parties to that, one of the critical things we talk about is digital cooperation.
How do we really ensure everybody is well and represented? I think within these conversations, I think I would like to see more conversations with, I would like to see one or two private sector partners here maybe as part of this panel because I think Those are the conversations we need to have for AI and open source and all of these that we're talking about all of it to achieve the full impact and opportunity for humanity.
Thank you.
Thank you so much.
Thank you so much.
I'm offering you two options to continue questions or we take one question from the audience.
It's up to you to decide and I think it's good to hear from the people who are participating.
Is there any question before? If you don't have questions, that would be very hard.
Oh, we have one.
I can see it.
We have a question there.
Anyone could help with a microphone.
Yeah.
Maurice.
A short question, please.
And if you sit by a microphone, you can press the button and use it.
So the gentleman back there.
So it's not working.
I'll come to you.
Thank you so much.
Really appreciate the conversation.
My name is Esanha Hagan.
I work with a nonprofit community centered data and Research Institute in Washington State.
The conversation of contextualization and also building a shared infrastructure came in this discussion, which was very interesting and fascinating.
I have a question about when we talk about the contextualization of the models or the infrastructure, um, Can you talk or share some of the samples of the work or the project that you guys are doing? Also, how can we have a shared infrastructure when we're talking about contextualization? This was basically the question.
Thank you.
I really appreciate it.
Thank you.
We will start with the Kingdom of Morocco because we talk about the contextualization, please.
Yes, this is a very interesting question.
Thank you.
When it comes to context, it means that we would like to deploy our algorithms in very specific manner that fit with the needs of the population or the city or the government.
I can give you one example.
It's about large language models.
There are two kinds of contextualization.
The first one is how to use large language models in some areas or under some infrastructure constraints.
When we think about, for example, languages, The existing models don't fit really in the needs of the population in Africa, for example.
Very few models tackle some rare languages or some very specific languages.
By rare, I mean the number of people using this language.
For example, in Morocco, we have Arabic, we can speak French, English, but we have Amazir, which is a language very specific to Morocco.
And we need to contextualize the models and sometimes we have to build the models in order to take into account some very specific structures of the language and also data, et cetera The second contextualisation I see is that sometimes we have to move from large to small language models because of resources because of logistics, because of power computation, et cetera I think we should really be careful when we want to use existing models, in particular, that's the advantage of open source that you use existing development by others.
We have to be enough aware of the limits of reusing things and be able to adapt and to contextualize this software to your needs.
Thank you.
Thank you so much, Your Excellency.
I hope that it was a comprehensive answer to your question.
P, unfortunately, we have to end decisions.
That was not in the plan to end it that quickly.
But I'm looking at you do you have any last message to the audience? Sure.
I want to say in Jamaica, AI adoption is gaining momentum.
But its use of open source AI today is still at the foundation building stage, not yet at scale deployment.
Open source AI should therefore be understood not only as access to models or code, but as a complete ecosystem that includes documentation, transparency, security, local adaptation, standards, and skills development.
Jamaica has important enablers in place, including strong youth talent through our AI Innovation Lab at the University of Technology.
And let me pause to say another great example and I see the intelligible CEO is here, Jamaica started to introduce hackatons at the last Hackon when we just started, we had 150.
The last Hackaton was 750 young people churning out to spend a day, 24 hours in one location building solutions for the problems that they see in the country.
We are pushing the youth policy and along with emergency policy structures such as the National AI Task Force and a solid data protection framework, alongside investments in skills and research.
Current initiatives supported by international partners such as the World Bank and the IDB and local institutions are helping to build and the EU are helping to build this capacity through training, research collaboration, and blended financing models.
The direction of policy is clear.
Strengthen domestic AI capability so that Jamaica can not only use EI assistance, but also understand, govern, and improve them over time.
Ultimately, success will be measured by our ability to move from dependency to active participation and contribution within the global AI system, ecosystem.
Jamaica's message is simple.
Open source AI must build capability, not dependency.
It must help to serve citizens better, protect trust, and create lasting Jamaican owned value.
Thank you, Mr.
Moderator.
Thank you so much, Minister.
It was very exciting having you with us today and very, very informative to learn about all the effort here, and I think this is how it should be done.
Thank you so much for coming.
Your Excellency, Sennar that, please.
Any closing remarks? No, absolutely.
I think I would like to close to actually answering that question around shared infrastructure because I do believe if there was any message to leave here with and is really on that is the opportunity because we talk about AI, one of the biggest things barriers towards that talk about data, when you talk about the infrastructure elements, I think that is very real for our region, so that's why shared infrastructure average actually provides us with an amazing opportunity to be able to potentially leapfrog decades in terms of our development and our participation in the global AI advancement.
I think it is critical.
One of the things we're working towards that is to make this possible because shared infrastructure, there's so many other factors that comes into it to consider and deal through.
For example, at the recently concluded ECOS ICT Council meeting, one of the things Ser Leon was able to bring to the table was actually for us to develop a regional data embassy framework.
This would ensure that because we talk about data, one of the quickest things a lot of would point to is issues of data sovereignty.
Issues of sovereignty, where is my data stored, who has access to my data, who's utilizing my data? How can I utilize it? I think we look at the global space now, I think more and more countries want their data a bit more closer to home.
But with the understanding that for West Africa, every country in West Africa can't build their own infrastructure.
But if we can develop the policy and framework that would allow us to take advantage of shared infrastructure, we believe will put us way ahead of our development goal.
This is really one of the things that we're pushing this Data embassy framework to develop because we believe it also would enable us to attract investment to develop infrastructure within the region that could be for the benefit of all of us.
So believe as maybe a final piece towards it.
It also by the way, fits within the open source framework for open source infrastructure.
So maybe that should be another part of the conversations around open source infrastructure civilian.
We were able to launch one of the first five G open access network save Leon.
So we know it's possible even within private sector, the idea to make it open within competition, but still it allows you the opportunity and to be able to leverage that.
So big push towards that.
We're keen in working with our partners and other regions.
We're keen to learn lessons and from what Morocco is doing, from what Jamaica is doing, what other regions are doing to really see how we can take on board those lessons, but also look at how we can collaborate for the benefit of all of us.
I'm really looking forward to it.
Thank you.
Thank you so much.
I think that Minister needs to leave very soon, yeah? Um, we want you to be on the picture, meaning this is not very often that we have app, so we want you to be in the picture.
I just have, if you have 1 minute for giving us the keynote so we can pick that very short 1 minute.
I would like to emphasize two things.
The first one is something we didn't talk about is governance.
Governance is crucial for AI, but also for open source AI and sometimes in many domains, open data and exchange of data may clash with the protection of private data and personal data.
This is why in Morocco, we propose some law not yet adopted by the parliament, but on the way to make possible exchange of data while preserving the protection of personal data for citizens.
The second point I would like to emphasize is about this idea that using open source, we can maybe improve the explainability of AI algorithms.
Open source can be audited and can allow us to open the black boxes of AI algorithms.
This is something very good but very hard to do because it requires very qualified talents to be able to understand this open source because opening source is not enough.
People should be able to really understand to understand what's going on.
Maybe the last thing I would like to say is that in Morocco, we play a very huge attention to the talents to the training of talents.
We also have a very big summer camp last week with 1075 people building solution on AI for ten sectors, education, health, agriculture, governance, et cetera.
I was really very impressed by the quality of the solutions proposed by the young people 20-25 mostly.
I think the challenge was to say, what's going on after 70 years of AI? What is the new? What's new? What can we do with this AI coming onto the table? The topic was to ask these young people to provide new vision of AI 70 years after the foundation of AI discipline.
It was really fantastic and I will be very happy to share with you the outcomes of this experience, one week in the desert of Morocco.
Thank you.
Thank you so much.
Next time, invite me, please.
I'll be happy to visit the desert tomorrow.
Your Excellencies, thank you so much for joining the panel.
That was the first in kind, very nice.
You show us the way how we should do it and hope that we see you next year, giving us more insights of how you have successfully built this strategy.
Big applause for all the ministers.
Thank you so much.
Let let me just say one thing before we leave.
We have to acknowledge the adaptability and readiness of our moderator.
Thank you so much.
Thank you.
Thank you very much.
Whilst the picture is being taken, we keep hearing that some of you cannot hear.
I feel like there's a joke in there, but it doesn't come to my mind right now.
If you are having issues hearing or if you just want to be closer to the stage, we have some empty seats around here, especially the ones in the red seats, feel free to come closer and also use this as an opportunity to maybe get up, just stretch a little bit, and Okay.
This is the last session before lunch and I feel like that's the infamous thing to say to lose people's attention, apologies, but I came up with a joke to keep your attention here.
It's time to talk about the elephant in the room, which I'm sure you've all seen.
It's the robot in the room, and we will be moving on to the session on open robots.
Please, speakers, do come to the stage.
It's a bit snug there, but you'll have to get close.
Let me introduce you to the moderator of the next session.
I see a lot of you moving down, so it seems to be working.
Yeah.
How are you? Yeah, the French at you? This session, it's.
I see there's a different name here.
Do you want to Let's turn them off.
Yeah, yeah.
We'll turn them off.
Matt.
Can you turn off the name? Can we turn off the names for the speakers because they're not correct.
We are seeing a lot of the people onstage and we see people moving closer.
That's great.
It's lovely to see.
It's lovely to see.
Hello.
It looks like my microphone died.
I think that's definitely a sign I've spoken too much.
So let's hand it over to the panel.
Let me introduce you to the moderator who will be moderating the session.
We have Lahari Chri from Amazon Web Services, who will be introducing the session and introducing the panelists.
Over to you.
Good afternoon, everyone.
Thank you all for being here today.
My name is Lahiri Chauduy.
I'm an open source TPM focusing on artificial intelligence and machine learning at Amazon Web Services.
Today on the panel, we'll be talking about open robots.
To get started, open source reshaped how we built software.
It reshaped Cloud infrastructure and today it is reshaping how we build artificial intelligence.
We are Okay.
We're asking whether the same can happen for robotics and what that actually requires.
Because the robotic ecosystem is still deeply fragmented with proprietary stacks, closed hardware, incompatible middleware.
Building a robotic system today still means rebuilding things that dozens of other teams have already solved behind closed doors and that's very expensive and it also limits who gets to participate globally.
And this matters more now than it did five years ago.
Physical AI is moving from research to reality very fast, and we are seeing how robots are entering manufacturing, healthcare, agriculture, and many other sectors.
The field is scaling and if the underlying infrastructure stays closed, and while that happens, the gap between who can build and who cannot will only further widen.
We have four panelists with us here today who are building this future from very different vantage points.
Over the next hour, we'll be going through the robotics value chain, and I'll let the panelists introduce themselves briefly.
Hi everyone.
I'm Matthew Mesler, co founder and Chief Executive Officer of Wonder Craft.
Wonder Craft is building in this space.
We started building exoskeletons, which are robots to make people work again.
Very recently repurposed that technology to build humanoid robots, the ones that you see here, and this one.
We do legged robots and have recently entered into the industrial space through a partnership with Rondo Group and helping in manufacturing tasks.
Hello, everyone.
My name is Christine from ASquare Robotics.
For those who don't know us, we are a AI native general purpose robot company and from Shanzin China, we build productivity oriented general purpose robots.
From day one, we're building robots with a brain, so we're building and invading the technology and embody foundation model, and we're also designing the most stable and reliable hardware platforms for real world deployments, and we're finding use cases to have our robots get to work.
So in fact, we are very happy to be able to bring our robots here.
It's called the alphabots.
So this is the generation of product that we have already achieved mass production and also real world deployments in manufacturing services and public service.
So if you have a chance to come visit China or come to Xinjin, you will find our alphabots working at some of the major parks and shopping malls as a retail service serving coffee and ice cream on average of 8 hours a day, making 100 cups of coffee and ice cream without errors.
Um, so my perspective today comes from bridging AI innovation to real world deployments.
As part of our commitment to making intelligent robots for every industry and community around the world, we have launched the Alpha brain platform, which is an open source community for embodied intelligence that help researchers and developers to connect models, data, training, and deployment all at once.
So we see in firsthand how hardware is rapidly advancing and intelligent is still the bottleneck and our vision for open sourcing our alphabrain is very simple, empowerment.
By providing a shared foundation, we want to lower barriers to overcome ecosystem fragmentation and to enable more developers to participate and innovate in embody AI.
We're very excited to join this discussion and to talk about how we can build a more accessible robotic ecosystem together.
Thanks.
Yeah, I'm Gary Bratsk.
I'm an entrepreneur and basically someone who just gets things going.
I did OpenCV, which is the open source computer vision library.
It's currently getting about half 1 billion downloads a year.
It's one of the major pieces of open source AI software.
But I took part in winning the robotic race, the Darpergrn Challenge, Stanley, that team became Waymo.
Recently, I did Space NG, which helped with the first and so far only successful commercial lunar landing that was Blue Ghost last year.
I was with Bonsai Robotics.
The AI, I currently switched to advisor there.
We produce agricultural robots, both automating large tree shaking robots and smaller ones for specialty crops and for mining.
But now I'm back at Stanford University.
Um, at their robotic research center, looking at the problem of home robotics and how to get robots for aging support and other things in the home, really rethinking how that's going to happen and what it would look like, starting for security and entertainment points of view.
Hello, everyone.
My name is En.
I'm the vice president and Head of Global Operations for Robot tool and Robot X at n in Robot Valley.
Where we are building a global collaborative and open source community for AI hardware and robotics in China.
Our goal is to integrate robust supply chain capabilities with open source deployment standards, providing a platform for global teams to efficiently build and scale using China's manufacturing ecosystem.
We warmly welcome everyone to visit our community in n in.
Thank you.
Okay.
Thank you.
Let's start with the stack then.
Where is openness proving most effective today across the robotic stack? Where does adoption remain limited? Is open source in robotics primarily driven by software or are we starting to see meaningful openness in hardware and manufacturing as well? Go ahead, Gary.
Robotics, there's been a lot of advance with the AI models typically using VLMs, vision language models, and also like on Lacoon earlier, he's pushing the pA models.
Those are having some impact.
But the software is not quite there yet, but it does open up this possibility for Again, somewhat like on the flip side of what Jan said, AI is a universal mediator, so it should be done universally.
Right now, hardware is done by people have the money or the infrastructure for doing the manufacture.
But if we had truly capable open source models that are say capable of getting a model and actuating anything, then hardware can start to be bespoke or produced for certain situations.
For instance, even to the level of your particular kitchen, there could be a robot that fits it best and if the software is there that can actuate it, then we could have a different industry where you get some motors or whatever that actuate and then three D print parts and you'd have a robot that's meant for that environment or or more appropriate for that society.
It also solves problems of how do you maintain these systems complex like humanoid robots will need a lot of maintaining and how do you fix things in the field? Well, if you can produce most of the parts near the source, then you can fix and maintain.
Yeah, I agree with Gary.
I think we already have a lot of open resources that we get access to and it is genuinely lowering the barrier to get started on robotics.
But the challenge is really the problem is about fragmentation.
We're building this type of technology with different data formats, different model architecture, different simulation environments, and different benchmark.
Even if things were open, it's incredibly difficult to reproduce someone's results and compare the system fairly and also take from a research paper and deploying into the real world.
That's why as our company, when we think about open source, our goal is to connect the dots to bring that data model, training, and evaluation and deployment all in one coherent framework, and we believe that open source should not just open up individual components, but it should give the whole community a share foundation to build on.
Yeah.
In our view, the most important word in robotics is deployment.
You're talking about impact.
The only way to have impact in the real world is to deploy actual robots.
If you look at where open source has been historically successful, you have software like Open CV, which, as Gary said is incredible.
I think it's been widely used in academics as well.
You have robot blueprints that can be built in every university, every college and it's doing well.
I think where it's not being able to crack anything yet is in real world or industrial deployment.
When you think about why When an industrial player is deploying a robot, they are looking for a couple of things up time, how reliable the system is throughput or cycle time, how effective it is to get the job done, and then of course, ROI price.
The thing is, I think from their perspective, they don't want software or hardware.
They want a system that someone can stand behind in terms of responsibility.
And because at the end of the day, if and when the robot fails, as was said, they are going to have a problem and they need to have someone that can be accountable for that.
I think that's where open source needs to progress or find its ways.
Maybe to think in terms of systems rather than a single brick.
Otherwise, it's going to remain limited in its adoption in the real world.
Okay.
Building on Matthew's point, many robotics innovations succeed in labs, but they struggle in real world environments.
What are the main barriers to deployment at scale that we are seeing today? Yeah, I think reliability is one of the key points.
Up time cycle time, and these machines are either very simple and then their deployment is contained to simple tasks, or they are very complex and complicated, and then we are not yet at the level where we can deploy them in the general context.
For example, in our view, Robots will not scale because they are general, they will become general because they scale, meaning that we take the approach of deploying robots on well defined use cases like moving boxes or moving tires or things that are still very difficult to deploy reliably but manageable with today's technology.
As we get more robots out there in the real world, we gather more data that can improve our models and extend the variability of the environment where they're able to perform.
Yeah.
Building on Matthew's points, the barriers between labs and the real world.
In labs, everything is controlled, but in the real world, everything is changing all the time.
You have the lighting, you have the object, the workflow, people working around it.
These are all the uncontrolled elements and the bar for reliability, stability, and safety is much higher in real world settings and also in productivity demanding scenarios.
The robots can't just work once, it has to work every single time consistently.
So we believe the key is to build a close loop between real world deployment and model iteration.
Stimulation helps, but we do need real tasks, real environments, and real users.
And at AIR Robotics, we treat commercial deployment as part of the learning process.
So the robots generate operational data, the data helps improve our model, and the better model also helps with the next deployment to be more robust.
So we believe that real world produce real intelligence.
We think there's a couple of reasons why there's a barrier.
One is because the factory environment is flawless, but the real world is unpredictable.
It's like what Matthew was saying.
Two is because the time and scale is very different.
Three is because the economics of the value chain is different.
We believe in the cluster hub of the supply chain is very important.
I just wanted to add the financial support and I don't know how you would do this, but obviously, I do open CV and it's probably saved billions of dollars in software development and yet we get approximately zero back for doing that and yet have to run this organization.
I wonder if at some level of scale of an industry, so they're successful, they're making millions, that they pay some open source tax and that's paid out on the radio song play model.
If a certain open source is used a lot, it gets a proportion of that money and maybe some is also set aside for new innovation.
Thank you.
The deployment picture today assumes current robots, but AI is changing what they do.
How is the integration of AI in terms of perception, reasoning, and autonomy changing what robots can do in real world environments? It's definitely changing a lot, especially in terms of flexibility and the number of tasks that you can consider.
Now that you have vision, perception, and models to be able to analyze the environment, it's actually credible to deploy robots that are going to evolve with their environment, meaning that in manufacturing operations, you use to deploy and fixed arm robots, like six robot that could do a single task, and it would take six months to deploy those robots, and it would take a lot of engineers and PhDs to configure them.
Once they were deployed, they would do one single thing again and again and again.
Now with mobile robots, you can imagine a world where you deploy robots way faster because you don't need the same level of expert operator training and whenever the manufacturing floor or the environment changes, you can repurpose or re target or re task the robot to this new environment and these new constraints.
It could be the parts or the piece that the robot is interacting with that does change like suddenly it's not a motor, it's a gear it's not that panel, it's this part.
It used to be very complex to adapt robots to new requirements and it's going to be increasingly easier.
Now we shouldn't sell dreams.
I think there's still a lot of progress to be made and especially in terms of reliability.
This is why we have this idea that robots will become general because they scale, meaning that they will increasingly address more complex environment.
Yeah.
I'll say a lot of traditional robotics, they're very sophisticated instruction followers, give them a very clear control environment with a fixed set of task, they would do it perfectly.
But the moment when something changes, it fails to adapt and that's what is changing with the rise of foundation model.
Instead of programming the robots with a long list of rules, you're giving it ability to perceive the physical world, reasons and act in a more flexible way.
Take for example, when the last time you bought a new coffee machine, you're not going to go straight to the user menu and read through line to line.
You're probably going to play around with it, look at it, recognize some familiar patterns and figure it out along the way.
That's like knowledge transfer and the robots are starting to develop these capabilities.
And our vision is to build robot intelligence that works similar to human brain does not by adding more rules, but have developed deeper understanding, and that is what will move robots from pure automation to genuine adaptability, and that's also what make it possible to deploy these robots faster and across far more industry and use cases than ever before.
Yeah.
Obviously, the world or most places in the world are already deeply into aging and not replacing the population.
There's an obvious need for the potential robot labor.
At Stanford, one of the things I'm looking at is aging.
As one of the example robots, of how to do this, how the robots help out an aging population, for example, is one area of deployment is the Japanese toilet.
It's a robot.
It's a medical device that's not very attractive.
It's for people who lost the ability to wipe themselves.
But instead, it's sold as a high luxury good that people aspire for rather than this sickly medical device.
I use that as an inspiration How do you get robots in the home? Well, you want them as something not as a sad robotic dog that pretends to be your friend.
But for example, if we can get them in the house in a safe way and I'm looking at ambient environments that make the robot brain or world model the entire space to make it more safe.
But in that, the robots would obviously have some utility like security system function.
But such robots could then enable embodied entertainment like if you knew the movie, the show West World.
There's robots that have an intelligence and play a role, but that could be brought and generalized instead of some pretend companion, their character in a story that you're involved in when you're being entertained, but they're also that same thing when you want to be educated or cognitively helped, how to cook, how to whatever.
This is a friendly way and a useful way of robots entering you know, the home environment that starts to mediate the aging society.
Of course, you know, this can be carried across to factories and whatever, but this is just the focus now.
They can start helping with healthcare.
There are things that aren't scaling well in any society, education, because it's an old school model of some dictatorial lecturer speaking at people, which doesn't really suit people.
Like a home entertainment robot can do using AI bespoke education for where you're at, finding what your gaps are and teaching you in an individualized way.
It can also be for health monitoring your questions, but also for exercise.
Exercise as a game rather than as a task.
I don't have too much to add, but I think in general, it allows the AI to leave the safe lab environment and entering the wild world.
At the same time allow more people to be involved in the interaction with AI.
We're happy to see this happens.
Yeah.
I think there's also something that's interesting too because we all talk about LLMs and foundation models and how this can enhance robotics.
And I just like everyone in the room to think, I'm sure everyone in the room does use LLMs in shape or form and interact with them on a daily basis.
If you think about your experience, think how often the answer that you got was completely missing the mark, doing something a little bit crazy, saying something was the wrong color or completely wrong.
This happens very infrequently, maybe 1% of the time or even less.
And it's okay because there is a human in the loop that's able to mediate the answer and to understand when to just disregard the answer.
Now, in robotics, it may not be exactly the same because if there's a robot that falls one times every 100 times, this is not something that you can disregard, this can actually be catastrophic, or if one out of 100 times it empties your fridge instead of replenishing it, that's going to be catastrophic as well for adoption.
We also have to think in terms of impact in the real world of physical AI versus LLMs and frontier models.
I think that's something that sometimes is overlooked.
Thank you.
That's really a great context to understand what the technology can do right now.
But what does it take to build a competitive robotics ecosystem at the national or regional level today? Can open robotics meaningfully enable participation from emerging economies or do structural barriers such as cost and infrastructure remain too high? It just reminds me of one of the things for scaling robotics and one of the missing parts is simulation.
Now, many people have simulators and Cosmos is recent by Nvidia.
They're still hard to use and resource intensive as far as the compute, but it is an enabler and for scaling things.
For instance, the auto industry, there's now a lot of autonomous driving.
They have a Tesla driving and many other cars spreading around the world.
But One would hope and this would take a government action, that any accident of any car that's autonomous should be reported into a national database where that becomes simulated, and then that becomes the software requirements that before you deploy new autonomous software, it should pass all these accidents that have ever happened to anyone's car anywhere, and then we have a rising level of safety, but the same is true in factories and homes, that if we have these open general databases as a level that you must pass, then we get increasing capability and safety.
So our perspective regarding this question is there are two things very, very important.
One is high density supply chain is essential.
Secondly, is collaboration with global talent is really important.
So our core standpoint is from our teams practice in Sangjin Loba Valley.
I want to talk a little bit about Sangjin Rob Valley.
The unique value of Sangjin Rob Valley is which it owns over 200 robotic enterprises and 3 kilometers for closed component supply chain.
So this solves the universal global pinpoint of inefficient, high cost hardware development for tech teams worldwide.
Yeah, I also wanted to add to answer that question, I think open source alone is not enough.
You can release the most powerful model in the world, but if it requires a massive compute cluster and team of specialized engineers, that in practice is still only accessible to a small number of well resourced organization.
Um, so for a lot of researcher startups university, especially in developing economies, the barrier is not talents.
I think we find brilliant people everywhere, and the barrier is access to access to compute, to infrastructures, to tools that let you actually do something with the technology, and it is not just about making the code available, but it is genuinely lowering the bar so that a team anywhere in the world working on problems that matters to their own communities can actually get started.
I think to build an ecosystem, you first need clients.
That's the first thing that's going to drive progress and innovation, people willing to get one of their problems solved, basically.
Then the question is what are these problems? You need to identify use cases where robotics today, not in ten years, but today can already bring value and you can deploy robots.
After that, of course, you need academics, you need, great teams, you need talent and compute.
And you need to have all of these together for ecosystems to meaningfully take shape.
I just wanted to say that I love the idea of having a shared database of accidents and issues because it's true that once you have that, you have automatically safety levels that rise that's structural.
I think it comes with a lot of complexities in terms of simulation because you may be able to simulate a lot of software issues and problems.
But I think it's going to be way harder to recreate the hardware conditions that led to these possible accidents.
I'd love to understand how you think about that in terms of how to create something that's representative or helpful for all manufacturers.
Well, one model for this is something like Carla is an open source driving simulator that was developed outside of NVIDIA, but now they support it.
One could just report the accidents.
Again, it takes a model, the cab model or whatever of the car, and then it simulates the physics to some level of fidelity and that That varies depending on the situation.
But, you could build this in all the areas that you want to do this in.
If it's some manufacturing robot, you provide a model and you provide whatever the sensing was and set up the situation that led to the accident or whatever.
This would also potentially be a repository of just data for other people to train and build their own systems or fine tune them.
One barrier that keeps surfacing is fragmentation, and that is that platforms do not talk to each other well, how much of that fragmentation across platforms and standards is slowing the robotic sector today? I I mean, in one sense, you have fragmentation, which is natural.
Different companies have their proprietary systems.
The one thing you do have though is data and if there is a way of collecting and sharing data, then other people, at least it theoretically prevents lockout that someone owns everything.
If there's enough data that someone can recreate and retrain a model to become competitive, then you never have that lock in.
I don't know if fragmentation is such a big issue because in every industry, you have a number of actors and they all compete in a way this drives innovation.
If you look at cars, all cars are different and proprietary and it's never prevented anyone from driving a number of cars.
I think what cars have is a shared infrastructure.
When we think about human robotics, we think of infrastructure because we believe it's going to be relevant to all industries and ultimately healthcare and then home.
I think the question is more, how do you have a shared infrastructure that allow all these fragmented players to compete or be used rather than for conglomerates monopolies that have never been good for innovation.
I just want to add from the perspective of the operation cost, the idea of high cost integration tag without deployment standards is high.
The cost is over the hardware itself, so we should be able to pay attention to this.
Okay.
What role could open standards play in unlocking interoperability across systems and markets? I think that's an interesting point because when we look at robots, this is not the same thing as LLMs again, and we've all seen what happened with the enthropic issue a couple of days ago, a week ago.
It's not such a big deal because in a way, everyone was able to reconnect their systems to another model, and that's the beauty of API.
It's not the same thing with physical systems because once you have 1 million robots deployed in all the infrastructure of the country or region, and suddenly someone has a button that they can push and that freezes all these robots.
It's not as easy as just writing another line of code and saying, Okay, now we're using another IPI.
No, the robots are there and they cannot be changed that easily.
Now with open source standards and interoperability, this may be a little bit easier, but the problem.
There will always remain a physical element that is going to be hard to overcome.
Yeah.
Today, the real interoperability, the challenge is the AI layer.
We talk about whether different models can work with the same data and can you take stimulation environments built by one team and use it to test a model that is built by another and can you actually compare results across different system in a meaningful way? Right now, the answer to most of that question is no, so that's a serious problem and I There's no common standard at the level.
The ecosystem is still fragmented, but we'd love to see how open standard can create a genuine share foundation, not just so things that can technically be connected, but also the whole community can move forward together more efficiently and transparently.
I want to probably abuse the common standards thing here to think about labor and it's just an idea I want to push since we're in the UN.
There's this idea of what happens when robots are ubiquitous? Are they going to cause the lack of work and whatever and and what's suggested for this is like a standard universal basic income, which I think is the wrong thing to do because it turns people into beggars.
And rather, I press the notion of universal ownership.
So let's say government or governments tax a corporation on a share of their ownership.
This is mixed into a mutual fund and distributed to the population.
So this is an open standard for a better way of doing income that converts the population into owners, not beggars.
Coming to our last question, in which areas may open source approaches face limitations in robotics and what role should governments and public sectors play in supporting safe and scalable open robotic systems? Again, just to abuse things for fun.
I'm more for replacing governments.
I'm not very happy with any government on Earth.
I think they're all underperforming or severely underperforming.
What if the AI and the robotics, you could swap out at least part of your taxes and whatever for the government of your choice, you would be under their rules giving money and they'd be delivering services and now we'd have a huge competition for who's going to deliver the best government.
But it's a different topic, sorry.
I think the most valuable role government can play is as a ecosystem builder.
There are things that are genuinely hard for any single company to build on its own.
Say for safety standards, data governance framework, share testing environments, long term fund fundamental research support.
These are public good and historically, the technological ecosystem has had the most lasting impact the ones where governments step in and build that shared foundation early.
So the goal should be healthy, open ecosystem where innovation can happen broadly, um, Yeah.
I think the number one thing that drives companies are their clients.
Now governments are one of the largest entities in the world, and so By all logic, they should be the one with the largest problems.
Actually, that's what I hear on my right.
The number one thing that they can do is to become clients of robotics today so that they can shape where those companies are going and what kind of products they develop because at the end of the day, that's what companies do.
They listen to their clients.
Hopefully, they should and governments have the ability to become large clients of those companies.
I think there's two things government should do.
One is encouraging the regional manufacturing density.
Two is invest in foundational open source infrastructure, both software and hardware standards.
That's what we see in Sangjin Nuba Valley.
Thank you.
Okay.
Do we have any questions from the audience? Okay.
Hello.
Maybe I can use this microphone.
My name is Nico Caballero.
I'm the G chair at CN.
Sorry.
I have a question for the distinguished lady from China.
You referred to fragmentation as a somehow negative thing.
My question is, wouldn't it be actually beneficial at the end of the day in terms of having different approaches to whatever problem there is on the table? Okay.
Yes.
When I talk about fragmentation, I was talking more about building some basic layers and tools for researchers and developers to get their hands started in the same starting point, less than having, I think what you're referring to is having many different players in the market, different types of products, which I think is amazing.
But what we're seeing now with embody AI is that a lot of building these embody AI foundation model is not about coming up with a clever algorithm, but it is about a uh, freeing the engineers from conterior to building the infrastructure, providing all the baseline tools for them to get started with this technology and making it more accessible for them.
This is what I was referring to with the fragmentation, but think having different players and coming at robotics and by AI from different angle is very healthy for the market and for different use cases and application.
This is what we also wanted to build with this ecosystem is to empower different players to come in.
Anybody else? Hello.
My name is Eve Neon from the open source Hardware Association, and to the panel, I wonder, there was a lot of talk about scaling robots, using them in multiple situations such as in the home, in industry, What do you think the level of technical literacy is going to need to be such that we can actually deploy robots in not just research labs and things of that nature, but in actual everyday situations such as being in the home, maybe working at a mechanic shop, something like that? How do you see open source the open sourceness of these robots of their designs, things of that nature, either aiding or complicating that ability to reach a necessary level of technical literacy? I guess there's the technical literacy for building and developing these systems and then for using it, and as you see, most software is pretty bad at being a usable.
I can't remember.
I got the Tesla and it has a vent icon and it's gray.
Does that mean the vent is closed or is it open and when you press it, it turns white.
Does that mean it's now open or did I just close the vent? That's just an example of stupid API design.
With robots, I assume you're going to have the same problems and if we have more open models, maybe more people can feed into just better design, more intuitive.
Obviously, if they can demonstrate their use or one of the big advantages of LLMs is simply using natural language to explain things rather than this cumbersome interface.
I could just say, I want the vents open.
It's easier and that's how we evolve to describe things.
Maybe with the language model that becomes the universal interface.
We'll take one last question.
Hi.
My name is Chair Mark Femi Adena.
I'm here.
I'm a PhD candidate at some Houston State University in Texas.
My research focuses on AI assisted crime scene investigations.
And my question is, do you envision robots being deployed to assist law enforcement investigators, to document crime scenes, to locate evidence, to make investigation easier for them, and also what are the policy barriers or the technical barriers you see that will make this a reality? Thank you.
I think that's probably one of the last thing that will happen because we will talk about robots that are enabled or authorized to exert force and violence on other human beings, otherwise they wouldn't be particularly helpful.
I think that's probably the last step if it happens at all.
I knew that there was a few years ago, there was an experiment by the NYPD with the Boston Dynamics dog.
I think they withdrew the experiment because there was a safe world for the dog or something that's bad guys could just used to stop the dog because, of course, the robot needs a safe world.
You run into all kinds of problems when you do that.
In our view, robots will first be deployed to help people rather than to exert violence on them and there are a lot of topics and subjects where we all need help for physical help in healthcare or other kind of help.
That's I think where we should focus as the robotics community.
I just wanted to I worked early at Willow Garage where we were developing early robots, and I think robots are going to help the crime, not the law enforcement because we once locked ourselves out of the building and we just telep the robot from the inside to open the door.
That just defeated the whole building security system.
But I I would imagine the first use would be in violent situations that a robot can go in and risk its life instead of a human's life in helping prevent things.
But I imagine like anything else humans do, it's a two edged sword and there'll be rich new areas for criminals to explore.
We are all at time now and it's lunch break.
Let me pull some key points together from the panel.
What we're all taking away from this conversation is that robotics value chain is still fragmented at different levels, but that does not limit us because we also have open source and we have heard the panelists speak about how open source can actually help us to build open robots.
But There are some key points that were also highlighted by the panelists, how we require governments to build national databases that could record AI accidents and then that could become a potential ecosystem for the builders and also be an ecosystem builder at the same time and also support the regional manufacturers and support the open source hardware and software ecosystems.
Work that is being done not only by the panelists and everybody in the room matters a lot right now because robotics will reshape labor, healthcare, mobility, and economic opportunity within the next decade.
Whether the transformation is proudly shaded or narrowly captured depends totally in part on the choices we make about openness now.
Thank you so much and thank you to all our panelists for being here.
Thank you.
Thank Thank you so much to the speakers, the moderator, and all of the speakers have moderators from this morning.
I know I'm your least favorite person in the room because it's only me between you and lunch, but we have a slide that we must show you and please take a photo of it.
It will be giving you information about the breakout sessions that are taking place after lunch.
Since we give you 2 hours for lunch, we assume that this information will be lost in the meanwhile.
I see a lot of cameras out.
That's great.
Just for information, if you're looking for the conference rooms, those are in the basement.
You go to one B.
Okay.
And we have the volunteers who are around in the yellow shirts and they can guide you if you're lost.
I've been here for less than a year and I'm still lost every day, so you're not the only ones.
Then there's the Ecos chamber, and that's the room we're in right now.
So if you're attending that session, you can come back here.
And as I mentioned in the morning, please take your things with you.
Otherwise, we're going to be in trouble with security.
So take your bags, take your belongings, go for lunch.
You have 2 hours, so you can obviously go outside, but we have a cafeteria on the fourth floor.
And then there's also food in the basement in the cafeteria.
So sorry, that's the cafe.
Cafeteria fourth floor, cafe, first floor.
All right.
You're free to go.
See you back here at 4:15.
Thank you.