How does a small team in the UAE compete with the world’s biggest AI labs? By building open-source models that rival the best—and doing it with purpose. In this special episode of Below the Fold, recorded live at the Dubai AI Festival, Dr. Hakim Hacid, Chief Researcher at the Technology Innovation Institute (TII), shares how his team is shaping the future of Arabic AI, open-source LLMs, and robotics—right from Abu Dhabi. From the development of the Falcon LLM series to thoughts on ethical AI, agent autonomy, and AI for robotics, Dr. Hakim breaks down what it means to build globally competitive AI in the MENA region. 🎙️ Hosted by: Will Snow
🎧 Topics include:
In collaboration with @BelowTheFoldPodcast, @DATEwithTECH, @launchfoundryai, @DubaiAIFestival, and @DubaiAICampus
[00:00:00]
Hi, my name's Will Snow, and today we're coming live from the Dubai AI Festival with date with Tech in association with Recon Below the Fold and Launch Foundry. Today we have our very prestigious guest, Dr. Hakim Hacid, chief researcher of the Technology and Innovation Institution. Welcome. Thank you very much for the invitation.
No problem. It's so good to have you here. Thank you very much. So, from what I understand and what I've been told, you look after pretty much everything, ai,~ um,~ and you lead one of the most ambitious AI research centers in the region for people that are not familiar with your business, what makes you work in large language modules and.
AI distinctive. Thank you very much for the invitation. ~Uh, ~glad to be with you,~ uh,~ today. So indeed. So, I'm part of the TAI Technology [00:01:00] Innovation Institute that is based in Abu Dhabi. ~Uh, ~TAI is research center that is composed of,~ uh,~ nine research centers from different,~ uh,~ in different domains. Including robotics, security directed energy, quantum computing, cryptography,~ uh,~ to mention just, just a few.
Wow. ~Uh, ~so the research center, I'm,~ uh,~ I'm, I'm handling is the AI Research Center and indeed we work a lot on the,~ uh, uh, ~generative ai. So what makes the work we are doing,~ uh,~ special I would say is first of all, so we are very small team compared to the. Big teams that are all, all around the world. ~Uh, ~I think we are extremely competitive in terms of,~ uh,~ performance.
~Uh, ~when we take our models that we, that we, we provide as open source,~ uh,~ we are extremely good,~ uh,~ in performance. ~Uh, ~we stay competitive with all the models that are there, including the,~ uh,~ cloud source. And, and, and when we look at the, let's say the, ~uh. ~The place from where these things are coming. So it's basically a small country that is producing these,~ uh,~ exceptional outcomes to [00:02:00] compete with the very big players that, that, that were there for,~ uh,~ decades and decades for sure.
That's amazing. It's, it's such a nice time to be in ai, right? Like Absolutely. Things are really flying. ~Um, ~you mentioned these LLM models. Tell us a little bit more about those. Those are obviously homegrown, they're accessible to everybody. Absolutely. So we, we have, ~uh. ~Two lms, I would say to two families of lms.
We have an Arabic model that is,~ uh,~ basically the no,~ uh,~ model. That was the first model we have released and we kept growing it,~ uh, uh, ~in house. And then we have the one that we release, and that's the Falcon Large Language Model. ~Uh, ~this one is an open source,~ uh,~ open weight,~ uh,~ large language model that is accessible for everyone.
We have like three families of that Fcon one, FCON two, and Fcon three. Currently we are at the Fcon three that we have released. Initial version back in,~ uh,~ mid of December, 2024. Okay. ~Uh, ~yeah, so we still continue working on that. ~Uh, ~and,~ uh,~ things are, are growing and we may have few news in the coming couple of weeks.
Ah, very exciting. I love it when there's a bit of a sneak preview [00:03:00] into those things. Maybe we can hear from,~ um,~ and for all the entrepreneurs out there in terms of getting access to these, ~uh. ~So we have two ways to access the models. We provide the model so people can download them and run them on their infrastructure.
That's, that's the initial goal that we had. It's basically to allow everybody to have control of their data. So they have the model running on their infrastructure, on premise. The data doesn't travel in the network, and then you get, you get the data host. The second thing is we provide an API and an interface for people to play with the model.
Or to connect their applications to the,~ uh,~ to, to the API. Amazing. So useful to have these out there, especially when you go the open source route. Absolutely. ~Uh, ~these things also generally grow faster, right? And they're more accessible. Absolutely. Yeah. Fantastic. And, and things are, are evolving extremely fast.
So we get new generations as we, as we move forward. So we keep, we keep things,~ uh,~ growing, upgrading all the models, and we make sure that these models stay available for the community. Do these [00:04:00] LMS specialize in any specific industry? We are building,~ uh,~ general purpose lms,~ uh,~ but then we work on,~ uh,~ fine tuning these LMS for specific domains, the legal, the healthcare, so on and so forth.
But the one that we open source is the generic model,~ uh,~ so everyone can fine tune it again for their specific purposes. Amazing. Very, very useful. Yeah. I'm sure our listeners will be so glad to find out about this information and, and find it online. Okay. ~Um, ~and then let, let's move on. Sure. ~Uh, ~coming from Bell Labs, Zane, and the academia side of the world, how did those experience shape the way you now lead r and d?
Yeah, I think,~ uh,~ you know, every experience is unique somehow, right? So we, we learn, we grow,~ uh,~ we, we improve ourselves somehow. So I think the. In each experience, there was a, a different challenge. I think,~ uh,~ bell Labs were a pure r and d ~uh, ~kind of work,~ uh,~ z and telecom. It was more,~ uh,~ sort of, ~uh, ~operational, I would say.
'cause it was customer analytics and this kind of stuff. But then it exposes you actually to understand the customer, [00:05:00] the needs. ~Uh, ~and then you have the academia where you have the, the pace that is much more sort of slow, where you need to think and you need to, so you have to combine all these sort of characteristics.
To be able to sort of e execute properly an r and d strategy where we are going extremely fast, to be honest. So, ~uh, ~you think at the, at, at the operation, like at Zane, things were fast, but currently things are not fast there, just the flying, you know? So I feel exactly the same with you. Absolutely. I think,~ um,~ every, every week just there's something new of groundbreaking.
You know, an releases 3.7 meta has a context window of a million tokens. Absolutely. ~Um, ~it just, every week there's. Feels like there's something groundbreaking. I agree. ~Um, ~and it's very difficult to keep up as a technologist. That shows also the, the, the potential of the field. Right. So we, we still, we still discover things every day, every week.
Yeah. So, ~uh, ~still we have a lot of work to do. Yeah. So refreshing, right? Yeah, absolutely. It keeps things interesting, for sure. Yeah. Right. ~Um, ~and, and then just like kind of coming back to the golf, really, how do you think the Gulf is positioned globally,~ uh,~ or the [00:06:00] UAE or Dubai specifically compared to other places like Silicon Valley or even some of the, the stuff that's coming out of China, which is so advanced.
Yeah, I think,~ uh,~ look, there is clearly a good positioning of the UAE. So now we hear about UAE everywhere. So everybody is coming to the UAE, either to, to, to invest, to participate, to contribute. I think there is clearly a message saying that we are serious about this AI and we are putting the necessary sources to make it happen.
So in my opinion, things will,~ uh,~ will come as we move forward. But, ~uh. Uh, ~the UA is becoming a serious hub for growing this ai and it is becoming extremely competitive. Yeah, it's brilliant. I've, I've kind of felt that and I've seen it, especially with all the guests we've spoken to at, at the festival this week.
Yeah. ~Um, ~so I, to be honest, I see it as for anybody that's interested in the subject, this is really a place to move to and Absolutely. And get going. Right? Absolutely. Absolutely. And, and we see that. We see those moves, right. A lot of startups are coming to the UAE. A lot of experts also are, are, are coming to the UA e we, [00:07:00] we used to have.
~Uh, ~difficult time to convince people to come and do AI in, in, in the UAE, but now they see, they see the things and we're getting it easier. It's not, it's not sorted out completely, but we're getting,~ uh,~ we're able to get more talents in a much easier way. I would say for sure. It's not, it's not just the salary that counts, but also the project and the ambition of the place.
That convinced the people also to come. Totally agree. And I, I think there's a lot of backing from private institutions as well as the government that's putting all these things together. Yeah. I think there is, there is a solid strategy behind that and there is a solid support and,~ uh,~ the project is a serious one to do something and to be,~ uh,~ let's say serious actor in, in this,~ uh,~ in this area.
And, and it's a big, it's a big one for us, right? 'cause we actually building smart cities. Yeah. It affects every dimension of our life really. Yeah, absolutely. ~Um, ~all over the, over the country really. It's, it's, it's a sort of a living lab, you know, so it's, it's a, it is a living lab lab and you can test many things and,~ uh,~ yeah, I like that.
I'm not sure how people would feel about being test subjects, but I, I like the analogy. [00:08:00] I, I don't think we would live to that point. Right. But I, I think it's a great segue into,~ um,~ you know, ethics and, and data. Yes. Like how do you see,~ um,~ really AI being implemented around the ethical debate around it and, and where's it going for us?
See if, from my perspective,~ uh,~ one thing that we tend to forget because things probably are going extremely fast, is that AI as we see today, is a new field. It's a very young field, right? It's, it's a, it's infancy. So we still need to do many things to get to the point where things are understood well. So as of today, there is just a hype around,~ uh,~ these, these models, what AI can do, so on and so forth.
But we forget that we need time to see, for example, the real deployment and implementation and how it affects the life of the people, for example. So I think we need to give it a little bit more time for the time being the things that we are having. It's, it's more mostly on the [00:09:00] hypothetical side, just that this AI will just,~ uh,~ bridge the privacy of the people.
Yes, it can be used for that if it is not used well. Right? Yeah. It depends on the use. At the end of the day. I think for the moment, there is a, there is the legislation, there are the rules, the laws in place to control the way this AI is used. Probably this legislation needs to sort of evolve in, in the future.
But I think again, and that's what I say all the time, we shouldn't be too much in hurry to add new rules without really understanding, having. A factual understanding on how this AI would affect Absolutely. The lives. No, I totally agree with you. I mean, with this stuff, there's a, there's a rush to commercialize.
~Um, ~and often the, the ethical side sometimes get ne neglected. But I do believe, like you say, the right infrastructure, policy making is in place Absolutely. ~Um, ~to keep these things sound. Yeah. Agree. Brilliant. And, and just for a, a question really, how do you use AI in your day-to-day life? How [00:10:00] is this some place that you might use some, use a form of AI that's made life easier for you?
Honestly, I think,~ uh,~ AI nowadays is,~ uh,~ almost everywhere, right? So for sure on your phone, on your car,~ uh,~ wherever you, you go to Metro Station, you have things that, that, that are linked to ai. You go to buy the groceries, I'm, you have things that are related to, to ai, just the way in the groceries. We have been doing it for, for a while, but at that time, nobody was paying attention.
Like the way, the way the supermarket is organized, the, the products are positioned and everything is, is in ai. So. For me, honestly, AI is, is, is a sort of,~ uh, uh, ~a pervasive thing that is in our life and it's, it's everywhere. We just see it. I think it's something that we see. It's,~ uh,~ the top of the pyramid, I would say.
Yeah. Is, is just these LLMs and chatbots that, that, that we see more and more,~ uh,~ you know, generating also,~ uh,~ images, videos, you know, artificial stuff for sure. But I think for me, AI is definitely everywhere. Yeah. [00:11:00] Amazing. It's such a nice way to see the world or, or interesting way or rather I, I've never thought of like looking at a supermarket 'cause I'm from a product background and user experience in CX Uhhuh.
~Um, ~and I, I know about like the signing and the aisles and where people go and obviously product placement. But you're right. This, it is, it's been in the background for a long time. Absolutely hasn't it? Absolutely. Absolutely. On every, a lot of part of our lives. Well, even when we navigate, we do search on the internet.
Just using a search engine that's all AI behind that. Sure. Absolutely. Yeah. Cool. Alright, let's move on to our next question. I've got so many questions 'cause like we, we kind of staring into this thing of adjunctive agents. That's kind of the new buzzword for 2025. Right? ~Uh, ~where do you see that taking us?
I think,~ uh,~ that, that's an initial,~ uh,~ it's an interesting,~ uh,~ paradigm, I would say, but I just want to link it to something that we had also in the past, right? So it's not something totally new. It's new because it's running actually on, on, on top of LMS and this, this is governing this, this kind of stuff.
Yeah. But we had the concept of web services in the [00:12:00] past, right? So those web services are actually just like sort of, ~uh, ~tools, agents that we're capable to. Solve specific problems, correct? Yes. What, so we were able to decompose actually complex problems into smaller ones. For each one. We assign a web service or an agent, and then we got into these issues of,~ uh,~ how to make all these things work together so that we solve the bigger problem, right?
So this is what we are doing currently and I think the nice thing,~ uh,~ at this stage is,~ uh,~ the ability to control these agents and help them support. In an automated way with lms. Right. That's an interesting part. But soon we will get into scalability issues, right? 'cause we cannot spend our time generating agents for anything we want to do.
So I think that would be the challenge that will be coming,~ uh,~ very soon. ~Uh, ~once we, we, we get all these protocols that are coordinating the agents and this kind of stuff. Now, how to go at scale. To be able to generate enough [00:13:00] agents that are capable to do the tasks that we want to do. Yeah. I think that, that, that will be an interesting challenge.
Ah, man, there's so much talk around this. I've, I've made a lot of people going into this building, these, these agents. Yeah. And I, I do agree with you. I think it's, it's gonna become a very red ocean space. There's gonna be a lot of it all around us. But you do, you are right. I think when scale comes in, we also start struggling with compute, right?
Absolutely. ~Um, ~and, and we, look, I don't even know how far we are with the quantum side of computing. I think we still have time for that. We still have time one not there yet, but,~ uh,~ wait to wait for it a bit. Yeah. Right. No. Fantastic. All right, let's,~ uh,~ let's move on. Just gimme a second. Sure. ~Um,~
cool.
Yeah, I think, look, those are the main questions I have, but,~ um,~ tell me more. When we talk agents, there's also, I've, you know, chatted with an Nvidia guys and a lot of things are happening, happening now in the RO robotics space. Absolutely. ~Uh, ~do you, can you give us insight into what's happening with robotics?
I see. Robotics I think, is the next field for,~ uh,~ like, it, it's an extremely good [00:14:00] application area for ai. So beyond the physical part of the, of the robots, of course you have to get the joints, you have to get things that also,~ uh,~ can provide the natural behavior. Physical behavior of, of the robot. You need the brain, right?
And this is where I think the,~ uh,~ the champions will be playing. So, ~uh, ~these models, first of all, I think everybody's trying to get a, a model that is capable to run on board, like on the robot itself, right? Because you have latency issues. So you don't want to take the risk to have the robot communicating with a distant server.
And then. Receiving instructions. Yes. So we want to run these things on the robot. And so we need to, we need definitely to have smaller models that are capable to have a reasoning power directly on the robots. And then you have energy issues because you are probably consuming much more energy than what, correct.
Yeah. One, one battery will not be enough for the, it comes down to power all of a sudden. ~Uh, ~absolutely. So there is a lot of work that is currently done. We [00:15:00] have, we are exploring ourselves,~ uh,~ these kind of,~ uh,~ issues we have, as I mentioned in the beginning. Robotic research center. So we're collaborating extremely,~ uh,~ tightly with them to be able to propose, I don't know if we have to call it an LLM or whatever, at least an AI model that is capable to control the robots in an autonomous way, for sure.
Jeep, there's so much in this space, like, ~um, ~absolutely. I, I don't want like an iRobot or a, you know, Terminator kind of scenario happening. I don't think, again,~ uh,~ these are all things that we, we, we need to pay attention to. We shouldn't ignore them for sure, but I think we shouldn't be also afraid to the point of,~ uh,~ stopping innovation or slowing down innovation.
Right? Yeah. It's, it's very interesting. I was having this conversation with a friend recently,~ uh,~ where we, we were chatting about where is AI heading, right? And AI is specifically very much built for humans, right? And we love to understand these things. At what point do we start giving AI the freedom [00:16:00] to kind of code itself or have its own language, which might not be for humans, but it can advance itself.
I see. ~Uh, ~for me, I got this question yesterday, right? I very similar one. So I think,~ uh,~ from my perspective, we should always keep the human in the loop. Right. So, because once we have a fully autonomous AI that can recreate itself at some point. 'cause we could, we could go further, right? So we will allow an AI to reprogram and recreate itself.
Yes. This is where we lose control. Replicate, yeah. So that, that's, that's the issue. So I don't think if it's a, if it's a wise,~ uh,~ thing to do, I think it is wise to do it, at least in the lab, see how it behaves, and then, ~uh. ~We can advise. I'm trying to be pragmatic, right? So, ~uh, ~of course it's,~ uh,~ for me, we need to keep the human in the loop because this AI at the end of the day is expected to be built for humans.
So we need to keep the human in the loop. Yeah. To be able to get [00:17:00] the, the, the, the super controller that is a human. Yeah. The irony of it all is like, we, we are replicators, right? Mankind, right? Yeah. We're an organic version of, of really what we're talking about. Absolutely. ~Um, ~and, and so I guess when, when we talk about AI and it eventually may be replicating,~ um,~ what is your thoughts on like AI consciousness?
I, that's a very, for me at least, it's a very, very high level,~ uh,~ discussion, to be honest. It's,~ uh,~ because right now we, we working with probability, right? Right. Everything now that AI gives us is its best guess. Yeah. It's not thinking, feeling. ~Um, ~you know, doing it itself, you see there is, there is such a diversity in, in these kind of, of concepts even on the human side, right?
Do we have a full understanding of these, of these things? Probably not. Probably not. That's the thing. So for me, the issue to replicate these thing in a machine, I think I, I get stuck quickly in those things, to be honest. Yeah. So, ~uh, ~and interesting topics,~ uh,~ don't get me wrong, but. [00:18:00] I mean, we, we don't have yet a very huge understanding for human behavior, for human, let's say,~ uh,~ nature.
And then we try to reproduce that. It's a bit complicated. Of course. Yeah. Again, this, the, the human intelligence, just for the intelligence, we still don't understand how these things work in the brain, right? So you co you compare our brain that is extremely small. To, I dunno, something that is running currently the big models that we have.
So there is no, there is no comparison, right? Sure. But still, probably we are more intelligent than the, the AI that we're creating today. So I think there is still a way to go from,~ uh,~ from those studies, probably other people in the future will be able to add this,~ uh, uh, ~consciousness to the, to the ai.
It's an interesting thing, but, but we need, it's fascinating advance first on the understanding human and then. ~Uh, ~formula, like formalize that to, to, to attach it to an ai. [00:19:00] Fantastic. I really love the subject. ~Uh, ~the, the, the next question I have for you is really around how do you, what's your one piece of advice for all of our listeners out there?
If there's something that they can take from you and, and you would give it to them freely just to help them in their lives, what would it be? I think just to the, the, the best thing to do nowadays and for the future. It just to stay updated, not being afraid with ai and make sure you use AI for your benefits For sure.
Amazing tool set. ~Um, ~I've added it to a lot of the things I do now,~ um,~ and it's really enabled what I do, so Absolutely. It's made me faster, quicker, more efficient. Absolutely. ~Um, ~but it's still most of the organic creativity and the, it all originates from you, has to come from you. Absolutely. I agree.
Fantastic. ~Dr. Kareem. Thanks so much, uh, for having us, for coming. Uh, that's pretty much all we have time for today, and yeah, thank you very much for the invitation. Fantastic. Thanks a lot. Thank you. That's, that's Dr. Hakeem, not sorry. We can, we can do it again. You can do it again, yeah. Oh, cool.~ Dr. Hakeem, thank you so much for being here on the, on the show.
We really appreciate your time. It's been fantastic getting to know you. It's my pleasure. Thank you very much for the invitation. Perfect. Thank you. There we go. Thank you very much.
[00:20:00]