Camille Morhardt 00:23
Hi, I’m Camille Morhardt, host of InTechnology podcast. And today, as part of the Intel Capital series, we’re going to talk about artificial intelligence and its role in gaming and game development. I’m going to co-host with Srini Ananth, who is a managing director at Intel Capital, and he focuses on deep tech, cloud infrastructure and AI applications companies. Welcome the podcast, Srini, and will you please introduce our guest?
Srini Ananth 00:50
Thanks, Camille. It’s a pleasure to be here, and it’s a pleasure to have our guest, Florin Radu. Florin is a VP of Corporate Development and partnerships at Inworld AI. And Inworld is a company that is the AI driven gaming platform for powering NPCs as well as virtual characters in a host of other verticals. It was founded by Ilya (Gelfenbeyn), Michael (Ermolenko), and Kylan (Gibbs), who are working together at Google and DeepMind and have a long history of working on conversational AI and conversational speech platforms.
Florin himself was previously working at the intersection of technology, media and telecom. He’s worked at Sony Gaming, and he’s been an investor himself, investing in startups in the music industry. He’s also been in technology, media, and telecom banking at Credit Suisse. So welcome to the podcast, Florin.
Florin Radu 01:38
Thanks so much for having me.
Camille Morhardt 01:39
Okay, so to jump right in, Srini, you said NPC very casually, and I’m just gonna say, like, a year ago–or at least two years ago–if you had done that, we would have to pause and say, “Okay, let’s define what NPC is.” I actually think that’s common in lexicon right now. And it’s Non-Playable Character. I think actually, most people know what that is, even down to the acronym. So I’m curious, like Florin, first of all, what is, in case somebody isn’t familiar? And then why do we all know what it is now?
Florin Radu 02:11
So, NPC is a non-player, or non-playable character, essentially, in a game, it’s the other players in the game that you can’t control. And I think it actually became very popular when the movie “Free Guy” came out with Ryan Reynolds, because the whole concept was, you have NPCs in this world that are kind of just doing whatever they were programmed to do. And one day they realized, almost like they have free will, and they go about, they try to escape the confines of this virtual world that they live in. So that was actually the exact same time that Inworld was started and I think we kind of tapped into a lot of this media frenzy, as well.
And then, of course, everyone’s playing games, whether it’s on their phones or on console and on computers at home. And everyone has interacted with an NPC and had these kind of interactions where the NPC runs into the same wall, like, over and over again, or responds in the same scripted dialogue over and over again. So there’s a lot of memes now on Instagram and TikTok of these NPC recreations where it’s like actual, real people acting like NPCs. So overall, it became a pretty big part of culture. And of course, that’s how we started our business, focusing on NPCs, but we’ve expanded far beyond that now with broader AI tools for game development.
Camille Morhardt 03:14
So, tell us briefly what the company does.
Florin Radu 03:17
So essentially what we built is an AI engine for games. It’s a series of AI components, tools and infrastructure that allows game developers to create content and develop games more efficiently without sacrificing quality, but at the same time create novel, new gameplay mechanics that make games more fun, more immersive, more engaging and more personalized for the player.
Camille Morhardt 03:36
Yeah. I think you said, “make games better and make better games.”
Florin Radu 03:40
Exactly. Yeah, both sides of the spectrum there, yeah.
Camille Morhardt 03:43
So how does AI–or is it Generative AI, specifically, or large language models–what are you tapping into to actually improve the mechanics and the NPCs in the game?
Florin Radu 03:53
Yeah, so it’s really a wide, wide spectrum. We definitely use LLMs and, you know, generative models themselves to create new content. But there’s also other AI and ML models that we’re using across the entire stack. What we call these things really are primitives. We have input models, which is, for example, speech recognition, taking in text, taking in state ingestion from the actual game itself, like, where are the players? What colors are the different players–in terms of their outfits or their HP levels, whatever really is in that game, state–taking that in. And then we also have output models. So for example, speech synthesis, text generation, which is more, more traditional, LLMs, which are very, popular at the moment, but also some pretty rudimentary actions, animations, really expanding what is being produced.
And then in the center between the input models and the output models, we have systems. So, to give you a sense of like, one of the easiest systems to explain is, for example, control systems, so making sure that the content going into the models and coming out from the models is safe, is in line with the IP that is being used to create these games, making sure that the brand safety is there. So those are some of the controls that we’re applying–in addition to making sure that, for example, the characters don’t go off on tangents about topics that they’re not supposed to be discussing.
Camille Morhardt 05:01
Are you collecting player data also and then training models or training future game development engines on their actions or preferences?
Florin Radu 05:10
It depends on the engagement that we have with the studio. Some studios don’t want to allow that. Some studios are like, “you know, we want the highest quality possible.” So, for us to improve the quality, we need to take in some of the interactions that are being had, figuring out how are these interactions trending, and using that information to actually fine tune those models, improve their quality, reduce costs and overall, ensure a higher and better experience for the end player.
Camille Morhardt 05:32
What is kind of your take on this centralized training and taking in a collective input and then improving these models versus, like a distributed learning kind of a model, or maybe even just customized AI at the Edge for an individual person, and then kind of retaining that on their individual device?
Florin Radu 05:52
We’re pretty flexible here. It really depends, again, on what the goals are of the game studio. I think naturally, we believe that the overall industry will move to having everything happen on device that helps decrease latency. It helps improve performance. It helps reduce cost. You know, if you’re playing on your Switch and you’re in the tunnel, you lose internet connection, you don’t want the game to stop working, or the characters to stop saying things. And that’s really the natural progression. In the near term, there will be some quality degradation, just because we can’t put these like massive models on device just yet. But over time, we really think that those problems will go away, and then games will be consumed with AI in the same way that we’re playing games now offline.
Camille Morhardt 06:29
Will it be going back and forth, like you’d be online at certain times, and then offline, downloading portions of the game ahead of time, or as it goes because, I mean, there’s a lot of multiplayer that would require online presence, right?
Florin Radu 06:44
Yeah, for sure. From our side, we can do it in a hybrid approach, for sure. There may be scenarios where, you know, you don’t want to be processing things on device, because you know that hardware is being used for other parts of the game, and there it’s easier just to tap into doing things through the cloud. But of course, again, some of the scenarios I mentioned where you’re losing internet connection, you can definitely switch over to on-device models, and maybe the quality is a bit lower for the five to 10 minutes that you have to be on device. But overall, we believe the quality will trend in a pretty similar direction where there won’t be any sort of noticeable differences between how that is actually being processed.
Camille Morhardt 07:16
So, what are you feeding into, say, a non-playable character, like training; what kind of inputs are you providing? And then, how is that able to take more real time input from somebody actually playing the game and then modifying their behavior, the character behavior, based on the person playing the game?
Florin Radu 07:34
So, all our models were designed and optimized for gaming use cases where conversation needs to be fun, engaging, immersive. And if you look about these like, you know, pretty large, traditional foundational, large language models, they were built to serve any use case imaginable, right? Whether it’s a customer call center or a student asking for help on their homework. So those conversations are pretty boring; a lot of rudimentary AI implementations now it kind of sounds like ChatGPT hooked up to a character. What we’re doing is really tapping into engaging conversation in game scenarios and other scenarios, looking at how that conversation unfolds, what’s being said and what makes those conversations engaging, and using that data to train these models to respond in a way that that is engaging, that is fun, that wants you to keep on exploring and building a relationship with these characters, or whatever the context may be. And that, you know, is very different from a lot of the traditional enterprise use cases that we’re seeing become really prevalent these days.
And then in terms of how you can actually adjust character behavior based on the inputs from the player, we’re actually able to pass through information in real time. For example, let’s say you’re a wizard in this game where you went and did this quest before interacting with the character; the character can take that information in, pass it through, and respond in ways that make sense based on how the character was programmed. An onboarding character will have very different goals and motivations than, for example, a villager whose village was just attacked by evil pirates or whatever, whatever it may be, and they should be able to respond to players in a way that makes sense for their role in that particular game. So we’re able to take in all that context and spit out behavior and dialog and actions that’s in line with the background of the character.
Camille Morhardt 09:07
One of the reasons that I think I enjoy conversations with my friends is they have a memory of a prior conversation, and they sort of, you know, check in again.
Florin Radu 09:16
Exactly. These characters have memory. They have knowledge. If you come back the next day, they’ll be able to understand what happened the day before, just like you and I would and continue that natural conversation. Or one of my–this hasn’t been realized yet–but one of my favorite use cases would be, imagine you have the first version of the game, and then you release a sequel. Imagine all those characters continue to live there in the sequel, you should be able to engage with them, and they’ll remember all the adventures you had the first time around, or what you did in that game. And that’s like, it makes a very natural transition between the first game, the second game, the third game, whatever it may be. And now in the games industry, are seeing a lot of IP, creating massive next generation games for using that same IP, including a lot of the ones that we’re working on right now.
Camille Morhardt 09:56
Can you tell us a little bit about the game mechanics that you’re enhancing or helping developers enhance?
Florin Radu 10:03
Yeah, actually, I can give like a particular customer example. We had this studio on this platform called Playroom, launch a game called Death by AI; it’s a party game, so you and a bunch of other friends are given a particular prompt or a situation. So, for example, I was playing the other day and told me, “You’re disintegrating like Marty in Back to the Future. What do you do?” So basically, everyone submits their own idea of, like, what the next step is. I submitted, I think, for something like, “oh, I drink like a regeneration potion” or whatever it was. We all submit those ideas, and the AI will basically play out the scenario for us and tell us, did we survive based on that idea? And you’ll go around and, you know, one person survived, the rest don’t. That’s a really novel game mechanic. It’s perhaps, like, pretty simple. It’s like an early implementation, but that game got, like, hundreds of millions of hours of gameplay and users. And it just shows how, like, AI, can actually bring the player into the game itself. It’s non-scripted. You’re actually deciding what happens, and it’s actually also creating, you know, a social interaction with you and your friends. So that’s like an entirely novel gameplay mechanic that wasn’t really possible before AI.
And then we have a bunch of other examples. For example, we worked on a project with Ubisoft, where you’re talking to one of these characters, and you actually have to build a relationship or rapport with them, get them to trust you, for you to unlock more information, to unlock more knowledge, more quests, and be able to actually unlock new parts of the game or explore the game. And before, that could probably be done in a way where it’s a bit, like, rudimentary, like if you ask this particular question, or you hit this buzzword, then you get this information. But this can be done in a much more natural way, very similar to how, imagine you were actually playing that game in real life, and your job was to get information, you probably wouldn’t give me that information until you felt comfortable with me and trusted me. And that similar mechanic can be actually implemented in games.
Camille Morhardt 11:40
This is to get a little bit more philosophical for a moment, if you’ll pardon it. You know, what impact do you think gaming has had on the evolution of the human mind?
Florin Radu 11:50
That’s a good question. I think especially the way gaming has evolved, it’s become a natural outlet for a lot of folks who maybe are not having the same kind of engagement and social interaction that everyone else may be having. For example, right now, you see all these online videos of folks who are meeting the person they were playing Call of Duty with for the past 10 years, for the first time in person; or they’re inviting them to their wedding. And I think gaming has really helped create these connections between people, which really gets to the core of, like, what makes us unique as humans is social engagement, social interaction, building relationships, building bonds. And that’s really what I think makes games engaging. What games taps into are those relationships between people.
So, I think I can see the argument where some, well, people are spending too much time playing games. But on the other hand, there are a lot of folks who don’t have an opportunity to have that engagement for whatever reasons or not, you know, maybe not confident enough to be able to go out and make friends, and they’re able to go home, play games, meet people, expand their horizons, and somewhat have the same interaction. So, I’m curious to get Srini’s thoughts on that, as well. But I think, from my side, it’s overall a net benefit and allows folks to build these bonds that otherwise may not be possible.
Srini Ananth 12:58
Yeah, I think what I would say is we’re starting to see gaming being used in some of the therapies as well, right. There are companies that have been using gaming in treating neurodivergent disorders. There are companies trying to investigate how it can be used in other mental health disorders. So, we’re starting to see some of the benefits of gaming when used right. And this is something we always worry about when it comes to our kids, how much time should you spend in gaming? How addictive it can get, and then, what are the sort of benefits? And I think, given the right atmosphere and given the right sort of game, it can be very helpful. At Inworld they do a great job of making sure that the content is always safe for kids to use. And so that is one of the things that we always have to worry about. How can we use AI to keep some of this non-safe content off of the gaming platforms? If we can take care of that, I do see a lot of benefits.
Camille Morhardt 13:48
You know, you guys both talked about human-to-human connection that’s made through this sort of digital platform that wouldn’t otherwise be made. But also, if NPCs are becoming more and more engaging, then essentially, people are talking with large language models or other generative AI models that are becoming, and they have their persistent memory, and they’re becoming more and more interested in you and retaining the information and customizing information for you. Are we sort of turning into the “Her” model of the movie where, like, “I don’t need friends because I have, not even other humans playing the game with me, but NPCs that are engaging in an interesting way.” I don’t want to even attempt to make a judgment call. I’m just curious, like, how do you think that is going to affect humanity or communities or people who are so deeply engaged in it?
Florin Radu 14:40
As it relates to gaming, and games, you know, NPCs, for the most part, are, I don’t want to say peripheral, but they’re side elements of the story. In some games, they’re more important than others, where you have open worlds and you have to explore, but for the most part, you’re kind of being pushed along a narrative that the game developers created, and your goal is to kind of go through that narrative. We’re using AI for the NPC use-case in particular, to make the rest of the game more realistic and more immersive. So from my side, I think there are certain use cases where, if you just go into open world and talk to world and talk to NPCs all day, I don’t see necessarily that much value in that; but in the way we’re thinking about it, it’s really about complementing the existing game narrative and making those games more compelling, more engaging, and allowing these games to feel more believable versus just having a chatbot conversation back and forth, which I think has been a use case that we’ve seen a lot of so far, because the quality bar is much lower when it’s just text being exchanged back and forth, and maybe a picture that you uploaded of the character that you’re talking to. But over time, as the quality improves, we’re going to see much more compelling use cases that actually, I think, complement our life and complement games.
Camille Morhardt 15:46
So Srini, this stuff all sounds really interesting and compelling for a gamer. Why in the world is Intel investing in it?
Srini Ananth 15:54
Intel and Intel Capital, we’ve been investing in some of the foundational technologies, such as speech and natural language processing for, I would say, probably close to 12-13 years, or even 14 years at this point. And we have also been enabling high end gaming desktops. Intel processors actually power a lot of the gaming desktops. And so, we have always been interested in the gaming industry as a whole, and we’ve been enabling the foundational technologies in that space. So, for us, it was a confluence of a couple of things. We had invested in Speak To It, and that was later renamed to API.ai; that was the first Android speech assistant. They had over 50 million users, and then later on, it became a developer platform, and after that acquisition by Google, it ended up being used by a few million developers. So here was a team that continued to work in the speech and natural language space. Along with it, they also spent a lot of time in generator models and the next wave of LLMs. And so, when this team came back out and said they’re going to focus on gaming, it was but natural for us to find a way to partner with them.
We looked at it as this is a platform to power NPCs, but more than that, power virtual characters in a variety of settings. It could be in other industries, too. And this was just predating that LLM wave. But the other part of it was we invest in a lot of B2B developer platforms. And we look at it as a way of enabling millions of developers to go and build the tools and to build games faster, cheaper, and get to market really early. And I think what has been amazing is to see the second piece, or in fact, there are two other pieces which have been added. One is powering the NPCs, is one thing, but there are all these other components that Inworld has developed over the last two, three years, where they’re also part of the game development flow. It’s not just the production piece, but the game development, the production and also the infrastructure. A lot of the AAA studios, they probably have the infrastructure to support millions of users, but the smaller studios don’t have that. And so Inworld now has this third component where they can power those games and provide what we are talking about in terms of inference costs. And how do you actually serve this with the lowest latency? And part of that is the on-device piece.
So, two years back, we didn’t talk about AI PC. That has become a new element now, right? How can Intel, with its AI core, enable on-device gaming? How can we reduce some of the cost for these game developers? Because if you can offload a lot of that AI models to the Edge, you reduce the cost of running the game. It is also productive for the game developers. It’s also a great experience for the players. So, for from all of these angles, I think Inworld has been amazing as a partner. They’re one of a very close partner of Intel’s Client Computing Group, and we’ve really had a wonderful time working with them.
Florin Radu 18:29
Likewise.
Camille Morhardt 18:31
Well, what is like, one of the hottest topics in gaming right now, from a developer perspective? What are they kind of arguing over or really excited about right now?
Florin Radu 18:40
From the conversations that we’re having with studios–so, assuming that’s representative of the broader industry–there are two or three main topics that kind of come up relatively often. The first one is the cost component, and there like, one of the common beliefs is that, you know every API call or call to a model has an associated cost. Typically, what happens is, you do the math and you say, “I’m going to have millions of players interacting with characters or making requests to the model. I’m going to go bankrupt in day one.” And that’slargely a false belief, because costs don’t actually scale linearly. They may be growing linearly in the very beginning, while we get more data and understand how the use case is actually playing out and how this engagement is shaping. But then there’s lots of things you can do on the back end to optimize those models, to fine tune them, to improve their performance, to use smaller models and basically help costs taper off. The second component, is the on-device piece, where, basically, the cost conversation will largely go away if you’re able to serve everything on-device. And those are all things that Inworld already does and then offers to our customers.
The second piece is around brand and content safety. One, are the characters or the dialog or the output of these models going to be unsafe? Obviously, we do checks for that. But also, more importantly, is it going to be on brand? If you think about companies like Disney or NBCU, where they have all these characters, they have all this lore, if a Disney princess talks about the US election, that’s going to be really, well first, obviously, it’s a really weird experience, but it’s going to break the immersion for that particular child who’s engaging with that Disney princess, and that’s like the worst thing that could happen to any IP holder. So, it’s also about doing robust checks to make sure that everything that is being spit out by these models is in line with, you know, how they should be if it was being, you know, controlled very finely on what they were saying.
And the third piece, I’d say, is probably IP ownership and just IP security, basically making sure that one, games can cost, like, a billion dollars to make, and they take six or seven years, depending on the scale of the game. And a lot of that information, if it gets leaked, very similar to, like, a movie script or something else, it’s gonna ruin a lot of basically all the investment that’s been made into that game to keep things confidential and then allow the players to experience it. So, making sure that there’s no IP leakage. And also ownership, like, who owns the data that’s going into the models and coming out and with Inworld with all our customers, we own none of the data that’s going in and none of the data that’s going out. It’s all owned by the studios. IP belongs where it belongs, and we’re more so just focused on ensuring the highest quality for the end user, for the end player, and making sure that you know all that IP is secure and not being leaked through our SLAs and enterprise level security.
Camille Morhardt 21:01
What do you think is next on the horizon? I mean, are we going to be living in this sort of LLM/Gen AI, kind of new tools available to us for like, the foreseeable future, or do you think there’s some other thing that’s going to crop up to make a huge difference in gaming?
Florin Radu 21:18
In gaming in particular, I actually think a lot of like, the killer use cases that we’re going to see be, like, the gold standard for how to use AI haven’t even actually been discovered yet. Like, that’s one of the most exciting things to me as a gamer. And also, as building this technology on a daily basis is like, what will it be used for in games that make someone laugh and makes them have, like, the best gaming experience that they’ve ever had? So from that perspective, we have some ideas like the ones I mentioned so far are pretty interesting, and I think those are easy to understand, and they’re kind of like the first step, but I actually think a lot of the use cases for AI within games are still being determined now, and it was up to the developers who have been building games for a long time to figure out, “What do I actually want to enable with this game that I can now do because of AI that wasn’t before possible,” and then seeing how that actually plays out when the players are engaging.
Kind of TBD on, like, what that killer use cases are? I think we’re already seeing some of them, but that, to me, is, like, probably the most, most exciting thing that I see coming. And there’s, of course, there’s other stuff, as I mentioned earlier, like on- device model serving, which are going to be a gateway to allow us, for all of us, to experience AI in games by reducing costs and improving latency. And in there and everything else, but those are all steps on the path to actually enabling players to have an amazing gameplay experience.
Camille Morhardt 22:31
Is the goal with gaming, pretty much universally to immerse in a different world and be a different character, a different person? Or is, is there any kind of interest or evolution toward a crossover between your actual world and your gaming world–like playing out a real-world scenario, but in a game or something like that? I mean, since our devices know so much about us already, it could, they could, sort of the two worlds seem like they could come together a bit more.
Florin Radu 23:01
They definitely can. And I think we’ll we’re already seeing like one customer, a pretty interesting use case where it starts to blend the character calling you on your phone, and you talking to the character that way, and then figuring out what, what that unlocks in the game when you actually log back in. So, there’s definitely a blend. But I do think one of the amazing parts about games is having this alternate reality that you kind of go into and explore and build relationships, and allows you to kind of maybe be even the person that you would want to be, but can’t, because, you know, there’s no dragons in this world, and on earth, or whatever it may be. So, I still think games would be largely separate, but they will become more realistic, more engaging, more fun. And I think that’s overall, a positive for everyone.
Camille Morhardt 23:39
With some of the most recent advances in gesture recognition and emotion recognition in AI, are we going to start to see games incorporate that into customized responses based on people’s actual mood?
Florin Radu 23:52
For sure, you can kind of get away doing some of that already, just based on the content of what’s being said. Like, you know, if I say “I hate you”, I’m going to be able to understand the overall intent, but it could be like, “Oh my god, I hate you.” Like, you know that is where the difference comes in. And there is some benefit to being able to take the tone of voice as well as the context of like, the broader discussion–and taking that in so the player knows how to adequately respond and knows what the impact should be on itself, whether it’s through a relationship meter or something else. So that’s actually gonna be very useful. Like I said, some of that stuff can already be packed because of the actual content of what’s being said and the context. But it’s only gonna get better. And then that’s kind of the kind of stuff that we’re actually gonna use probably intensely for the future.
Srini Ananth 24:33
What I would love to hear is, when do we start to see some of these experiences? Can you talk a little bit about the development cycles for games like this? Because there’s this expectation in the Gen AI space of everything getting to market very quick but that’s happening in these sort of consumer apps. But the people’s expectation as the cycles move so fast that if you don’t see a use-case, that whole space is dead. But as we realize in certain verticals–whether it’s a drug discovery or gaming–it is not that quick, because there are tool chains and there are a lot of processes and all those things that have to get integrated. So, I’d love to hear your take on what does it take for this, you know, the first game with AI to be seen in the hands of a user?
Florin Radu 25:16
Yeah, I think it depends on how you’re defining game. There are already some examples, like the one I gave earlier, with Death by AI, where that’s a game. It’s not a AAA game, but it’s out. It’s being interacted with by players. People are loving it. It’s blowing up. It’s definitely engaging. And so that one was obviously very quick to market. And then on the other end of the spectrum, you have a AAA project productions, which can cost more than a billion dollars, with the budget and marketing and everything else, can take more than five years to make. Those games will take a while to come to market. All those design decisions are being done now, and how to implement AI and how to test AI, because it will take so long to actually integrate it in a way that doesn’t degrade the quality of these games, because that’s really what makes them stand out.
And then you have kind of the middle section, which, depending on how you define it, could be indie games, could be AA games, and those will maybe take a year or two to come out. And we already have some early examples of partners like Netties Games and Niantic, smaller games, demos, but these are, like, the first touch points that people are actually having, and in a way, like warming them up for what’s to come at the same time, giving about the developers, like very interesting insights into what’s working what’s not, so that they can iterate, integrate that into the game and when the game finally launches, hopefully it’s, you know, a very fun, engaging experience for the player.
Srini Ananth 26:29
Awesome.
Camille Morhardt 26:30
What was the first game you ever played, Florin?
Florin Radu 26:33
Uh, to be honest, I don’t remember the first game, but the game I played the most was definitely the one Srini mentioned in the beginning, Empire Earth. It’s a real-time strategy game. I think it came out in like, 2001 and my dad, my brother and I would just have these, like, massive LAN parties. We’re just all in our own rooms, in separate edges of the house, building up, like, cities from the stone age all the way to, like, the Nano age, where you have like, laser guns and you basically, that we would end the night at like, 2-3am and my dad would always win, but my brother and I would always like, try to team up and make a pack to, like, attack his city. So, it was like a very fun game. It brought us closer together and, and, yeah, that’s probably the one where I’ve logged the most hours. But it’s quite old at this point, so I had to look for the next one to start playing.
Camille Morhardt 27:14
Well, thank you, Florin. Really interesting conversation. Florin Radu is a VP at Inworld, which is a platform that allows developers of games to actually use AI to make the games more interesting and more compelling, have the NPCs come alive and make for more interesting game mechanics. Thank you so much for your time today.
Florin Radu 27:37
Thanks so much for having me.
Camille Morhardt 36:26
And Srini, thank you for joining me to co-host from Intel Capital, Managing Partner there.
Srini Ananth 27:44
Thanks for having us, Camille. And thanks for Florin for joining us.
Florin Radu 27:48
Thank you.
