Beyond Agents: What AI Strategy Really Needs in 2025

Speaker 1 [00:07]

Thank you for being here. So today I want to talk about what I currently think about AI strategy and what we really need and what we probably should start to think about. Not unintentionally, you see like a lot of whirlwinds in these pictures. And who feels like that sometimes if it's about AI and stuff? Just raise your hand. Yeah, many. Okay. So this is me. You can find me on LinkedIn. You can find me on the Discord here. Don't hesitate to reach out. I want to tell you about my journey to NVIDIA GTC, what I learned, and what strategic implementations I have drawn from this experience. So first, why did I decide to go NVIDIA GTC? NVIDIA GTC is a really big conference. It's the premier conference of NVIDIA. I think everybody knows NVIDIA, graphic cards, GPUs. It's very important for anything we do with AI nowadays. I thought, AI moves fast, and the conference will give me a broad overview across many industries, what's happening in AI, because I mostly, my bubble is applications in finance or applications in business, and I don't know much about robotics. I don't know too much about other domains, medical stuff and so on, and see, okay, what's really happening? And the best way to go is, just go there to get deeper insights, what's going on, maybe see, touch things, because we probably know, if you look at your LinkedIn feed, you see, oh yeah, this is happening, oh, this is great, so many applauses, and okay, but it's LinkedIn, and it's too many people just talking about things they probably don't really understand. But something's going on, so there's something behind all these messages, and let's give it a reality check and um and i also felt like yeah i was a little bit also like there was some ai fatigue because i've been talking about ai and data for many many years and then you basically eventually learn okay sometimes people don't really listen they want fancy they want to have fancy prototypes and don't really want to fix the real problems um and said okay let's go it get a new perspective fresh ideas let me tell a little bit more about gtc if you don't know it um i was really happy to be back at this conference center because actually 10 years ago i gave a talk there for mongodb what's the different thing but it's 17 000 participants inside is really big it's like it has a keynote with like 25 000 people and the fun fact there zero gaming so we said okay there has to be some gaming and we get nothing at all so it's all all about applications, AI applications. And of course, there's the keynote. Here you see me in the keynote. It's in the stadium close by in San Jose. And there was the keynote, and there were like 25,000 people listening. We have sessions, talks, tutorials, trainings, all days, networking events. You have also a large exhibition with people showing robots, selling services. So it's not like a conference like this where it's all about, here we talk about, hey, we want to exchange knowledge, we want to network, we want to help each other, we want to discuss things. There's, of course, also like a big sales part there, which I don't really want to cover. But what were the main topics at the conference? Of course, disclaimer, this is a sales conference as well. It's a technology sales conference. is in state it's also like of course a little bit of old maybe overhyping things but it's still like a tech company and one of the most important like most important and most successful tech companies in the world and there has to be something behind it so what were the main topics main topics were ai models and intelligent systems another topic was like ai models this includes like the latest advancements in foundational models generative ai autonomous systems from from large language models to intelligent agents. Infrastructure, of course, because NVIDIA does graphic cards, they do stuff for data centers. Of course, data center infrastructure, I have no clue, actually, to be honest, about data center infrastructure. It was a big topic, and what's happening there? What are the trends? Where does it go? Simulation and robotics, and it's not unintentionally like a combination here. So we're talking digital twins, to applications and machines that basically do stuff in the real world. Developer tools and ecosystem is one of the topics because of course this is all run by code. So of course there's a lot of developer stuff going on and of course also like applied AI solution. So you have many, many different talks and sessions, over a thousand sessions. And so I want to tell you a little bit what were my takeaways. So I made a new friend here. Cute, isn't it? He didn't follow me home. He wasn't that autonomous yet. But a good message first is for Python, Python is a key language in NVIDIA. Because if you think Python, oh, everything is under the hood, rusty. No, no, Python is the most noted language in the NVIDIA ecosystem. system. NVIDIA puts a lot of effort in that stuff works in Python on NVIDIA, and Python is like a first-level citizen in the whole ecosystem. And just like fun fact, because there were too many talks to attend, also like some talks were just like crowded, that's why I decided, oh, I can also like scrape these 560 videos and have them transcribed and more run analytics on this, and this is just like an easy part when . Here you see the talks with most mentioned Python. So if you want to hear Python for 80 times in one talk, go to CUDA New Features and beyond. But you see, it's important. And also, of course, with all the fancy CUDA stuff, which we know is important to optimize GPUs and performance. Of course, AI is everywhere and in everything. And this is also like a different perspective I want to bring to the table. Because, of course, if you are in business, analytics, figures, numbers, natural language processing. Of course, you have like a little, you're inside that bubble. And if you look like on a broader range, where is AI nowadays? You basically see it's in each and every industry. It's in healthcare, it's in cloud, it's on edge. And here you see how often was AI mentioned. And if you wanna hear the term AI for like 260 times, watch the keynote. But you see, it's like the most said word probably at this conference. But it's everywhere, and the message here is it's in each and everything. And what were actually the tech trends there? So the tech trends actually were, of course, agents are everywhere in anything. That was no surprise at all, because AI agents in large language models was very clear this is going to be one of the hype topics. But agentic AI is also in everything, agents, agents, agents, but agents are not only deemed software agents. So also if you say you have also natural environment agents, like a car is an agent, you have like these production lines here, you have, this is like a simulation where robots learn to walk in the more or less difficult environments, and you see like this is also regarded as an agent. the next thing that is on the table is synthetic data so um first time i heard about synthetic data it was at the time in deep learning and deep learning synthetic data it was like yeah if you don't have enough data to learn on you can just like use synthetic data and i said okay this is like a really bad idea and basically i think almost everyone agreed but here synthetic data is something different so you have generative AI creating these new videos and I think that really makes sense because you can have like a street different environments you can have like different color sets and I think this robot really like in the kitchen is a really nice example you can have different settings and train models on that and you don't have to basically set up everything you can record in the world as real data so synthetic data is bringing a lot of more data for training to the table for these use cases. Next thing is simulation. This is a really big stage. You see, like, here's Jensen, Juan, the boss, this tiny person. I mean, this is like Katy Perry, no, how's it called, Taylor Swift-sized stage in a stadium, and all this was live simulated. So you see, like, okay, this is like the NVIDIA and this is a life-generated simulation. Here, also, it's not a video game. It's also a life simulation, and you see how, with generative AI, new environments were just generated live, and you really have to imagine, this is a really big screen. I mean, this is a tiny screen. This is... It was really insane quality as well. This is another thing NVIDIA really brought, or like promote a lot of simulation and then we have digital twins and I said oh digital twins are back because if you're in Germany you have heard a lot about digital twins for more than 15 years probably it was one of the big topics in the industry for zero and I would say oh probably if you say digital twins in Germany nowadays you say oh no we cannot listen to anymore like does it really go And here, BMW showed a battery plant, and they actually can fly with colleagues in the battery plant and show other people around the world. They have like six battery plants all around the world, and then they can connect, okay, bring the people from the colleagues from Mexico in and say, hey, this is how we do it. This is how we do it. And they also plugged in large language models to ask questions to the simulation. So, of course, the simulation does not live, but it can help you, the large language models can help you to navigate and find stuff there. And, of course, you can go in there, you can collaborate, you can discuss things, and I think this is a really great approach. Fun fact I have to add, I had the pleasure to meet somebody at Easter who is also in the automotive industry and said, okay, yes, this is a little bit overhyped. It made a wave in the car industry. It was a little bit overhyped, but I think this is absolutely the right direction to think. You have like a simulation. You add other AI things like large language models to answer questions, to help you navigate. It's like a mixed, it's like a huge remix of many, many tools and technologies we have that bring real new value to the table here. And yeah, so I think that's probably not fully done yet. Maybe a little too overhyped, but definitely leads into the right direction. Of course, we also had humanoid robots. So, human robots. Oh, how stupid is it? Hey, what can you do for me? I can help you with information about the GTC event, give directions, or just chat if you want. Oh, that's awesome. Can you tell me why it's so troublesome to sign up for the trainings and there's all these queues? I think that's pretty badly organized. What do you think? I'm sorry to hear it's been a hassle. Organizing big events like this can be complex. But your feedback is important and can help improve the future events. Make any sense. yeah so of course humanoid robots are for attention and for sale but they're not like the application we are looking for in the near future that's my take because you basically see this is like one of the top top ai companies in the world and not even like the gestures were a good fit and this was just like this really narrow large language model interacting with me complaining about overcrowded rooms and trainings but you also saw other probably useful robots um so here we see like a robot like as we know from production we see a robot operating on on stuff um we also see autonomous robots oops that should play or not No, it doesn't play. Let me describe it, sorry. So, okay, this is, unfortunately, video is stuck. So this is also like an autonomous robot, and it delivers you the envelope. And this is a fully autonomous system that can come here and say, hello, Alex, thank you for the envelope. And then I put it over here, and the robot came and picked it up from there. And it was fully autonomous doing. And, of course, this is not something you basically need to know, but it's also leading in an absolutely interesting and right direction. We already saw Qt robot. And apart from robot, everything, and also robots, everything runs on open source. Like, NVIDIA released an open source model for robotics. So, basically, anything they do, or companies like this do nowadays, everything is open source. And the models are also open source, of course. Full disclaimer, we probably don't know on which data they were trained. They have probably not the weights. They are like open-sourcish, or like we could call them rather open models. But there's a starting point you can get stuff doing. You can get a starting point to do stuff. And the other learning is your software. Software is worthless if you want to sell hardware. You give the software for free to sell hardware and stuff. And this is another driver for open source. but also like an enabler for many others. So this is not a black and white thing. A really nice thing you probably look for if you, who's into AI development and is short of GPUs? Yeah, a few, okay, yeah, thank you, yeah, many, yeah. Okay, NVIDIA is releasing this box here. It's the DGX, they announced it under a different name during the Christmas break already. this is a tiny box with 128 gigabytes of VRAM like graphic memory and it has one teraflop and it's not there to replace the laptop this is still your development laptop it's basically something where you can run your models locally and just connect it you have basically a very powerful machine to assist you on AI stuff or train stuff or do whatever and you can just like plug it in now it's going to be released in summer uh like the good the good part was i could pre-order one because i attended gtc um but also another interesting thing is there's also like a really huge one the the dgx station basically where you see you have uh i don't know i can't read it anymore like many more teraflops and of course this is also like really interesting because people for example clients why in finance they they don't can they cannot easily go to the cloud um you still stay on premise and then you have like this insane power mission like a really power machine under your desk and you can do your number crunching and you don't really have to worry about all the governance stuff and many all the paperwork and everything if you move stuff for that in the cloud and also locally often is also faster than with cloud delay um so i also want to give you some more insights into data center stuff they announced because this is another thing because very often overlooked is with all the generative ai um there's some voices posting hey we shouldn't waste that much energy on basically regenerating the same pictures on linkedin because hey we've seen this already um but one of the big messages and the one that's the stuff that's really moving fast as well is the shrinking data centers so this is this has a power consumption of one megawatt this is just like one rack and i took a picture at the exhibition so you see this is the rack with like just regular people stand next to it so it's it's not really big it has an insane amount of computational power you can you can see the details here or watch the keynote it's at length but one of the main messages was also basically jensen said on stage oh don't buy our stuff we currently sell wait for this because this is way more energy efficient so making inference cheaper and more efficient is one of the main topics and they also make a lot of progress here and it's seems to be really like an impressive piece of machinery this is built out of one million parts so let me see and then yeah you can yeah so that's why honey i shrunk the data center, so data centers become more efficient, smaller, applications become cheaper to apply because less energy consumption. And what's the strategic pattern? If you put all I showed you together, the strategic pattern is everything's converging. Like AI is no longer siloed in this application or that application. It's like everywhere. It can be in anything. It can be on edge, it can be for your business analytics, it can be in a robotic. I also analyzed like almost 600 GTC talks and also like when you did the clustering, it revealed all the topics and were really clustered really closely together. There were only like three clusters and everything was very densely connected, even up to applications in medicine, biology, research, yeah, basically everything. It was just like quite insane to see it all come to one place together. And the innovation actually happens at the intersections. I think it's very important to think cross-domain, to work cross-domain, because it's just important people who know how to build machines robots whichever application in biology or medicine they we need to be able to communicate to work together we totally need new and more improved working frameworks and setups for collaboration to really get all this together because not everybody can become an ai expert and we have like like a lot of machinery which is not too easy to build so there's also like really deep tech experts and we really have to think cross-domain learn from each other and compilation collaboration is the key in all that was coming um so yeah car engineers need ai biologists need cpu and we have to all operate this so there's a lot of work to be done and we have to tear down many walls and let me show you a little bit where this is heading if If you put all this together, like AI, according to NVIDIA, is moving into the physical world. And we must design for that. And you see like a circle here. We have like simulation. You have a digital twin. You simulate. You train it on simulated data. You train it. You deploy it in the real world. Of course, you also take like inside sensor data from the real world back to retrain and prove your model just as we do. And you see, like, this looks like a next, like, superstorm to me because, like, oh, now we simulate, we have a digital twin, we simulate, we eventually bring a digital bring to life in the real world, not necessarily like this humanoid robot, but applications. And I think this enables, like, really fast development, but it only enables really fast development if you are, like, working really agile in a very effective team. It doesn't work if it takes like half a year to get an answer from the other department or if you have inner department fights. So you really have to tear down many, many walls. Otherwise, the rest of the world will be just faster. And one, this collaboration really got my eye because also they have to really rethink everything new or like a lot like new, everything perfect. So this really caught my eye. Again, unfortunately, another cute robot. And probably this looks a little bit like Star Wars, right? And this was a collaboration announced in the keynote. And it's a collaboration of NVIDIA, no surprise. Google DeepMind, yeah, why not Google DeepMind? They already have like a Nobel Prize, reinforcement learning, they're one of the top people for anything AI in the world for many years. Oh, and Disney Research. What's the question? Like, what does Disney bring to the table here with all these fancy AI things? Actually, Disney brings two things to the table. So this is like a video where this robot is autonomous. It walks autonomously. This is not a movie. So actually, it came on stage. And this is an autonomous system. And the question, what is Disney bringing to the table? Disney is bringing two things to the table. Amusement parks? Who has been to Disney World? Well, okay, I have some catching up to do, I've been at seven, and yeah, but I'm not a Disney fan necessarily, but I can tell you like everything they do in animatronics, entertainment, like the systems they build around, they do a really good job and they do a really good job and they have their own research department since like the 1950s, so if you go there and you go to Indiana Jones ride or Rise of the Rebellion, Star Wars, it's really impressive. Illusions, animatronics, animatronics is like other robots. It's really good what they do, and they bring a lot of experience on the table. And the other thing they bring to the table is Pixar, because Disney bought Pixar, and Pixar did what? Pixar does a lot of movies, and the movies are made in the computer. And Pixar has a really interesting physics engine, which you can use to simulate. So Disney brings two things to the table. Maybe three things, like how to make robots look really cute. And this is like a totally new approach, like to have entertainment and tech and research-heavy company like Google DeepMind collaborating together in this weird mix. and this might be of course, you can always say this is a keynote this is a prototype, we don't have to worry about it because of course we always say there's so many other problems that are not unsolved because we see the problems because we're German it's right, we also point and see problems but I can also go just like two weeks back because fun fact, a few weeks earlier I happened to be in Los Angeles as well and we just took a Waymo and if you see, this is an autonomous car driving in Venice Beach you see it's really dark if you know the US there's no sidewalks people just like there's a car and sometimes pedestrians have to walk between the cars and stuff it's really dark you see here you have a display in the car where you see there's a pedestrian crossing you really see what's going on and I would argue this car has a better oversight at night than any human has And we were driving it, and I had an electrical engineer in security in the car, and he had zero complaints. We all felt really safe being picked up by a robot taxi. This is also taken just on the streets in Los Angeles, one of these tiny delivery robots. So, yeah, we read about it maybe five years ago. I don't know, sometimes I tend to forget because we have all these ideas, some things announced, and you see, oh, this will be on the street and in production tomorrow, and then it never happens, and you say, oh, it's probably never happened. Here you basically see, just a few miles away in Los Angeles and also parts of San Francisco, the future is already dripping in. So this is not a waterfall thing. It's dripping in, and it's definitely happening. So some final remarks. And sometimes in the press, the GTC said, oh, it's like AI Woodstock. I didn't get any Woodstock vibe here. I would argue we have a lot more Woodstock vibe here because we talk to each other. We have crazy ideas. We work very agile. It's a business conference. But regardless, you see, like, long-term strategic investments in open source continue to build value for companies. And we really see huge companies who did, they already see the profits. They already see it in sales. They see it in market dominance. And, of course, we have a significant competitive advantage in the US because they just take a little bit more risk than we do here. And, of course, the convergence of simulation, AI, data, many other technologies, probably other technologies I haven't even mentioned here, probably technologies I'm not even aware of, brings a new paradigm shift, like how we develop things in the real world. and i think of course we have a lot of knowledge of ai business data things we do here um yeah and sadly also i see like technologies like digital queens are also like now bringing extra value to the table because they help to develop for they for rapid development you can do digital twin you can try you can simulate this is also not something completely new of course if you have a production line in the automotive industry of course like you simulate robot arms like building stuff already but if you think it a little bit further for further applications with simulation synthetic data and ai mixed in with generative ai large levels models ai technology you name it in this mix there's there's many new things and a lot of um extra value and totally new approaches and speedups happening. Yeah, and that's basically it. Thank you. Thanks for listening.

Speaker 2 [28:10]

So, thank you for the talk, Alex. We have a few questions on Slido, and I want to prompt everyone, if they haven't already, they can go into Slido and ask their questions there. So, the first question was, why in a talk that is about AI strategy was NVIDIA, why is NVIDIA so central, if you talk about AI strategy?

Speaker 1 [28:32]

AI strategy? I think it's just like very good to explain stuff. Of course, strategy usually works on a higher level. But I think this is like really very visible strategy. You can really touch, see and understand because I see one issue or one thing that is very often hard to solve. You have like high level strategy with people deciding it with zero hands on. And I think it's very important to bring the hands-on and the deciders closer together because then you can make better decisions if you actually have a deeper understanding of what you do. And I think the NVIDIA conference, because I don't have any time to NVIDIA or something, disclaimer, I think it was a very good hands-on oversight to give to understand strategic implementations.

Speaker 2 [29:24]

So the next question is, what is your take on what comes after agentic behavior in a technical sense?

Speaker 1 [29:32]

Insanity? Insanity? Insanity? Yes. I mean, I don't really think one thing comes after another. We have to think in parallel. Of course, we are very biased towards agentic AI because many people drive it. If you listen, if you hear this and you think about, like, not hiring software engineers or data engineers because it's all being done by agents soon, I assure you will make major mistake in strategy. It's just like one hype topic and also there's research that says how much value can generative AI bring to the table? It's the estimation is less than 15% and the rest is still like other parts of AI analytics, machine learning, prediction. It's just like become a bit boring compared to agentic AI dreams. Because if you, who has played a lot with agents already? Yes, so did it solve any real problems, big problems for you? Like a real big problem, no, not so, yeah. So I think it's great for boilerplate, but when you have like a little bit more complex problem, they already start to struggle. So I don't think there's the next. I would suggest, apart from all the talks and stuff about Engine AI, listen to Jan LeCun, what he says about large language models. And he has a lot of really good things to say. And he's the head of AI for Meta. So it's not just like any exotic researcher. He says large language models are like just parrots, and they don't learn anything, and they cannot really build a strategy. And we really need to separate ourselves from, oh, I just wish I do a little typing, magic, create solutions of your solution. It's just like a dream in my head. My brain wants me to believe this, but it's not happening. If you work with agentic AI for software development, it makes a lot of really stupid mistakes. I, as an expert with a lot of experience, say, oh, yeah, tiny mistake. What about using the GPU for training? stuff like that. But new people won't be replaced by that. You still need the experience to see what's working, what's not, and what's the bigger picture.

Speaker 2 [31:56]

So, many companies didn't start with AI yet. Would you say they should skip many steps and directly dive into autonomous agents, or do you recommend an incremental adoption of AI?

Speaker 1 [32:09]

I mean, you always have a learning curve. What I suggest is talk to other people. Forget about your domain. That's like regardless which domain you're in, forget you're in the domain and you have to talk to other people in your domain. Do that. My better advice is talk to people who are not in your domain and look for people who probably already solved your problem and have experience because we've seen this very often. We have different tech bubbles or research bubbles. So my favorite example, who knows what econometrics is? Yes, maybe 20%. Okay, it's big data for economic people, for economists. And we see this in many parts. You have like a different terminology, because of course it was analog books. We don't talk to each other. and my best advice is talk to researchers, talk to startups, talk to your competitors, and really think more like, how can I be fast? This is the biggest question. How can we deliver and implement fast and not who owns what and all the stuff? Because everyone else will be faster.

Speaker 2 [33:27]

And what would you say is the key message for AI strategy?

Speaker 1 [33:31]

I think we need even more out-of-the-box thinking. Even more, we need to basically think, I wouldn't really, I don't really like the term AI first, because it's just a tool or a helper, but you basically don't, I think you have to let go from how do we solve it in real life and now what can AI do as assistant here, we have to bring new approaches see okay what's in our lego box as as parts we have ai we have this and we have probably also good data that's of course still the most overlooked part data quality and what do we also have a skill set in people because you won't probably also won't be able to find the right skill set and right people and the right motivation and the right mindset and this is a mix and then you have to think okay what's your goal what do you want to reach and then bring the different parts together in the right order so that's uh sorry it's a very generic answer but this is how things nowadays everything is being remixed we really need new ways of thinking tearing down all our walls and the hats how how we used to do it which department is responsible for what we just need closer closer and even closer collaboration and real talk and less fancy demos.

Speaker 2 [34:58]

So many people were also interested if you have other examples of companies that heavily profited from open sourcing their Software products or research efforts especially companies whose main product is software or technology

Speaker 1 [35:12]

Software technology, I mean, like, who has profited the most? Meta, Facebook, Instagram, YouTube, Google, Microsoft, NVIDIA. And why is that? Because they put billions in the system. And they put also, like, a lot of money in developing. And, of course, we say, hi, thanks for PyTorch's open source and free. We did, or Wolf did a question in your SciPy keynote, your SciPy, there's many maintainers and developers of scientific open source packages there, and he asked who's being paid for developing open source at this scientific conference, 60%. So open source is not like people doing things at night. invest in open source because they have understood software is you don't need the software to run your business you need the speed you need the reach you need the product like instagram and the rest you can basically rather share bring the right people in because if you look large language models and everything how how fast did it happen of course yeah there was the chat gtp moment but it was not oh there's open ai and just jet gtp releasing the software basically next day microsoft announced something a few days later oh there was also google releasing something they have all been working on this already for many years and also like a lot of speed up is also because not only open source software there was also open research in in natural language processing. This is why we also see these large speed-ups. Again, they even shared knowledge. They shared the software. They even put money in having better software because we have the best engineers for PyTorch. You have the best people to implement it into your product. So if you, I mean, in Meta you don't ask, do we know anybody who can do PyTorch? Oh no, actually we dominate a system. We have the best people. We drive where PyTorch is going. We have the top experts to solve the tox problem because we built the software or we're basically one of the major drivers of this software. And of course, open source is always free. Don't romanticize it. Open source is a big business and it's just like a different approach about software. And I also want to remind you, one angle is Microsoft was the evil player that started, or no, IBM actually, that started making money out of software when they realized, oh, we can sell it to enterprise customers. Because in the very beginning of computers, when it was at research institutes and universities, everything was open source. And then they just realized, oh, we can sell it to banks. We can make an operating system, Unix. Oh, and the banks will pay a fortune for that. And then Microsoft came in and realized, oh, we can even scale it like Windows to a consumer market. And then software became a business. And now, you see, I buy this MacBook, the software is for free, already here. It's not open source, but still for free. So it's like a totally different approach, because you think about, what's my service, what's my product? And the software is something you need, but you don't necessarily need to control it. You probably need to navigate it, but you don't have to basically keep it as a secret.

Speaker 2 [38:52]

So you were also talking a lot about robots and robotics. When will PyCon DE have a walking robot on stage?

Speaker 1 [38:59]

Oh, actually, we had that already. I mean, you can always have, like, one of these humanoid demos and stuff, and I don't really think, yeah, so we don't have plans. But I would like to see if you're in a startup, open source, anything robotics-related, let's talk. I mean, we love stuff like that, and we would love to bring stuff like that or more stuff like that also to the conference because what you probably don't know about PyCons in the Python community, we were always thinking cross-pollination. So the astronomer can talk about image, like industry can learn from an astronomer about image problems, noise problems in images. Many problems have already been solved by other domains. And this is the place to talk to each other and say, yeah, I have a solution already. We figured this out five years ago with less hardware, with less computational power with more smarts.

Speaker 2 [40:01]

So if you want to do something in robotics, you can also come to PyCon, right?

Speaker 1 [40:05]

right yes sure you're very welcome um where yeah you find me on linkedin message me tomorrow now

Speaker 2 [40:05]

Yes.

Speaker 1 [40:12]

yes no definitely uh let's make this let's let's let's let's drive because like pi con and pi data is always like about innovation and open source like we have something about we have our values opens open source innovation and we want to have tech fun right and robots is not the best the worst part to do have fun with right but also drive innovation and learn from each Thank you very much, all of you.

Speaker 2 [40:36]

So do you think AI will be more commodified in the future with the removal of AI silos and collaboration of different fields?

Speaker 1 [40:45]

I I don't know. I don't know. We have to see where it leads. I don't have any insights

Speaker 2 [40:54]

And then a question about how should an enterprise store its corporate data today to be prepared for upcoming AI technologies of tomorrow?

Speaker 1 [41:04]

You should prepare your data in well-structured, well-documented ways, invest a lot of time and get good people for doing your data engineering. You basically have immediate and fast access to your data. You need to be able to experiment and do new stuff with your data. And of course, this is very complex and you probably need one of the most smartest people in data engineering because the mixture of having good usable data in combination with governance is a highly complex issue and you should really invest in that because if you have a new idea a lot of the time is wasted in ai projects or machine learning projects since forever because yeah get the data get the data right clean the data until you basically can do something with the data and if you have the data ready you can do that and i also want to have a footnote. I've been preaching that for more than 10 years and I'm getting tired. Just do it, leave it, but leave me alone if you don't do it.

Speaker 2 [42:08]

So you touched on governance and how do you see the topic of AI ethics as part of this conference and also of the future in general?

Speaker 1 [42:15]

Well, we've been talking about AI ethics for many, many years. I think it's very important, and it's very important to have a basic understanding to see every AI, every model has a bias, and you need to be able to deal with the bias. And, of course, we have to work towards, like, minimizing bias. There will be always some bias because you never can put the whole world and everything in every angle and every future, past, present angle from every single into an AI model. There will always be a bias and there's limits, but I don't think it's more like how do you act on these biases? What do you do? Because we will never get rid of them anyway. So it's very important. And one fun fact, who has been at Feminist AI already? No one? Oh, yes. How did you like it? Yes. So if you want to learn about stuff, hacking, large language models, AI models, biases, go to Catherine and Ines on the third floor to the Feminist AI Alarm Party. And you can do really hack some stuff and learn hands-on and get out of the theory. And yeah, I can really recommend it. I'm very, very proud we have this satellite event within our conference.

Speaker 2 [43:36]

Yeah, so we have one minute left. I think that's enough for one quick question. So one question is, many people believe a major profitable application for Gen AI is still missing. What do you think about that?

Speaker 1 [43:49]

Yeah, maybe there's no profit, I mean, maybe it's just like, yeah, it's a nice thing to have. I mean, for example, like, you probably have seen people like putting them like into these toy boxes with the tools and everything, and yeah, you see it in the social media feed, you see it for like three people, and then it's boring already, and this is like a thing, oh, yeah, we had like this viral thing for one day, and please stop doing it because we're already bored seeing everyone in your network being like a toy figure. Yeah, we got the idea. This is like, with generic AI, it's too generic, actually. It's really boring. I mean, you really have to think, what brings a new idea to the table? Full admission, did you see, okay, I have to admit, the first slide was actually AI generated, but with all the Gen AI, I used generated images for all my slides for since 2017. I stopped because the pictures you generate are quite boring at the end of the day. And what's your real message? And this is not generated. This is actually hand-drawn, and it's from Captain Future, if you know that. Yeah, so, yeah, it's also a thing, if you give a lot of people without inspiration just tools, don't expect to get anything new or something inspirational will come out of it. And I think with generative AI, it's for AI art, it's for using it for anything you do, and it's using anything like brainstorming, everything. You really have to see, okay, it also might make you lazy, because every time I ask Jet CGP, okay, can you please do some Baylor Pro for me? Of course I wish, please do the perfect code for me, I just want to get this going, it's worrying, my attention level goes down, probably you do copy paste and we know this problem is this like is this not just like copy pasting from stack overflow on steroids and sometimes because if you know stack overflow people google it stack overflow gave good advice code examples many people make their living googling copy pasting it into applications but we see what happens if people do that not good things happen and you have to think yeah sometimes it's thinking is hard and generative AI has makes us lazy thinkers and we should be very aware of it and we should really sometimes think okay turn off the television turn off the chat bot and read a book and think because that helps because generative AI is parrots it's there's no new ideas this is all just recycling stuff that is already there. That's why I, that's my final word, everything with large language models, I think, yeah, it's just like more fancy search and stuff and prototype writing and things. But that's the end of the matter.

Speaker 2 [46:50]

Thank you, Alex, and thank you for the talk. Thank you.