Enterprise AI - Current and Future Frontiers

Hello, hello everyone. I hope you are having an awesome Monday. My name is Daniel. I'm a part of the FI marketing team and you're here with us for another very, very exciting event. And this one in particular we're very excited about because we've had Dr. Vijay with us before and that was a pretty thrilling webinar as well. So before I intro Dr. Vijay, let me just talk about FI a little bit. If you're here for the first time, FI is the world's largest network that turns ideas into fundable startups and fundable startups into global businesses. So today we're here with Dr. Vijay Khedadi. He is the CEO of Syntegral. He's also the author of the upcoming book Ultra Human. So he's going to be here talking all about AI and its impact on large corporations, what's the current state and what the future holds. Before we bring Dr. Vijayak on stage and he introduces himself and he gets right into the topic, a couple of housekeeping rules, right? So if you look to the right of your screen, you should see a Q&A button right under the chat button. So if you have a question for Dr. Vijayak, pop your question, click the button, pop your question in there, and we'll kind of go through them once Dr. Vijayak is done with his presentation. And once the presentation and the Q&A is done, we have a networking portion. I encourage you all to please stay back. If Dr. Vijayak has some time, he'll also stay back. I'll be here as well. It's a great place to kind of network and get to know people from all across the globe. Speaking of all across the globe, I see that we have people joining in from Australia, Finland, Milan, the Netherlands, Germany, Morocco, More people from Germany, Milan, India. Awesome, awesome. A lot of people from Finland as well. So yeah. And yeah, so before we begin, right, if you could check, see right at the bottom of your screen, you see an emoji button. If you like what Dr. Vijayak is saying, please show him some love. And yeah, without further ado, Akshay, if you could bring Dr. Vijayak on the stage. Hi, Dr. Majak.

Hey, Daniel. Hello, everybody.

Awesome, awesome. Thank you for joining us again, Dr. Majak.

Yes, it's great to be here. The first time around was a great experience already, so very happy to have this session today with everyone who's here, and I hope to connect, as you were saying as well, afterward with whoever wants to chat more and just explore and make connections and so on. I'm going to be open to that.

Awesome. If you could share your screen so we can kind of begin.

Yep.

So I'll just move back and the stage is all yours, Dr. Vijayak.

Okay, do you see— are you seeing me as well still, or are you just seeing my screen now?

Yeah, we can see you as well.

Okay, good. Well, that's good to know, so I shouldn't do any silly things because I think I'm not on camera. Good. Okay, no, there. Okay, good. So yes, uh, welcome. Hello everybody, and I can't see you guys right now, but Again, you know, I know that you're all there and you're listening. I can see that there's people here, so, and I hope to see you actually afterward. And, and as I was saying before, to, to, uh, exchange a little bit, you know, that's actually where the great value of these type of events lies, uh, when we can have some kind of collective, uh, brain, uh, power going on and create some collective, collective synapses across the web. But, uh, I'm gonna start this whole process by sharing with you about my learnings and insights from the, what I call the frontiers of AI adoption in 2023. Obviously the post, the big LLM explosion and how and why can I do this? Why am I, you know, what qualifies me to do this? A little bit about my background. So I've actually, I've been in the digital innovation and digital transformation space for many years. I have led the digital innovation of a university in London. I also led an incubator, an award-winning incubator in Toronto for a couple of years. So I have been involved where I was involved in some as well in some corporate spin-off ventures and had some insights into the adoption process of new technologies, but also in other roles, you know, I've been involved in various digital transformation initiatives. So the process of, you know, seeing how, seeing through innovations and new technologies and new applications when they come out, how that happens, how it's integrated into an organization, what the resistances are, what the challenges are, and so on. I've seen that from that side, but now I am the founder, co-founder, and CEO of Syntegrol, which is a startup where we are leveraging the latest stack of technologies centered around large language models, but also several other technologies in the stack to complement the LLM in order to tackle a particular problem which we saw as especially important and challenging and interesting and rewarding, which is the challenge of compliance and reporting specifically for sustainability purposes. So we're working right now with several firms in the space, uh, that are helping others to generate compliance reports and insights on sustainability practices. If you know a little bit about the space, there's lots of different reports, reporting standards, there's lots of different frameworks out there, and, and vast data sets and different data sources and you need to really read through all of that and be able to have some accuracy in the outputs that you produce. And this is something that we're focused on with Syntegrol. So given that that's the case, you know, we built our MVP earlier this year. We had it done around May or June, May or June. And since then we've launched pilot projects with lots of different companies and those processes have been, very insightful for us and have allowed me to see what's going on inside the enterprise right now when it comes to adoption and the implementation of this brave new world of what's possible with AI, particularly with generative AI. Okay, so before I go and share about, you know, I want to basically, this talk, I want to get to a point where I'm sharing with you guys some of these experiences from the front lines basically to say like, look, this is what we're facing, this is what we're seeing, here are the challenges and here are how we are overcoming these challenges and here is how others are doing it, right? So, and then we could start a conversation around that hopefully. And again, I'm not here to, this is not, I'm not sharing any kind of final solutions or anything. We're experimenting and tinkering in that space and I'm here to share some of those insights from that lab basically. So that's the idea. But before I do that, let me just give you some general context so we know, you know, obviously most of you know the general context I'm sure, but still I'm just going to give it for those who maybe want a little bit of context. Obviously we know that there is a vast explosion of generative AI capacities this year, which has gone into the culture and gone into, you know, the collective consciousness, everybody's talking about it everywhere. People are applying ChatGPT and lots of other models out there, generative models for images and videos. And now music even has just been launched. It's amazing. You can, both Google and Meta, they've launched a generative music tool where you can just hum a melody and it just turns it into a full thing for you. It's really a crazy world of new capacities. And what's the difference between, for those maybe who want to, who come from a, not from a tech background, who are kind of looking at this, what's happening here, what's the difference? What's the difference between these new models? I'm saying it in kind of, let's say simple terms, but in terms that are nonetheless very correct. The difference between these new models and what we were having before is that these are generalist models, right? So before the models, if you're thinking about the big waves of, kind of predictive AI, which was 2017, you know, and so on, '16, '17, where people were investing heavily in predictive AI, the models that were being built in that kind of previous wave, they were specific for particular purposes and they understood particular domains well and they could do a particular thing. These models are generative, you know, these generative, the foundation of generative AI on LLMs, these LLMs are generalists, right? They are trained on vast data sets. They are sometimes billions or even trillions of parameters and they contain vast amounts of the entire knowledge that's out there in the world. And they're sometimes even, in many cases, in most of these cases actually, the new ones, they're actually multimodal. They can switch between text and image and different modalities basically, right? Multipurpose, multidisciplinary, multimodal. They can solve many different problems for you. Like they're just kind of a generalist tool that can solve so many problems for you. But are these already, are these tools already what you should be putting to use in an enterprise context? Yes or no? Do you need to do lots of work? Is there a whole other stack that you need to build around it or can you just take these generalist tools? Those are questions that people I think are having right now and the space is very, very exciting right now. I would say if you'd ask me like where we are, With regards to the generative AI explosion and its adoption within enterprise, is this the big kind of, you know, is this the top of the hype? Is this like the bubble? If you ask me, I think it's actually early on in the phase, in the development, in the cycle of adoption because there's still so much to figure out. There's so many different things things to figure out. There's so much experimentation going on. There's intense competition between all these different models. There is intense competition on who's going to capture us the sizable chunks that exist within the enterprise space of inefficiencies that could be solved and who's, with whose models are we going to do this and with which architecture and how will those different parts and architecture actually combine together. So, Every week something new is happening. Microsoft with Copilot, Google Workspace has just launched Duet AI, which integrates all across Workspace. You know, you can have generative AI within your Google Docs and your Google Meet, even inside Google Meet. You can have an assistant coming inside the Google Meet and listening to conversations for you. So it's some crazy stuff happening which people have actually not even fully figured out. How this will transform and change the workflows on a day-to-day basis within the enterprise, right? But one thing that we can see is everywhere in every industry, people are trying to find, people are playing around with it. So if you talk to people, if you talk to people in various different industries, it's very likely that you come across, if they're not yet now, very, very averse to technology or very conservative people kind of who are following late. The most, most people who are more open, younger people and so on, they're already using, they're already using, uh, the capabilities of LLMs, of ChatGPT and, and other LLMs that are out there, Bing and, and Bard and so on, that are public that people can use. They're already using them in various ways, either just personally or at work or a combination thereof. And this has happened obviously since the beginning of the year when these technologies became widely accessible. But where, what we see, the shift that we have seen after summer is that before summer, before summer '23, people were mostly doing their individual experimentations and everybody was experimenting with it individually. And so people we talked to at different companies, everybody was saying, yeah, this is crazy, we're trying this out, I wanna find out what it is. But there was not yet from top down, from the directors, from the exec, there wasn't yet the mission or the objective set to look, we need to figure this thing out. We need to figure out how to implement generative AI in our enterprise. Otherwise competitors will do it faster. They'll be much more efficient than us. And that has changed now in summer. So after this summer, when you, you know, once, I kind of sort of around, in my experience from customer development calls, it was sort of, you know, after this summer, July, August, as people came back from their summer breaks, everybody had it top of their agenda. Okay, we now are CIO, CTO, whatever, you know, we have now gotten the objectives, we need to figure out how to implement generative AI. We're now launching lots of different pilots. We're now experimenting officially and there is budgets for that too. Okay, so there is different applications. I was saying, where are the applications? Where that we're going to talk about that later in a bit. Where do you find useful applications and what are the challenges and pitfalls and how to go about it? But before I go into that, I wanna share something quickly with you, which is just an example. And this example is actually generated by Synthelligence, which is our product at Syntegrol, which is a tool that reads 10,000s of news sources and finds signals within those sources and then throws them into a processing mechanism where they are looked at for signals and those signals are then measured and ranked according to confidence levels and noteworthiness. And yesterday, actually, I was actually yesterday. I shouldn't say that, but it was actually yesterday. Yesterday I asked one of our lead engineers to, hey, can you just quickly ask the platform for me to give me some outputs what's happening in enterprise space or what's the latest in enterprise space? Give me what's noteworthy, right? And he just did this quickly. We don't do this usually. We usually do this for the sustainability space. We have lots of different very, very specialized signals there around lots of different things like compliance and green certifications and supply chain, circular economy, and so on and so forth. But here, what you can see here, he just did this quickly for the enterprise AI space. And as you can see, it says here, for instance, yeah, Salesforce. Salesforce is huge. They just made some big acquisitions. They're positioning themselves really, really well, obviously, with all of that. Legacy, you know, knowledge that they have and the positioning, the domain expertise that they have, that's a really strong position to act from to enter into the space. But also major consultancies, McKinsey, BCG, Bain, and so on, they're gearing up to really get strong and get into the space. Either they're building in-house solutions or they're acquiring, now starting to acquire startups and third-party service providers. Or a combination thereof. Okay, so not gonna dwell too much on this. Let's move ahead to sharing the challenges, the challenges and the opportunities. In the Gartner hype cycle, generative AI is sort of at the top right now, and that means that there's gonna be a trough of disillusionment coming. Even though I said before, in my mind, generative AI or general, you know, the adoption is still very, very early days. And this can both be true because even though there's gonna be a trough of disillusionment, you know, then there's gonna be a next wave and that next wave, you know, it will be a combination of generative AI with some other capabilities, right? And then we're gonna have the next round and the next round and the next round and so on. But we're still very far away from really seeing what the potentials are in adoption in enterprise, right? Because there's vast inefficiencies. Anybody of you who has been in a large enterprise, you will agree with me that there is vast inefficiencies in so many different parts of how work is organized still today, right? In logistics, in knowledge management, in HR, in so many different parts, right? And those are all opportunities basically. Those are all opportunities for, for entrepreneurs and intrapreneurs that can be tackled with the new tech stack that's there. So I'm going to talk about the challenges that we're seeing, that I'm specifically seeing with Syntegrol right now when we're talking to companies. And again, we're talking to companies in real estate, supply chain, asset management, and mostly to the consultants, consultancies who are servicing these industries because they are our kind of natural client because they're helping those firms be compliant, right? They're helping them with their compliance, their sustainability reporting, and so on and so forth. So one big thing that we're seeing is when it comes to, okay, you know, we're going to experiment with generative AI, we can now make this process faster, we can make it better. One big challenge that we're seeing is what happens to the humans who were doing this before, right? There are some people who are doing this right now. Maybe they're making the reports or They're making the analyses, they're making the, you know, they are creating the outputs which we're looking to automate right now, right? What happens to these people? Are they going to be replaced? Are we looking to replace them? Are we looking to create solutions that are working without humans entirely? So in my experience, that's not a good process. That's not a good idea to have. First of all, because the outputs, especially in complex domains, they're always going to be lacking if you try to ignore the human. If you try to kind of build a solution, you try to say like, look, give me all your data, we build a solution for you, and then you can just go and use that, you don't need those analysts anymore, right? That's not a good solution in my experience in complex domains. It could be that if you have intelligent document processing, automated document processing, some things that are very, very, where it's very clear what you need to do, right? Take this, read this, put this out there. Yes, that you could do that and you don't need necessarily the human there. But in complex domains, which is most domains that are value generating for enterprise today, in complex domains, you always gonna need the knowledge and the input of the human because there's a couple of reasons why. There's something called tacit knowledge. Tacit knowledge is non-codifiable information, right? It is kind of, what you just know how to do it, but if you ask somebody how they're doing it, they couldn't be able to tell you. That's why also it's hard to codify that knowledge, right? This whole thing between system 1, system 2 kind of thinking, if you look into psychology background, Kahneman and so on and so forth, right? So the question is not, so can I, that's the one thing, right? There is that human knowledge which is tacit and hard to codify, And you need that if you want good outputs, right? Things like value judgments, things like experiential things where you just need to have seen that situation. Yeah, it's really, yeah, you could say, okay, just feed me this X number of situations like this and I will recognize the pattern. Sure, but still there's situation, there's a lot of situations we find when we try to generate specifically for our reporting, sustainability reporting purposes, when we try to generate the reports without any human in the loop, the outputs are kind of, there could be, you know, there's something missing there which sometimes the human knowledge can nudge it in the right direction. And when you get that human input, it becomes really, really strong, the outputs, right? So therefore, and as the second point, it's not just that your results are better. The second point is, if you include the human in the process, you're going to have less resistances in the org. With adoption, right? So if you have done this internally in a company, you will know, right? If you are talking about building coalitions for inside the company to have adoption, and if you're trying to sell, it's the same thing. If you're trying to sell, you need champions inside the company. You're gonna have more champions if they're also thinking that they're on board, they're not being necessarily replaced, and so on and so forth. Although it depends sometimes on the dynamics. In some companies when they're outsourcing, these analysts, they're actually happy to cut down that cost, right? So that cost point, they don't need that necessarily. But yeah, it depends case by case. But generally speaking, having a human in the loop is best practice, what's evolving right now, in addition to the capacities of the LLM. So the new systems that are being built, they're consisting of LLMs and with some features that have a human in the loop. In our system, it is a sort of an augmentation of the results through a feedback loop where we have our user constantly give feedback. But you could also have things like reinforcement learning reinforcement learning through human feedback, RLHF, or other kind of types of more really advanced forms of improving the outcomes constantly through the human reward function and the human feedback. And that you can build that on top of the LLM, right, with regards to your application stack. So that's one thing. The next thing that we're seeing, the next thing is accuracy. This is a really, really big one. This is a really big one, accuracy. Because obviously, as you know, even though GPT-4 is now better than GPT-3, GPT-3.5 and even the, you know, some of the other models, you know, there's been some testing and seeing like, you know, some of them are getting quite good at avoiding hallucinations, but nonetheless, you still have hallucinations, it's still a problem, right? And hallucinations being a problem, it's not such a big problem if you're using it for yourself or if you're using it to find out something cool for yourself or if you're, you know, if you want to try and whatever, get some inspiration for writing an essay or whatever, it's not such a big problem, right? But you could just check again and again and so on, you could just make sure. But it's gonna be a problem if you're using that for a domain where you need highly accurate outputs, right? Where it's absolutely mission critical. If you're trying to do compliance reporting, right, which we're doing, you can't have a system that is sort of 85% accurate, right? That's not gonna fly, that's not gonna fly. You're gonna do a pilot and then that's it. You're never gonna go anywhere else from that, right? So, and actually here we have a problem. The problem is that it's actually, you know, it's easy to use the tech stack that's there right now with LLMs that are there right now to build a solution that's 80% accurate. That's not so difficult. To build a solution that's 99% accurate is very difficult. It's exponentially more difficult to get that incremental increase in accuracy. Right? So if you wanna go from 90 to 99 or 90 to 95, it's not just 5% more difficult, right? It's like 50%, 100% more difficult than doing, than staying at that lower level of accuracy, right? So, and there's a lot of things that you can do, you know, there is, and this is, look, I don't wanna get too much into the philosophy part of it and the kind of the theory side of it, which I always love to do, But you know, I need to do that elsewhere. And then, you know, as you know, Daniel was saying in the beginning, I got a book coming up, I do some talks and so on, some speaking on futurism and things like that. But just this much, yeah, it is a little bit, it lies in the nature of the technology because LLMs are about approximations. They're probabilistic and they see knowledge as continuous. It means knowledge is like in this continuous spectrum. The way that the LLM works is that it just kind of approximates something in that, close enough. It just wants to be close enough to the output, to where you need to go, right? And that's good. That's good if you need to have something that sounds good, that's great. But if you need accurate outputs, real accurate outputs, it could be a problem. So what can you do? Well, you need to, and this is a huge question, there's lots of people who are trying to solve this in different ways. We are solving this by combining the capacities of the LLM with the capacities of knowledge bases, knowledge bases, ontologies, these kind of things which are made for domain-specific exact knowledge. So they have discrete knowledge structures, not continuous, right? So like if you have a knowledge base, especially if it's structured like RDF or this kind of discrete structures, then you're going to have exact outputs. You're going to say like, okay, this is this, this is not this, right? And it needs to always check with that to make sure that the outputs are exact, right? So we have something like that. And I can't get too much into the details of that. My CTO always tells me to, yeah, like keep it, you know, don't talk too much about it. But yeah, we have something like that, which is basically our way to ensure the accuracy of outputs in the standards. Compliance, sustainability standards, and so on and so forth. But there's different ways to tackle this. It's a big problem, right? It's a big challenge, how to get these systems to be more accurate. Now the third thing is scope. And I don't know why I put this, so many words on here, but it's just scope really, right? So scope, and this is what, what is this about? This, the question here is that how do we get the solutions that we're building on top of the LLMs to read these large amounts of data that we need? Very large amounts of data because in the enterprise, you know, if you, you know, if yourself, if you have looked inside the enterprise, you have all of these different parts of information, all of these different departments and functions which all work with their own knowledge base or which all work with their own sets of documents, their archives, repositories, and so on and so forth. And if you build punctual solutions, if you go in the enterprise and you build a solution with, LLM that you make functional, that you integrate for that particular department, you run the risk of creating these, perpetuating these data silos just by making the silo smarter somehow. So what you need to do is for most use cases where we need to read, where we have vast amounts of data, we need to be able to read Imagine you want to parse and embed 200,000 PDFs, right? And that are based all across, spread across the enterprise, right? Or you need to look at external data, which is new external data, which is coming in on a fast basis, like real-time data providers, or maybe they have some IoT systems that are providing some reports that you need to know and so on, right? So how do you do that? You need to build an architecture that can provide you low latency, access to large amounts of data. In our case, we're doing this with a vector database in a particular— the way that the database is set up. But this is a problem. This is going to be a problem in any application that can actually function within the enterprise to tackle these big problems that exist where there's big inefficiencies and therefore big opportunities to solve problems and, you know, and have those money— have those money flows around those problems that you solve, right? So this is gonna be a problem. You need to be able to tackle, to work with large amounts of data. So all of this together, put all of this together what I just said, and you will see that this, what you're seeing here, this is a diagram that was published by Andreessen Horowitz just a couple of months ago. I think this is super cool and super interesting. And it's called Emerging Architectures for LLM Applications. I'm not sure if you can see this right now, but you can just go and Google it. It's all over the, you know, it's easy to find. And here, Andreessen Horowitz, you know, they present this, what they see as the emerging LLM application stack. And this is actually very similar to what we're doing in many ways. They have here a vector database. They have a, uh, you know, they have cert, you know, they have a kind of what they call an orchestrator in the middle. They have APIs for external data. They have Various layers on, you know, inside this, various layers and parts within this kind of architecture that they're seeing. The whole idea is here that the LLM itself, it's a model. The model is not already the application, right? So you're looking at building applications which utilize these incredibly powerful models, the LLMs, but by integrating them into an entire stack. Right, so this is what these guys are seeing, what Andreessen Horowitz is seeing as the future of usable applications that can have impact in the space. I thought it's interesting. Obviously, again, this is an emerging space. There's many different ways to do this. Very, very exciting space right now to be in. Okay, I think, I don't know how I'm doing for time, but I wanted to still share something at the end. I don't know, I didn't even check. I don't know if it was 20, 30 minutes, but whatever it was, I think as we're drawing to an end for this conversation, I just want to share at the end because I'm very interested myself in futurism and in scenarios and kind of the philosophical side of how, what does this technology mean? Where is this going to go to? You know, how are our enterprise decisions right now impacting what future scenarios look like? And I think, I actually think that business leaders and enterprise leaders, CEOs, C-level, XX, or those who want to be XX, they should be too, and most of them actually are, the good ones are, they are interested in this, because they have larger horizon visions, you know, they're not just looking at what's the problem I need to solve right now, you know, what's my end of year, no, they're also looking at where are we going to, where is the whole space going to, and you can only answer that by seeing where is all of technology going to, where is the whole economy going to, you know, you need to ask these kind of questions. So again, as I said before, I think there's gonna be a trough of disillusionment setting in with the capacities of LLM, obviously after the initial huge hype. And the disillusionment will be around those topics that I mentioned, maybe others, but those definitely, right? So the accuracy, the lack of human, intelligent human interaction, Right? And when this disillusionment happens, what happens is that there's also kind of a dialectical answer and you get the hype on the other side. You know, so one thing that is, I think, very likely is that the next wave is gonna be symbolic. Very strong, you know, very strong. I would be very bullish on symbolic AI. So symbolic AI is sort of the other side of the spectrum, right? LLM is the side of the connectionist, probabilistic type of thinking and symbolic is the type that is structured around concepts and reasoning, right? So, there's already some great experiments there coming from neuro-symbolic AI and so on. So, that would be one thing. And then it's interestingly also, you could also say that these challenges that we're seeing right now in the enterprise space, they are actually parts of this larger dialectic between these different drives of technology socially and economically as we're moving towards the next stage. And the next stage, what is it? Is AGI, right? AGI, if you know, if you've, and if you should be familiar with the term artificial general intelligence, it's too early to say that already ChatGPT is AGI. Obviously, it's not even though it's passing the Turing test, even though, right, in most contexts you can't tell whether you're talking to a human or not, which is the Turing test. But it's not AGI obviously, right? It's not AGI. There's still a lot lacking there. And what is lacking? Lots of it is about the things that we talked about. So integrating human, once you integrate human expertise, if you integrate human expertise, you will actually have a pathway towards one particular type, one particular type of pathway towards AGI. So symbolic symbiotic human-machine intelligence integration. There's— this is going to connect with neurotech. It's very interesting to note as well, Neuralink just launched trials this year with humans, is launching trials with humans this year. So we're starting to connect brains directly to the net. This is happening. This is not in some kind of distant future. No, this is happening. This is now, right? A whole brain simulation. And other paradigms in the space which are leading to this kind of human-AI integration, that's one pathway to AGI, right? And the second one, when I talked about mapping exact domain knowledge, having basically these ontologies where you really have exact symbolic knowledge of one domain, this taps into one proposal for AGI which is called Oracle, the Oracle AGI. So this is something proposed by Nick Bostrom, for instance, where you have the idea that we create these systems that know everything about a particular domain, And they're set up like an oracle. You can question them about that domain, right? They know everything about that domain, but they can answer you. They can have full and complete reasoning capacities about that domain. And the last point about holistic integration, when I talked about scope. So this is part of a movement where we're talking about really connecting all the intelligence and all the data on the planet eventually, right? So this is gonna create something which Peter Diamandis, from Singularity University, I'm sure you must know, he's called meta-intelligence. So meta-intelligence is this idea that all the data flows and all the types of both computational, machine and human intelligence are increasingly gonna get interlocked and interconnected, and you're gonna have this emergence of this kind of higher-level meta-intelligence, right? So this is just something that AI researcher and pioneer Peter Russell, he called the global brain, years ago already, and I think we're moving towards that. And some of these, again, some of these experiments that are now happening in enterprise AI adoption, interestingly, even though people are not aware of it necessarily, they're contributing to these kind of next pathways. Okay, I'm gonna stop it here because I know this is a lot and all of this could be its own talk for itself. And we could continue this some other time, but just this much that if you are interested in these things that I was sharing, I invite you to connect with me and to follow me on social media. Just send me connection requests. I actually have podcasts coming up. I just launched again, or relaunched again, a conversation series. It's just me talking to people basically because I'm just, you know, I just have fun doing it. That's all. It's not, there's no, I'm not trying to, you know, become a Joe Rogan or anything. I just, I like talking to people, that's all. And I have some podcasts coming up which I just relaunched with some of the world's leading AI scientists. So there is— or AI thought leaders rather— Seth Early, who is one of the founders of ontology-based knowledge in the enterprise, Andrew Gill, who's a 5-time TEDx speaker, Bart De Wit, a founder of Hippo AI, Charles Martin, one of the leading tech and AI influencers. So I got conversations coming up with these guys where we're just going to be talking about some of the stuff I talk about now And if you wanna be in the loop with this, follow me on, or add me on LinkedIn. Don't follow me, add me, connect with me on LinkedIn. I got a YouTube too that I just launched again, as I said, with these kind of conversations. And I'm being very happy to, again, I love having these conversations. If you wanna talk about actual business and about, you know, what I was talking about with our company, with my company, Cintegrol, then we can have that conversation too. But if you're just interested in tech and the future of technology, the future of humanity, I'm always happy to connect as well. Thank you, guys. I don't know if I'm waiting for our host to return back here and just to share what's happening next, but yeah, this was cool. Thank you.

Awesome. Thank you so much, Dr. Vijak. I think you can see the emojis on your screen. The folks loved it. It's always a super, super interesting conversation to have about AI and more so with you because we've had you here before and last time around it was just as much fun. We have a ton of questions here and they're very interesting as well. But what I do want to do is I just want to spend a tiny bit of time going through the questions because I think the folks would want to love to talk to you one-on-one as well in the networking. So I'm just gonna quickly run through a couple of questions. Yeah, so we have, let me just share this, right? So let me just bring this question. We have Thomas who asks, how do you overcome initial skepticism about the benefits of generative AI in a B2B setting, right? And then we have another question which kind of talks about, which is asking about a similar thing, but that's also, it takes into account the high costs when it comes to implementing a Quip in that room. If you could maybe address both of those things in your answer.

So I think the great questions, and this is very practical questions, which are objections that you need to overcome if you're selling any product in the space, right? That's something that you need to overcome. And the way to do that is to have clearly defined proof of value experiments in the enterprise, right? So initially you're gonna find somebody who is open for this, who sees that, who's gonna be your champion or the person who places this with you, right? And you're gonna have several people who are not open to this for whatever reasons, right? Maybe they wanna build their own internal solution. Maybe they're just always conservative, right? They're just always closed, they're just conservative personality, right? Maybe they're just like whatever it is, all the fearmongering that's going on in the media about AI and so on. So maybe they're just impacted by that. Whatever it is, the way to tackle that is to get inside and instead of selling the big solution which will solve everybody's problems, you're like, look, let's define a clear experiment, right? And in this experiment, you need to define the right KPIs. You need to define the right KPIs. How long does it take you guys to do this currently? Right? How long does it take you guys to do this? Can we define a KPI? Can we do an experiment to then show you so we have a benchmark and we can decide based on that benchmark, we could see what your ROI would be if you do a full solution, right? So for instance, we went into one of our clients, we built them a solution where they didn't know, they had some people who were like, "I don't know what this is. I don't know if this is any good for us." and the other one was like, "No, I love it. I got a good feeling. We need to try this out." And they had no idea. We built them a solution and afterward was like, "Okay, we're spending a million euros a year on these type of compliance analysis and it seems like if this thing would work, it would save us 70% of that." Right? So, okay. Like that's sizable. Now nobody's going to say, "Oh, is this valuable for us? Yes or no?" Right? It's going to be very clear. It's going to be like, "Okay, how do we get this thing?" implemented, where are the hurdles, right? How do we get this implemented now? How do we work out the kinks? So again, I need to tell you, to person who's asked about initial skepticism, go and find the places in the enterprise where there is opportunities for efficiency gains. Where is opportunities for efficiency gains? Where people are doing tedious, repetitive work that takes time. Tedious, repetitive work that takes time. What is their billable hours of these people doing that work, right? Add it up. You're gonna know, okay, there's this many people that are doing this work, right? Can I make that faster? Can I create efficiencies for that? Yes? Okay, if so, let's define a proof of value. Let's define a proof of value, and then you can have the ROI calculation, and then you can overcome any type of skepticism. And this answers the other question I think too about the cost.

Yeah, thank you for that answer. So we have another question, and this— I feel like there's some intersection here with, uh, your answer. Uh, we have this question from Deepti. She asked, uh, what would be the biggest challenge in successfully implementing GAi systems in enterprise? Data being the main challenge, what can be done to overcome this? And I kind of want you to maybe talk about some unforeseen challenges that as someone who's ventured into AI for the first time might encounter?

That somebody might encounter who would?

As somebody who works in a large organization, right? I want to implement AI. So just kind of based on this question, what are some of these challenges that I might encounter, right? Something that I should be mindful of.

Well, look, if you're— I think if you're in an organization right now, if you're an intrapreneur, there is going to be vast opportunities for you, you know, vast opportunities for you, because you know, as a person who has been in the organization, you know a lot of the— you know how things work in the organization, you know where there are inefficiencies, and you know who's going to benefit from taking away these inefficiencies and who will not benefit. There's always going to be somebody who doesn't benefit from reducing those inefficiencies, right? And you need to maneuver around them somehow. You can't go to them with your proposal. You need to somehow maneuver around them or get them part of your coalition or somehow maneuver around them, right? So you will know that, which is you have a great advantage over anybody who's an outside third-party salesperson who's coming there, who doesn't know that. Maybe he can feel a little bit in the conversation, but he will not know the internal background. He will not know who will benefit from this, who will not benefit. So you have that benefit. Now, and you said the biggest challenge, I would say because the biggest challenge is to, like in any other digital transformation, sell, right? You have to identify your winning coalition. You have to identify your alliance within, you have to identify who in this company benefits from this, who will be able to sign off on it, and who will be able to be, you know, who would probably stand in the way of me doing this right now. This is just generally speaking. This is just generally speaking. But just more, and now more specific to generative AI systems, I think one of the big challenges is gonna be around evolving legislation. So evolving legislation, it might be, because right now it's a wild west, so there's not much legislation. We don't know if OpenAI and the other LLM companies, if they're violating intellectual property rights by just reading everything that's on the web. We don't know if the fact that you put their data, even if you have a GTP Enterprise or even if you have models that Microsoft is building something similar inside their cloud environment and so on, you don't know if it will be compliant with the new legislations that are coming out, especially in jurisdictions like the EU, which are more strict. The UK is making new legislation around it right now. That's, I think, is gonna be a hurdle. So you need to be wary of that a little bit. Where does the data go? Where's the, you know, who's gonna, where is the data going to be and how secure is it and how much can I guarantee that this is gonna continue to be compliant, whatever solution it is that I'm building right now? I think that's a challenge that's good to think about early on already.

Awesome. Thank you for that answer. What I want to do is I want to leave 12-15 minutes for the networking, but just to close things off, because FIB primarily deal with startups despite this conversation being about enterprise AI, we've had a ton of people asking questions as well. How should you as a startup founder think about the role of AI, right? Particularly, I just kind of wanted to pick out the things that you talked about where there's this new horizon, right? We have AGI coming in. There's, you know, a ton of velocity here, right? So as a startup founder, I'm just kind of getting my feet wet here. How should I think about it?

As a startup founder?

Yeah.

So look, as a startup founder, I think it's probably, you know, I mean, it's the best time ever to be a startup founder. Right? Uh, on the one hand, I mean, the general economy, recession and so on and so forth, is not necessarily— it's not so rosy. The outlooks for somebody who wants to go and work in a, you know, in a corporate kind of career path, the outlooks were much better in the past, let's say. Let's say this way. And on the other hand, as an entrepreneur, you're going to have so many tools at your disposal right now, right? So you can have, with very minimal I mean, you need to be smart, you need to learn how to use the entire stack that's out there, right? You need to be extremely smart, driven, ready to learn all the time, ready to experiment all the time, it's not easy, it's actually not easy from a personal development type perspective, like you're getting yourself to have that motivation and that focus, right? But if you can do that, you have so many tools at your disposal, you can create like decks and websites and, campaigns and email campaigns, right? And all of this, you know, research people, contact them, speak to them, test your hypothesis, build a no-code MVP, right? You can do all of these things so quickly right now, which you couldn't do before. You could get some validation. You got some validation, you can get people behind it, other people. As soon as you have validation and you can tell the story around it, you can get other people to come behind it, right? Whether it's collaborators or investors or whoever else it is, right? I think there's a lot of opportunities, uh, and if you're, if you're an entrepreneur today, you need to be, you know, AI needs to be your, your, your prosthesis almost. It needs to be like, you know, you need to plug it into yourself as much as you can.

Awesome. Yeah, so folks, there you have it. Um, if you're a startup founder, AI needs to be a part of your world. Uh, with that, thank you so much, Dr. Vizjak, for taking out the time to be here. Again, folks, you can check Dr. Vijayak's info out in the comment section. Akshay from our team will have all his info there. We'll also be sending you an email in 48 hours with the recording as well as the presentation and Dr. Vijayak's details. Before we, you know, end things and go to the networking, Dr. Vijayak, any parting words?

That was great. It's always fun to be here again. I would, you know, I would enjoy even more to have some different people sharing their ideas, so I look forward to going to the chat section afterward, and everybody, if you want to connect with me, send me a connection request on LinkedIn. That's the best way.

Awesome, awesome. So folks, before we close things out here, we have a very exciting event featuring the CEO and co-founder of Founder Institute, Jonathan, on Wednesday. Akshay from our team will be posting the link in the chat. And if you want to, you know, keep an eye out for all the events that we have happening at FII, you can check us out at fi.co/events. With that, folks, thank you so much for taking out the time to be here. I hope you have an awesome day.

Guys, shout out to Daniel and Akshay for organizing this. Thanks, thanks a lot. It was great, great work from your side too.

Awesome, thank you.

Thank you.