Ep. 55 AI Hype Versus Real Value with Artur Kiulian

00:00 — Artur's Journey Into AI

Matt Pacheco
Hello everyone and welcome to Cloud Currents, a podcast that discusses cloud computing, cybersecurity, emerging tech and how they affect businesses across all industries and all sizes. I'm your host Matt Pacheco and I lead the content marketing team at tierpoint where I help businesses understand cloud and security trends and AI, of course, to better make decisions about their IT strategy. Today's guest, Artur Kiulian, the founder and CEO of Principal, a frontier AI research company based in San Francisco, has a really remarkable journey in tech that we've come across on the show and we really are excited to talk to him. He's Ukrainian born entrepreneur who's built over 65 startups and it's kind of pivoted into this AI driven humanitarian work. He's helped coordinate response systems during COVID We'll talk about that and the invasion of the Ukraine.

He's convinced Nvidia to donate supercomputers to wartime efforts and much more. He's the founder and CEO of Principal, as I mentioned before, where they build digital twins and scenario planning tools to help organizations stop reacting and start processing, preparing for what's coming. And he's also Forbes Technology Council member, an AWS partner, and someone with a really clear take on where hype ends in AI and where real business value begins. So we love talking about that on the show. How to tie AI to business, help our listeners make the right decisions. So we're really happy to have you,.

Matt Pacheco
Welcome to Cloud Currents.

Artur Kiulian
Thanks for having me. Thank you. Like so many, you know, accolades and accomplishments, obviously I, I've done a lot of work, but sometimes I, I listen to it. I'm like, man, like I'm just 34 years old, you know, when did it happen?

Matt Pacheco
Yeah, when you put in a lot of work, you do a lot of things and it's good to hear it back sometimes. Definitely. So you've done a lot of things. Let's talk a little bit about that. Let's talk about your journey in tech, where you started and where you are right now.

Artur Kiulian
Yeah, so I have a degree in Systems of Artificial Intelligence, originally from Ukraine, and I kind of got bored out of coding very early on, primarily because I worked full time during my university days. And then when I graduated I was like, that's it. I was already senior developer managing people, and I was craving for something more meaningful. And this is when I learned about starting startups and entrepreneurship and I was like, wow, there's this whole world out there. So I got into that and this is how I essentially moved from Ukraine to US and through my entrepreneurial journey I had a couple of startups. Obviously first time founders always fail, so learned a lot.

And then I ended up basically co founding a venture studio in Los Angeles and I ran it for six years with a whole thesis of almost like building a conveyor belt for startups so that I first of all could learn much faster and through other verticals, industries, applications. And it was just a beautiful time during that mobile app era where there was so much to build all those vertical applications and kind of like the boom of mobile app era. So I did it for six years, built a bunch of startups, had some exits, and then I wrote a book in 2017 called Robot is the Boss and it was kind of like a hilarious title, but the topic was Implications of AI in Business.

So I did that and I just reflected back to myself how crazy is the current world and how much AI is going to influence that. And since then, it basically kind of pushed me to think about social impact of AI and all the implications. And essentially it took me three years of pondering about it until the pandemic hit. And when the lockdown started, it was basically my excuse to actually do something about it because the whole world froze, the venture capital industry froze. And I saw this call to action from White House Technology Office basically calling to the AI community to help navigate uncertainty about COVID pandemic, both on the policy and medical research side.

And I saw how respectfully vague that was and I was like, wow, no one's going to understand what to do, so let me actually craft some structure out of it, because I have the superpower to take uncertainty structure. It's so people can actually become useful in it. So I did that and it immediately took off almost virally, grew to 1500 data scientists, AI engineers from all over the world, jumping into initiative and basically trying to build tools to help navigate the uncertainty. And we got into Wall Street Journal, into TV interviews, and that was my perfect excuse to leave venture capital and jump full steam ahead into this beautiful world of social impact of AI. So that's kind of like how my transition from VC world actually started.

Matt Pacheco
Very interesting. So you built many startups and you've been involved in many companies, a lot of initiatives. You just mentioned a few. What did that experience teach you about how technology gets adopted within these organizations?

Artur Kiulian
You know, 2020 was an interesting inflection point because it was almost like the culmination of some of AI technology finally becoming useful, meaning were one of the first organizations who got access to GPT3 back then. And it was still quite stupid, but we definitely saw the potential in it. And as I was progressing on that social impact infrastructure, Basically I spent 2021 pulling free resources, free compute, free infrastructure from technology companies like Google, Amazon, Slack, zapier, webflow, all of them to give us infrastructure pieces to be able to launch initiatives very fast once the next crisis happens, basically to be prepared for any future that actually arrives. And it was definitely a good timing because technology was getting there. AI models are becoming better. End of 2021 was actually quite influential on the AI, let's say market timeline.

07:00 — AI in Crisis Response

Artur Kiulian
And then, you know, as were preparing for many, you know, let's call it unfortunate shitty futures to arrive, one of them actually arrived in my home country. So when the full scale invasion happened, were prepared, we had the infrastructure and we rolled it out in the first hours of it. And it was technology as unfair advantage to be able to do that because not only we had the key components, we also used AI as a decision support system to run the most critical operations. We started from evacuations that went viral and people were sharing thank yous online. And basically that scaled our evacuation request to tens of thousands in a matter of days, you know, adding up to thousand drivers on the other side who were willing to go into war zone. Like there's no way human can manage that, like literally no way.

So we started using AI triaging decisions and all of these things. And it kept bubbling up to other use cases like humanitarian aid, housing support, all the way to working with governments on high stakes decision making. And this is where those famous supercomputers came from, Nvidia. And we kept basically building more of the decision support systems. So what it taught me in terms of the key learning was you never have the perfect infrastructure, but once you have something, you are able to very quickly use it and adapt and create that feedback loop. This is where the real magic is. So don't wait for the perfect start using something, but also don't ever overestimate the capabilities because that's dangerous to you.

Matt Pacheco
That's really interesting, kind of that using AI and emerging tech in crisis, government and certain initiatives. You just talked about working with Nvidia for donating supercomputers to the effort. How do you actually make that case to a company focused on such a booming commercial market? That's quite a feat.

Artur Kiulian
Well, it was an impossible thing, right? Highly improbable event. And sometimes I joke to people that I'm in the Business of low probability events and preparing for those is one thing, but actually designing those to happen is even harder. So the way I think about this is that everything is possible given certain constraints and resources. So your job as a decision maker is just to paint the outcome, the end outcome, where you want to be paint the starting point, the current state, and use technology to help you navigate all that complexity, how to actually end up in that successful outcome. And it's not about predicting the future, because a lot of people think that I'm in the business of predicting the future because of the current work that I'm doing.

But essentially I'm in the business of action, being able to see and model which futures we're arriving at and how to act on it, so you can still get to the outcome that you're planning to. So my current work, I'm CEO and founder of Principle and what we do is basically scenario planning and simulations to help businesses get to their desired outcomes. And once again, a lot of people think that I'm selling a crystal ball, telling them, oh, you know, something will happen and that's why you should prepare for it. And it's not what I'm selling at all, which is counterintuitive because a lot of companies in Silicon Valley are actively selling, you know, crystal balls and prediction of market events. The whole prediction market thing is booming like crazy. You know, the whole gambling on the future, it's like all over the place.

So it natural to feel pressured to buy crystal balls. But the way I talk about this is that the key capability for the modern enterprise or even human decision makers is being aware of the environment, understanding what is happening and how it helps you or prevents you from getting to your outcomes. And this is where technology is magically capable of doing it. Because all of a sudden you can simulate thousands of pluralities of the future and you don't sample it for the highest probability event so you can predict the outcome. You're actually very interested in highly improbable events. Because if you think about it, invasion of Ukraine was a highly improbable event, at least for Europeans, let's say for Ukrainians it wasn't highly improbable event, but for the rest of the world primarily it was.

And if you're not having a plan towards those Black Swan doomsday scenarios, the moment it arrives a very improbable event, you're done, your whole business is crumbling, your supply chain infrastructure, it's all going to crumble like it crumbled during COVID which was A highly improbable event that no one prepared for. So that's kind of like the showcase of how technology helps us nowadays.

Matt Pacheco
Yeah. Wow. You've done a lot of work in these kind of highly unprobable events. Hopefully they stop happening. But I'm curious, how did your work in kind of this government and crisis settings between the invasion of Ukraine and Covid, how did that inform some of what principle does today and what did you learn about decisions you're making under pressure and a lot of the instances that you can now bring to the corporate world?

Artur Kiulian
Yeah, it's very interesting because the higher you go in the high stakes decision metrics, the higher are the risks and the actual resources and constraints for you to work with. And it was quite fascinating to start the work within the government because we started as a social impact initiative. It was a blessed time, unfortunate timing when there was a lot of support for Ukraine and there was a lot of funding, there was a lot of compute, there was a lot of free infrastructure and it rapidly started changing. End of 2023, support for Ukraine was dropping and then AI market was taking off, meaning all of these, you know, Nvidia and others were basically utilizing every single GPU that they could, you know, get a hold of. So all of a sudden were left with no resources to actually do something.

And governments have lots of resources, except they have their own priorities. So sometimes this kind of scenario planning work is not a priority, especially if you're dealing with an actual full scale war, for example, or to give another example, in 2025 we spent time primarily in the Middle east helping governments with national AI strategies. And we got invited by Dubai Future foundation to deploy AI policy decision making simulations for Dubai government. And we'd done it for healthcare and kind of like behavior analysis there. And we pitched potential scenario planning for other high stakes decisions, but it was not priority because it was highly improbable that Iran war could break out. And guess what? Two, three months later, since our presence there, it happened. Same shahed drones were flying over Dubai as they did in Ukraine.

So what it taught me is basically that given the resource constraints, there's always almost like the hesitance to dedicate it to planning towards low probability events. They always get discarded and it always bites you back. So that's kind of like the government work that transitioned into the corporate strategy work that I'm currently focusing on.

Matt Pacheco
I mean, there's definitely a lesson in there too, like for a lot of our listeners who are in the IT industry or in it, trying to plan strategies for their companies and there's, it's very interesting kind of planning for those low probability events. There could be a lesson in there. So you talked a lot, a little bit about AI and there's certainly we're in kind of this reality where it's like, hey, is AI hype or is it bringing actual value? Some people argue both ways all day. But today I kind of interested to get your take on hype versus value when it comes to AI and kind of how it relates to business. Because I know a lot of companies are being mandated in some instances to adopt new AI to stay competitive for new products, new features.

16:27 — AI Hype Versus Reality

Matt Pacheco
So I got a few questions for you about that. One of them is a really big question we constantly get. You've been outspoken about the gap between what AI vendors promise and what companies actually experience. Where do you see the biggest disconnect right now between the two?

Artur Kiulian
I would say the biggest disconnect comes from the fact that AI actually works, but it's been hyped up to such heights that it's almost impossible to decipher what's real and what's not. Especially when AI penetrates the actual executive decision making and has a yes man type of effect where you start using Claude and you start asking it, do I need 100 people in my department that does this and that or can AI do that? And very quickly AI spins up a lot of sub agents, goes on a hunt for reports of consultants and all of that and it gives you the report. You know, it's estimated that the work that they're doing is going to be automatable in six months from now. So you got to start preparing, laying off all those people. And this is what we're actually seeing in the market.

There's hundreds of thousands of people being laid off based on the expectations that I can deliver on the same level of the kind of work scope that people are doing right now. And you know, who knows if it will actually happen. We're seeing AI becoming so much better that it's entirely plausible. But the human in the loop is a critical constraint. If we start removing that, we'll actually fail to assess whether AI is successful with that. So I think that's the biggest kind of gap right now. People's awareness and deciphering the reality from the hype.

Matt Pacheco
All I know is chat GPT says I'm the smartest person ever all the time. Makes me feel great. Genius. Yeah, it's hyping the AI hypes you as you hype AI. It's, it's really funny. But you made a point, a good point, about how LLMs can be kind of sycophantic. They confirm your biases for business leaders using chat, GPT or Claude or any of the LLMs for kind of strategic thinking. What's some advice you would give there to avoid some of those risks that you mentioned?

Artur Kiulian
So a lot of people, when I pitch principle as AI simulation engine, they're like, oh, but I can just use Claude and ask it to simulate what's going to happen in my industry. And of course you can. And Claude will give you an amazing showcase. You know, same with Copilot, same with Gemini, Grok. They will all tell you this is what's going to happen and this is what you should be doing to prepare for it. The problem is when you do it with a generic LLM, even if you feed it all your company context, even if you connect it to all the drives and slack and everything, it will still be a sycophantic yesmin.

It will reconform your biases, reconfirm the biases of people who created those documents and instead of helping you navigate the uncertainty, it will just collapse it to the most believable thing that you think will happen. So the best way to do it still is to triangulate that with multiple stakeholders. And that's why humans are actually quite useful, right? We used to rely on consultants to do this kind of work. You invite a third party like McKinsey or you know, KPMG, Anderson, Deloitte, EY, all of these, they assess, they triangulate feedback from different decision makers grounded in a market research reality and sometimes grounded in the competitors because sometimes they work with competitors of yours and they present something that is a triangulated, highly biased feedback, but very triangulated against those realities. LLMs cannot do that, unfortunately, because they're this singular stateless machine.

So the only way to go around it is to either still use human parties for it, except humans tend to move super slow. You know, you hire McKinsey, it's going to take three, six months to get to something realistic and workable. In those three months, the world will be reconfigured so fast that report will be outdated on delivery. So you have to build real time systems that actually model out all of the actors in the ecosystem, all the players, all the regulators, all the governments, everything that affects it, which is insane level of computation. So we decided to solve that Exact problem. And this is what we're rendering as an actual product to companies.

21:42 — Research on AI Bias

Matt Pacheco
Kind of awesome. You've also published a peer reviewed paper on the biases of AI market prediction. But I'm going too deep because I know papers get very detailed. Can you tell us what did your research find and why it matters for business leaders?

Artur Kiulian
So we, what we published was basically a research paper to highlight to decision makers that LLMs are highly influenced by the context that you feed into it and the actual training data that they had. So what we've done, we grounded the predictions about the world based on the poly market data. So, you know, prediction markets are pretty interesting platforms. I personally think their business models are quite predatory and they're not good for society. Even though the technology by itself is very useful because this is the first time we're able to see how certain events in the market propagates through belief networks and actually changes people's beliefs. Someone says something geopolitical news or something, it completely reshapes the landscape. Certain outcomes become much more probable, certain becomes less probable.

So for us as researchers that work with prediction of the future, let's call it that, we are able to use these platforms to actually showcase that it's impossible to predict the future, that LLMs are not the solutions to prediction of the future, but it's also useful to understand their limitations through that type of research. So that's what we've done. We published the actual paper. It just got into one of the biggest conferences called icml. So our CTO and chief AI scientists are flying to Korea to present that paper at the workshop, at the forecasting workshop actually. So it's going to be quite interesting because there's so many academics that are trying to predict the future and they truly believe that's possible and computationally possible. And we're on this little crusade of saying that what if that's impossible? What do we actually do?

So that's long story short.

Matt Pacheco
Cool. One question before I go into that because I want to talk about Principal platform and kind of scenario planning and all that. But one still on this whole general AI piece, when you're talking to businesses, what's your honest assessment of where agentic AI is today versus where it's being marketed and like is the cloud will run your whole company pitch realistic anytime soon? We keep hearing about this like one person, billion dollar company that will eventually exist using AI for everything. Is that even realistic?

Artur Kiulian
You know, it's quite funny because when we're talking with investors and you know, people in Silicon Valley. And we're telling them that we're scaling the team, the human team, they're like, why do you need so many people? Like, well, like you can just use AI for it. And I'm like, yeah, good luck sending AI agent to talk with, you know, a chief strategy officer at Enterprise and their whole team and explaining, you know, the product and everything around it. So I think there's a lot of expectations for what AI can do or should do. And in reality I think AI is amazing right now with code and actually building things, like so amazing. It's fascinating. When I was writing a book 10 years ago, I could never imagine living in a world where engineers are no longer the highest, scarcest resource.

And AGI can do that. That's the reality we're living in. Then if you think about other functions in the areas that are called unverifiable domains, where there's no specific calculated way to verify whether a solution works, AI is still not that great. It keeps bumping into different pathways and most often is wrong. It's still useful in combination with a human operator because human operator can highlight and reroute AI. So this is where AI is still struggling and probably will never be able to fully automate these kinds of domains. Happy to be wrong about it in five years from now, but I think the human agency will have to be preserved. And that's both the beauty and the real value of it. Because LLMs are still backward looking machines. If you think about it, there's nothing else that the compression of the past data.

Sure, it's a lot of data, it's a lot of physics of the world encoded in this like super compressed weight of weights of the model. But still it's a backward looking thing and it really helps to have someone who's forward looking and can reason about the future, which is humans.

26:59 — Prediction Versus Simulation

Matt Pacheco
Well put, I love that. Let's talk about scenario planning and the Principal platform. Let's start off with this question. A lot of people hear that AI predicts the future. You mentioned it before, you talked a little bit about it and are kind of skeptical and are rightfully so. You'd argue what's the difference between prediction and simulation? And why does the distinction matter?

Artur Kiulian
Great question. So when it comes to prediction, you're literally sampling a model on highest probability of a certain outcome happening or not happening. So literally probability like 90% this event will happen in the market. When it comes to simulation, what you're really designing it for is modeling of plurality of the futures. All of the Potential ways that the future may develop. So you're creating this cone of the future. By the way, this methodology has been developed in like 1950s at Stanford Research Institute, Air Force. Then Rand Corporation came in, and then Shell came in and introduced scenario planning as a way to use it for business. So the methodology is super old, except we didn't have tools to actually build these kinds of representations of the future.

So when you're simulating all of the actions in those many worlds under different constraints, under different what if scenarios, and then you make it live. And that's where that real magic start happening, because you have a thousand worlds with your specific strategy and then signals keep arriving. OpenAI releases something, OpenAI makes an acquisition of a startup, and then you see how that affects those futures. And maybe out of thousand futures there is, I don't know, 26 that resemble this kind of development. So you're adjusting in real time and you're seeing which feature we're converging. And most importantly, you're understanding how to act on it, which is an actual competitive, unfair advantage. So that's the difference between trying to predict and gamble on something and simulating it in real time.

Matt Pacheco
Wow, really interesting. Walk us through a real example. Like how does a company actually use principle to prepare for a scenario that might seem unlikely but still needs to be on their radar?

Artur Kiulian
I'll give you an example of a paying customer that started as basically answer to the question, hey, we're in a software business, traditional type of business, where we have a bunch of apps, and obviously this is being disrupted by AI. So now how do we actually transform ourselves so we're not stagnating, but actually winning in this crazy AI race. Whether we do M and A, whether we sell some of our apps, whether we actually hire equi, hire departments, actual AI companies to build AI capabilities internally. There's so many decisions to make, right? All the bets are quite plausible and all of them may get you to the final outcome that you're shooting for. But which one? So this is how we start simulating.

And as a case study, we ended up not doing M and A because we highlighted a certain direction where it was actually about a completely different way of navigating the future. And then it's all about adjusting that strategy in real time. So a great example is just recently Apple WWDC event happened which definitely reshaped the landscape. They released some interesting things that a lot of companies were afraid of because there were rumors, there were kind of like public statements about it. And the goal of using principle was not to predict what exactly that company will actually release. It was to create a strategy that survives all of those and when the future arrives, actually quickly calibrate against what actually arrived and move as fast as possible.

Because, you know, previously it would have taken them two, three weeks, you know, month to actually recalibrate, get everyone on the team aligned and actually start moving in that winning direction. Now you're able to do that in six hours. So that's kind of like the magical use case.

Matt Pacheco
Pretty cool. So you work with a lot of companies, you were talking about AI a lot. Back to the AI conversation. When a company says they want to go all in on AI, but they have no with the, I guess they don't have the talent to manage AI or they don't have the people on staff to do so, what's the first conversation you typically have with them?

32:09 — Preparing Organizations for AI

Artur Kiulian
A lot of people think that they can reinvent their organization like next week just spin up Claude code, you know, give the tools to everyone and all of a sudden we can vibe code platforms, we can vibe code solutions. All of the, you know, customer support will be replaced by AI. So this is where the original kind of like hype versus reality comes in. Because the reality is the fact, sure, you can accomplish all of these things in theory if your whole organization has resources for, has capacity, there are people that are knowledgeable how to use these tools and all of these components are in place. So it's not just having the tools, it's being able to execute on them and having capacity to do that. So that's kind of like the biggest, I would say, constraint that I see these days.

Everyone is afraid of what is called SAS apocalypse that is already happening where AI is eating up software as a service business model and basically everyone can by code, SaaS businesses, sure there is an effect that you know, happening in the market and obviously it's represented by stocks falling. And that also as a signal propagates, it creates push on investors that are dumping the stock of SaaS market and are basically funding other things in the market to bet on agentic vertical AI. So what is happening in the market is that it creates this ripple effect of things, like almost butterfly effect of many things happening at once.

Where big incumbents are trying to buy AI startups, we're seeing that like Salesforce just bought fin for like 3 billion, which is an exact representation of Salesforce being afraid of being incumbent and losing the functionality to all of these AI startups. So they're trying to grab those startups so they can create the capacity for themselves as a large enterprise to innovate. So it's less about even buying market share, it's actually about building capacity inside to innovate and be able to utilize these AI edge. So that's kind of like, you know, a very quick TLDR on my thought.

Matt Pacheco
Yeah, so you mentioned something interesting with a lot changing in the market. So AI companies being purchased by bigger companies, new AI startups popping up and a lot of companies are adopting this technology and some of these startups and these companies are going to eventually have to prove profit. And I'm very curious your thoughts on how organizations like IT leaders listening to this show can kind of prepare for some of those changes. How do you create a budget strategy that actually prepares for the changing pieces in the market? Because it's constantly changing. The technology is changing, but also the owners, the players are changing. How do you get some stability or change? Plan that out.

Artur Kiulian
It's funny that almost everyone in the market right now is experiencing turbulence, almost like motion sickness because we're flying so fast that everyone's just trying to grasp for some stability, something that they can understand in this current abundance of information. Like literally every single day there's so much happening. Like even for me, if I wouldn't have principle I would not know what is happening in the world. So it's a real information overload problem. The stability on the other side is almost impossible to accomplish because everything is turbulent, everything being reshaped. And in fact I think AI is the cause of that because all of a sudden it's much cheaper to produce software. It's almost like costs nothing to build apps these days, especially because it's subsidized by VCs, you know, venture capitalists and the market overall.

So that turbulence is self inflicted, which will not pause and stop unless we reevaluate our kind of like approach to the whole, you know, AI race. But let's assume that's not going to happen anytime soon because it doesn't look like we're stopping or pausing for a bit. So what do you do with that? You accept? I think acceptance is a very important part of it that is very hard fortune 500 especially to kind of like adapt to. You accept the inevitability of this, that everything's going to shape in a different reshape in a different form. Second thing you become much more aware of not just general AI applications, but you go super niche and super granular you go into specific business units and figure out what are the key functions that we can automate using AI and where it actually makes financial sense.

Because the conversation that no one is having. Well, sometimes people are having these. And I think the last three months has been quite an epiphany to the market where Uber Seal made a claim that they burned through the entire yearly annual budget of tokens in just four months. And they were like, whoa, what is this? Like, where's the roi? Actually, so it's very easy to get excited about. Hey, just burn more tokens and the value will show up. Except it's getting more expensive. Not because the compute is kind of like scarce, even though it's becoming scarce. Not because energy is scarce, even though it is scarce. And the data centers are not catching up to that. But the question is how good the AI models, how fast they are becoming and how capable they are in burning so much tokens.

You know, you get access to Fable, that famous model that got, you know, taken out from the market. It burns so many tokens, it's crazy. Like, we had access to it, we tried it, and, man, like, the unit economics on that thing. Like, not sure there is convergence on it. So, yeah, that's kind of like the biggest items that I would bring up to executives.

Matt Pacheco
You mentioned moving fast. That's. That's a real. That's a trend. Not in just one way, in many ways. There's a real tension between kind of this moving fast on AI and doing it thoughtfully. And you talked a little bit about this. But for the IT leader, the CTO listening right now, what's one mistake you see most often that they should avoid when it comes to AI?

Artur Kiulian
I would say moving fast is something that we once again should accept, mostly because if we are trying to preserve stability and kind of like fight for the past, we're going to be always losing because the world is moving into the future very fast. So for an average IT leader, the questions that they should be asking is that given the current environment, given the fact that everyone is moving so fast, am I moving fast into the right direction? Because if everyone is moving fast somewhere, I want to be the one who's moving in the right direction. Because if I'm moving super fast in the wrong direction, then my company is done.

If everyone in my company is moving in the wrong direction in many different wrong directions, we're going to burn all those tokens, end up nowhere, because we're not aligned on that outcome, that success state that we're hoping for. So that's the biggest roadblock I see right now is moving fast. It's how do we make sure we're navigating in the right direction?

40:30 — Predictions and Delusional Bets

Matt Pacheco
That's a great point. All right, we're getting to the fun part of the episode, the kind of looking ahead. I know you're a simulation, not a predictions guy, but I'm gonna ask you predictions anyway because that's the fun.

Artur Kiulian
As long as you accept the fact that they will be wrong, I'm totally cool.

Matt Pacheco
Okay, let's be wrong. So you described yourself in the past as still delusional when you're pushing the frontier, even while about its current limits of AI. What's the delusional bet you're making about where AI goes next?

Artur Kiulian
I think what we're seeing right now within organizations like entropic, like OpenAI and some of other big labs, is the fact that they started shipping really good products really fast and no one can figure out how. And I think this is going to be actually unique to every organization. Not because it's some magical AI harness and there's some rules about how to build software. It's because we're converging to a point where unique company culture is driving unique, almost like world models that are being built within organizations. And when you're all aligned and you utilize AI in a way to get you faster to where you all collectively want to be, it's going to be a magical spearhead towards those outcomes. And we're seeing that with IPO push, with Entropic and OpenAI, though, as an effect on the market.

It's going to have very negative effect because there's going to be more layoffs very rapidly. There will be a showcase to other companies that, hey, Entropic can ship it so fast and they don't have 50,000 engineers. Maybe we don't have those. We don't need to have those. So there's going to be this interesting kind of thing that is also, I think there was a guy running Citrini Research and he made a statement that we may converge into this almost like ghost gdp, displaced gdp, where economy doesn't catch up as fast as our AI capabilities.

And literally economy will start struggling because all those people that were kind of like, let's say high value employees, like engineers that were buying all the Teslas and all the fancy items, you know, the real middle class, they will be in a disadvantage and all of a sudden there won't be anyone who will be able to afford all Those Teslas and you know, other middle class items and there's going to be this weird, you know, effect in the market that almost like screen. If we imagine that future, and I don't want to live in that future because that's quite dystopian, like we should do our best not to end up in that future and hopefully we collectively create a better world model so we don't end up there.

But if that happens and we're kind of converging on it right now, then it will almost develop the need for a prescribed future, almost authoritarian way of just designing the life for humans so they can survive. Right. And this is where, you know, extreme delusion, you know, disclaimer comes in, where we're seeing LLMs being used across, you know, our daily lives. All of people are using ChatGPT in some daily decision making.

If you imagine that there is almost like a alignment of individuals towards certain objectives, you can see how AI becomes a prescribed future engine where humans almost lose agency because they have to, because they have no other options, because they don't have jobs, they don't have enough utilization in the market, they don't produce enough value because AI can absorbed and extracted all of their mundane work and started doing it and they didn't have enough time to retrain themselves. So it kind of sounds like a very dark future that I wish were not converging on, but this is something that we're seeing in the market right now.

45:16 — Preserving Human Agency

Matt Pacheco
That's very interesting. So given everything you've seen, what do you think is the most important thing AI can do for society that we're still under investing in?

Artur Kiulian
It's to preserve the human agency. It's not sure if you're seeing this, but the more companies try to automate certain things, the more they push certain systems on people. It's happening all over the place at all the big tech people are forced to use all kinds of agentic systems because they're being mined and extracted. Oh actually a couple of weeks ago there was this big news item about meta, basically forcing every single interaction of employees basically being recorded and then laying off people because they're gradually automating some of those functions. So if you think about preservation of the human agency, the more you rely on just automating, bluntly automating mundane activities, you lose agency. And this is where I hope we'll figure out how to create a balance to it and actually help individuals preserve more agency.

So we're not just having a headphone in our head and just blindly following all the instructions, what to do, where to go, and all of it. Because there was this TV show called Westworld. Not sure if you've seen it, but the last seasons were exactly about that type of dystopia where everything was managed by a machine. And even the main guy in the show running that machine had a headphone and he was just saying exactly what machine was telling him. So I hope we'll figure out that.

Matt Pacheco
Type of a problem. Yeah, I hope so too. So let's wrap it up. One last question for you. What's one piece of advice you'd give to a business leader who's genuinely try to figure out how to use AI strategically, not just chase hype. What is that one piece advice you.

Artur Kiulian
Would give strategically speaking? I would say start thinking about AI as your thought partner. Not something that helps you actually make decisions, because AI is great at expanding your decision space, your action space, helping you map out some blind spots that you haven't thought about. But the moment you start blindly relying on it, you're kind of collapsing it to the generalized capabilities of LLMs. Because remember, they're generalization machines when it comes to edge cases. They're not great because it's literally a normal distribution of things that it's being trained on. So I think strategically speaking, using AI as a thought partner to figure out the financial ROI of utilization of AI in business is the critical point. And this is probably the statement that most of Silicon Valley will hate me for. Because everyone just keep funding these things. There's no tomorrow.

And someday we're going to figure out the utility, the ROI will come, you just gotta burn. You just have to burn a little bit more tokens and then we'll figure it out. Meanwhile, no AI works out of the box. Like that's why OpenAI recently bought that organization with 150 forward deployment engineers. Because they realize that there's nothing that just like magically works. You still have to have humans to bridge that value of machine to the end applications, to find that roi, to bridge that context of real business application and the workflows to amazing LLM capabilities. And it's funny to see this dichotomy where entropic CEO goes on stage, says all the blue color, white color, all the work will be automated. And then they keep hiring so many people and they keep sending people to organizations to actually deliver value.

So that's what I would recommend to enterprises and decision makers to actually figure out that roi, where it maps to business processes and how to move fast, but in the right direction.

Matt Pacheco
Excellent advice. So I wanted to thank you, Arthur, for being on the episode with us and sharing your insights. This was a great conversation. We learned a whole lot from you. So thank you.

Artur Kiulian
Thanks for having me. And I'm looking forward to the follow up conversation.

Matt Pacheco
Awesome. Yeah, I'm sure we could talk about this for days and we'll have to regroup not too long in the future to talk about all the things that have changed ever since. So thank you and thank you.

Artur Kiulian
Yeah, the magical thing is that next week the whole AI space will look different. You know, we'll have plenty to chat about.

Matt Pacheco
I'm thinking probably by Friday of this week it'll probably be very different. But yeah, always a changing world. Always great conversations to be had. So thank you for being on the show with us and thank you for our guests for tuning in and listening to another episode of Cloud Currents where we talk about all these cool things in tech and how they affect businesses. So follow us wherever you get your podcasts, and we'll see you soon. But I'm Matt Pacheco, your host, and thanks for join.