Skip to content
PodSubs
ShowsPeopleGuestsJobsNewsEvents
More
BlogPodCred
Menu
ShowsPeopleGuestsJobsNewsEventsBlogPodCredNetworksSign in
Join nowSign in
  • Shows
  • People
  • Jobs
  • News
  • More
    • Guests
    • Events
    • Blog
    • PodCred
    • Networks
    Join nowSign in

Deep Learning with PolyAI

Deep Learning with PolyAI artwork

035: Artificial General Intelligence, six months later

May 30, 202428 minAudio

Damien and Nikola from PolyAI chat about the past six months of artificial general intelligence (AGI) progress. Key points of fiery discussion include the definition and goals of AGI, the recent advancements in multimodality in AI, and the significant progress in the industry over the past six months. They explore the exponential improvements in context windows, cost reductions, and the potential for AGI to equal or surpass human capabilities across multiple cognitive tasks. The conversation includes the normalization of AI in everyday life, regulatory challenges, and the potential societal impacts. They conclude with thoughts on the future of AGI and its integration into daily routines. Follow PolyAI on LinkedIn Watch this and other episodes of the Deep Learning pod on YouTube

Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.

Transcript

From the show. Tap a time to play from there.

Nikolawe need to educate ourselves a bit about statistics and just understand what it is we're using you know, electricity should be grounded, like rubber gloves are a good idea, or you might get electrocuted, right? That's something we teach children, or at least we ought to. Similarly, with using these models, A sharp knife can cut you, a stove can burn you, and, using an LLM that's freely speculating over something that hasn't been trained on is also dangerous

Damienhello friends and welcome to another episode of the deep learning with Polly AI podcast. The only podcast. That is going to be talking about artificial general intelligence today. That's probably not true. I probably just made a bold case lie, but that's okay with us. As per usual is CEO and co founder of PolyAI, Nikola Mrkšić. Nikola, welcome.

NikolaHi Damien. Thanks for having me. I'm not sure we're the only ones talking about it today, but we're probably in. Top 1000.

DamienIt'd be the most handsome talking about it today. I'm going to go ahead and stake that claim. We'll put that in a little stick.

NikolaYou're looking pretty rugged. I don't know if I'll take that.

DamienWe're going into a holiday weekend. I'll take that. But no, thanks for being here. It's been about six months since we last had an episode about artificial general intelligence. First of all, awesome. We've been doing this for at least six months. Incredible. Second of all, We committed, we

Nikolacommitted.

DamienIt really is. And we've been getting these things out weekly too. So, hopefully you're enjoying them, fair listeners or viewers. I know we enjoy putting them together, but it's definitely just this, the state of this technology in general, with regards to generative AI and advances therein. It's been leaps and bounds. It's exponential. So six months seems like a lifetime. And so I figure it would be worthwhile to go back and revisit the conversation a little bit and talk about AGI maybe how the stakes have moved a little bit, what has happened in the industry to get us closer to it in general. General thoughts around the concept of AGI.

DamienIs it still the North Star? So maybe to get started why don't you tell us a little bit about what the concept of artificial general intelligence is and where it's sticking point is in the general zeitgeist? I

Nikolamean, listen, I think at this point AGI is what you make of it, right? I think that Definitions vary. It can be that it does, that it outperforms humans in many different tasks, or that, we can go further and talk about the singularity and the ability to, self improve. Really, when you boil it down, it's something that really impresses us, right? And at some point we find both, that we're both in awe and maybe we're also a bit scared of. But, it's really just the pursuit of. Superintelligence, right? It's something that maybe even participates in our evolution as a civilization, sentient beings, I don't know.

NikolaYou can go all sorts of ways with this.

DamienWhat would you say, as far as the most widely accepted Clarifications or at least check boxes that a given model or a given system would need to meet to officially qualify itself to be that AGI. Like you said, there's the extremes, there's sentient self learning, self improving, et cetera. But what would you say is the baseline where someone could stick their hand in the air and say bingo. I think we've got AGI over here.

NikolaWell, look, I think the formal definition of the research thing would be something that matches or surpasses human capabilities. across a number of cognitive tasks, so, it could be language communication. It could be vision. It could be, listening. It could be other things. It could be solving math problems, And I don't think anyone said the clear number of things, but, in, in speech and communication I think you know what the Turing test where you're basically supposed to, I think, in half an hour of communication to not be able to tell if behind the screen there is a human or, an artificial intelligence.

NikolaSo yeah, there's a lot to it. And I think that with what's been happening recently with multimodality, I think we're getting closer and closer to people almost expecting the thing to be able to, Perceive the world the way we do. I think there are a lot of people working on robotics and, building foundational models for it, where, it's supposed to do other, maybe mechanical tasks as well as we do that, there's of course, like mastering cognitive tasks, like how well do you understand, but also can you solve math problems? Can you do other kinds of, maybe intellectual labor, can you write, can you read?

NikolaCan you see

Damienso it'd be safe to say that the finish line hasn't extended outwards, but it's gotten broader with a number of different tracks. Do you think that you mentioned multimodality? Obviously, with 4 coming out recently, and a focus on things like integrating visualization with, data absorption and as a channel for input. Do you think the multimodality is at least now seen as part of the canon of what a GI could become versus let's say that plus the mechanical aspect of having a physical manifestation?

DamienYou, you need to have the Terminator droid in order for it to be a GI Do you think at least multimodality is maybe in order for us,

Nikolatake it seriously, I don't know. Look, I think that I dunno if we need the mechanical aspect, right? I think that like multimodality really helps. The general perception of these things move past just like it being a chatbot. So I think it's phenomenal. I also think that when you think about that, like vast array of cognitive tasks where it could do better than humans, we already had computer vision being better than humans at, detecting patterns and things and stuff. Sure. Putting it all together, it's starting to feel like we are building something that is much more like us rather than just a tool that takes text in and produces text back out because we're more than text in and out.

NikolaRight? We are multi modal creatures.

DamienYeah, that's true. So I guess if we're going to say that AGI signifies meeting and is superseding or at least exceeding. Human capabilities, then we motor model has to be a part of it. What do you think has been just in general, what would you say has really progressed in the past six months? Obviously we talked about mobility and a lot of the new models are focusing on that. What else do you think has really progressed in the last six months as we like move towards this concept?

NikolaYeah. It feels exciting things happen. Like every month, I think, with Sora, with mostly coming out of open AI, Google releasing other things. It's very meaningful, right? When you think of Moore's law, it feels it's surpassing that level of improvement, right? Context windows have grown by 10 X in the past six months, right? Where, that whole, you Initial spiel of like, how are you going to do rag? And does it even matter if you have huge context windows? And I think, does it work as well? What does it really mean to use these things in what application?

NikolaI think that slowly people are working on deploying it in a bunch of places and learning which things are like phenomenal, superhuman, and not work as well. It may be unpredictable. So, the other obvious part is just cost. Speed, like every three, six months we get 2x reduction in price. And things are getting faster and faster. If you look at the demos of 4. 0 the voice mode is not at least in my chat GPT app. It's not out yet. It's still the old one, but I think that we'll see just how fast those things are.

NikolaAnd as the improvement speed, I think that it opens up, I think what's going to be really exciting maybe we should have done this podcast in a month or two when GPT 5 is released, because I think, we'll see just how much slower versus how much better it is and what we can use it for. So,

DamienHey, who says we can't? We'll be around. So

NikolaI think so. I think so. I think Terminator 1 gets us that fast, right?

DamienNot that fast. Not that fast.

NikolaYeah, well, let me

Damienask you, it's, it seems counterintuitive, Niccolo, from an economical perspective. Why are prices decreasing as efficiency increases? It seems like that efficiency comes at the cost of additional resources of necessary computing powder or power, excuse me. So why are costs dropping for utilization? Well, I

Nikolathink, Moore's law, right? Chips improve. There are more of them. Their availability allows, companies to achieve economies of scale, right? I sound like an economist, but I think a big part of it is data, just like optimizing, doing good work with these models, distilling them, making them, I, I don't think that full data is disclosed, but, training a better model with fewer parameters means that maybe it has the same performance, but it's a smaller model. So I think like with for turbo and then for, oh, I think that we're probably I think a lot of people are speculating and saying for, oh, maybe the 3.

Nikola5 for the upcoming release of five, right. And that'd be cool if they released it in that order. But it's it's really just a matter of. Like a lot of work being done to optimize and, you've seen, I think the other part that has to be mentioned is just like further proliferation, improvement and open source models. They are behind, but they're not behind by more than a year. And then, when you look at the debate and how it's evolved I think that, um. People are concerned about what this will do to the economy to the world at large, like who should have access.

NikolaThose debates are ever since the melodrama with Sam Altman being ousted and reinstated, I think that it stayed relevant. A lot of companies are disappearing. It doesn't look like they're going to be very many winners, right? Inflection's gone. Stability is garnish, right? So it feels like the number of those foundational players there aren't that many, right? So,

Damienbut then if we're going to go back to those macroeconomic models, doesn't the reduction in a competitive field lead to increased competition? Doesn't it, or are we talking about, because most of these dominating leaders are open source models. We're looking at the commoditization of AI in general. And so therefore. Yeah, I think,

NikolaI think that, well, look, I think it's pure capitalism, right? Thankfully, there isn't one winner, right? If there was one, then your thoughts here could be used very, very effectively, right? But as long as you've got at least two players, they will compete with each other on price, on availability, on speed, and other things. So, like the invisible hand of the market will hopefully get us better products, better and better products. Prices, right? And the faster it is, the cheaper it is. We'll use it for more. So you'll end up using more, right?

DamienOkay.

NikolaThat's. As long as Eurasia and

DamienOceania are still in play, then we're set.

NikolaWe've always been at war with who this time, I don't know.

DamienI don't know. It changes in the new chapter. Well, yeah, no, that's cool. Let's talk about it from a consumer perspective as well. Do you think that in the last six months, like regardless of whether or not we're talking about AGI, but do you think that general consumer adoption or the late persons Perception of AI and automation in their day to day lives and automation enterprise application, and the steadily increasing appearance of AI and embedment of AI into their lives is something that they're going, this is actually really great. And they're warming up to.

DamienOr you think they're cooling off? Or are they pretty much the same as where they were six months ago?

NikolaWhat do you think? I,

DamienI think that I'm seeing it become more normalized. I think that people have an increased curiosity about it. I think that the initial naysayers about the robots are taking over are actually recognizing daily benefits from it and seeing like its capabilities and actually becoming a little more comfortable with it as a concept. And so I'm actually seeing a general warming and this is just, this is going off the grandparents and the parents test, right? What are my mom and dad saying? I think that's the easiest way to gauge it. But I would say the general sentiment is an increased curiosity as to what the capabilities are and what it can do for people as individuals.

NikolaSo, yeah, look, I think there's definitely been an increase and probably, multiples over the last six months, but not, I don't feel like it's been an order of magnitude, right? I feel like the excitement among the first layer of people, this touches people working on AI technology, places like the Bay area live and breathe this stuff, and everyone else is, when I look even at like our clients, when I look at the general population, I think we have to accept that it's not reached that level of, Intuitiveness that you have when you like, hold an iPhone and it's immediately clear that this is new form factor for communication.

NikolaI think I was Justin showing the voice mode and it was at home. And I think I had the thing make me guess the top 10 largest cities of the European country, and then flip the country and go on. And I was sitting there was holding my newborn. So I can really use my hands and I was just like wasting time. And then like my wife tried it and my wife is a notorious tester of all PolyAI systems. I think that even five years ago when we had like restaurant recommendation thing. Work for everyone that I tried it with and she picks it up and the thing goes Hey, what kind of restaurant would you like?

NikolaShe goes Hmm, anything except Indian. And that didn't work. So she has ability to just break these things immediately. No matter what we do and how good we make them. But I think that like in her criticisms of it lies very simple truth. Like she started playing that thing and she said something, tried to interrupt couldn't dead. It accomplishes, it doesn't work, right? And I'm like, of course it works. You're like obsessing over details, but she's right. It is not seamless. It is not yet nearly good enough to be everywhere and easily usable.

NikolaAnd there's always that one person who pulls out their iPhone and uses Siri for things that like you shouldn't use Siri for. And they're like, Oh my God, nerd. You're doing it to prove a point. And I think, we're not doing this to prove a point. We are, I think I'm infatuated by, like GPT 4 and everything that you can do with it it's incredible. And the multimodality, the demos they've had, like I'm really, really excited. But I think the truth is everyone who is that enthusiastic, It's a bit of a, like an AI practitioner and they know what they cannot do.

NikolaSo we're all effectively thinking of skillful little demos we can show to amaze ourselves and others. And in many ways it's useful, right? Kids use it to cheat in their homework and they're caught out by the use of very fancy SAT vocabulary used in ways that only. A non native speaker like me might do because I'm not interacting with people enough to know which words are used and which ones are, Shakespearean English. As long as you

Damiendon't tell people to get thee to a nunnery, you should be in good shape.

NikolaWell, absolutely, right? I think that you see a lot of it on Twitter these days. It's someone writes something, it's you use the after this, you use the word, whatever. Here's the fancy word twice within two paragraphs. Well, there

Damienis, there is the concept that the internet as a whole is now populated by something like 90 percent bots so that nothing is actually being manifested and created on all of the internet. So that's the new birds aren't real.

NikolaThe amount of junk emerging is insane, right? I think that I was on a client call this morning and then we were debriefing with the team and one of our solution consultants was like, Is it true that there are as many Bosnians in Albania as in Bosnia? And I was like no, there aren't like, that's not a very common, like population flow, but he's up with Google saying, yeah, damn, like some word vector got confused somewhere and the LLM spat something out and hallucinations at large, it's going to get really fun. And. Don't forget. We have so many elections coming.

NikolaWe have a British one on the 4th of July, imagine. And it's going to get really, really fun. When you just look at the comment sections everywhere, that's, Hey, you know what? Save that

Damiencommentary because that is an episode unto itself. We have to, at some point get into the impact to elections and how there's already like people are flipping out over deep fakes across channels, omni channel deep fakes and what that means to the stability of politics,

NikolaAt this point, I think I've heard Joe Rogan say. Anything. It's just I don't know what this says about my recommendations, but it's like literally, things that I'm seeing that I don't, none of it looks real anymore. So it's very dystopian. It's very cool. I think that, it's it's going to be interesting to see how we even step out of it. Well,

Damienso, so the chief breaking officers, like your wife, who are initial skeptics or find ways to poke holes in it right away. And then the acolytes that are out there, like yourself, who are promoting the concept of artificial intelligence that are promoting advancements towards AGI that are saying this is going to be great for humanity as a whole, what about the folks in between, where do you think that sentiment has gone in the past six months, do you think that people are just getting more used to it and seeing it as a way of life and sort of the means to the future of tech,

NikolaThe general sentiment is like, Oh my God, AI is advancing. It's all really cool, but it's not really there. That's the general mission and the buyer. And, I think people are excited. Many people are excited. I think it's on, the layers in between to make it just a little bit more accessible, bulletproof, easily usable without having to know and account for all the details. It'll happen, right? It'll happen slowly and then it'll be everywhere all of a sudden, right? I think many people are working on it and applying these things in every walk of life, right?

NikolaBut I don't know if you feel the same way, but, walking around, looking around. We're not there just yet, okay? Yeah, oh no. We're not watching Netflix that observes, your, reactions and changes the plotline based on what you like or not and stuff or like lets you interact or it'll all come, it'll all come. Soon ish, right? It's just it feels As

DamienCharlie Booker has predicted for Black Mirror yes, it will come. That element of Netflix interaction will come eventually. Yeah,

NikolaYeah, exactly. Look, and I think it's just as you said, like it's accelerating, right? From all the omni channel demos you're seeing, that it feels like Moore's law is a conservative estimate at this point, right? Nvidia keeps growing and, the share price just, I get daily notifications of 5, 10 percent jumps. So you know something good is happening and, we're accelerating and I think it's going to be really interesting to see just how fast we can get it to, you Like do more and to get more people using it because I don't know about like people around you But I don't think more than five percent of the people I know Open the chat gpt app on a daily basis.

NikolaDo you?

DamienI do not and I work in the industry. I mean I use gpt I use gemini just ancillarily as a curiosity, but it's not a part of my daily life. You're a hipster Well, yeah, it's my rugged appearance in my Devil May Care hoodie. And so it's just that's it.

NikolaHow to use, like the Azure enterprise edition of open AI. Okay, sure.

DamienI don't need the bots, man. I got my mushroom coffee. Remember? What about, okay. So what about a real key, maybe just high level thoughts on the big opposition of progress, which could be seen regulation, we haven't seen a ton of like really sticky regulation. There, there've been some things coming out of various regions, the AIU act and the And similar legislation, like co opted by here in North America, like nothing that's really too granular and enforceable. So it mostly is pie in the sky.

DamienHow are we going to approach this? Do you see any like key changes or key impactful regulation or legislation coming out?

NikolaI don't

Damienknow.

NikolaDoesn't feel like we're seeing like a Napster moment. Yeah, we really aren't. And there were those lawsuits and New York times and people just making sure that. The value of their data set is not like completely used and abused before they have a chance to monetize it. I don't see, I don't really see, I feel like there's just too much excitement and it's all moving way too fast for regulation to keep up for now. I'm sure, the taxmen will come eventually and that's okay. Like it should happen, but no, I, and I'm glad it's not because I think that like it's still exciting to, to think about regulation first.

DamienIt is. It is. I agree. Any other, you mentioned a few of the key players in the space. And obviously we've talked about 4. 0 from open AI's recent announcement. You're exploring on the daily when different functions that were demoed recently are coming out. What about some other key players in the space? The Anthropics of the world, the Alphabet Models that they're rolling out. Are they indicating progression towards from a technical set from a technical architecture and infrastructure place? Do you think that they're getting? Are they doing leaps and bounds growth?

NikolaWell, I think the interesting thing about Gemini is the kind of one million plus context window, right? Yeah, that's serious, right? I think at that point we're looking at how many A4 pages is that? It's very many. It's a few books worth, right? So at that point I think It gets easier and easier, right? I think that it's going to get really like what you notice when you really load up things in the context window is that inevitably more and more things that you've put in contradict.

NikolaBecause the structure of what applies to which part, et cetera, it gets a bit mushed together. Imagine if you're reading a law textbook, right? Different facts from different parts could be applied and used to contradict each other, and that's where Sure. I think we need stronger, multi step reasoning that, we hope will work. As an emergent property from like later LLM, something like hopefully GPT 5 that the whole thing will work better. Because even though the context windows have been growing, I think that at least in enterprise applications that we do, we see the importance of rag and also just of other things that create those safety controls and benchmarks, right?

NikolaSure. Are you sure that it works? That the right information is used? Yeah. Yeah. Supporting facts used in a different, say, part of the conversation in our case are the right ones, right? Do you do it with function calls, or do you, just stuff it in the context window and hope for the best? We'll see. I think that like stronger reasoning power is needed to really make use of those longer context windows.

DamienDo you think that's a, do you think that's a tipping point from excitement to fear? And maybe not a recursion of skepticism, but like a little bit of, Oh, wow. This is really huge. This is really big. I was excited to see how large it could get. Now I can see it. Now we need to figure out how to control it. Do you think there's a little bit of a tipping point there?

NikolaWhen you say control, what do you mean?

DamienThings like you're saying, things like motions towards retrieval, augmented generation, and putting, safeguards and guardrails around the capabilities of generative AI. Is that where somebody goes the perceived invasiveness, the capabilities are now a bit of a threat,

Nikolalike a tangible threat. I don't think it's actually really a threat. I think that there is a risk of the performance seeming more and more impressive while actually Not delivering and guaranteeing performance, right? Okay, at that point, let's say that, in one point of your, in the context window, let's say we're talking about like jaywalking, right? It's okay. Jaywalking is legal. In the UK, right? Like it's perfect. It's not punishable by law. Whereas I don't know, you go to Brussels and it is. So what if those things get muffled together and you start getting advice on the basis that jaywalking is okay in one place because you

Damienknow,

NikolaI'm a Brit and like I'm whatever, and like it interprets that as okay, that part is relevant rather than something else. It doesn't maybe, Okay. Realize that it needs to qualify where you are to give some advice. And this is all fine, right? There's no expectation that it should just have that superintelligence, but I think because it's so good at freestyling and doing all these things, the risk is that many people will just assume that it will process it the way that a human would, if they reread that context window a gazillion times and went back and forth and carefully painstakingly make sure that everything it says is correct.

NikolaWell, it's really, it's much more like a human who's. Read it, has pretty good memory, and then like just blurted something out, right? So, the tighter the model, the better it is, the less this would be the case, because the internal representation of the context window is the parameters of the neural net will be better and more precise, right? But like the ability to internally think and create those connections, right? Like the problem is solved effectively on the basis of all that like that's gonna be like we don't really have a clear understanding of how it happens It definitely does and that's why it's fantastic and interesting and amazing, right?

NikolaBut we don't really have the guarantees and it's always been like from the start and now more than ever, like the magic of deep learning has been in the fact that it works so well and that we can't really, and that nobody knows why, yeah, well, we know why we know a lot about the why, right? We know, but it's the fundamentals,

DamienBut the actual like process by process, it's still very mysterious.

NikolaYeah, it's yeah it's fantastic. So I think that just on the basis of that, we just have to figure out, and like you could try to regulate, right? Like you could have gone and regulated the airline industry right after the Wrights brothers. But maybe you shouldn't have, maybe you should just actually get on with it because the upsets too maybe something will happen. Maybe a plane will crash. But I think it's usually either the pioneers, the people who really want to push it. Like they're the ones who are okay taking that risk, right? We evolve we take risks, we do reinforcement learning over You know ourselves, right?

NikolaSo, we'll learn from mistakes and yeah, we might make some Difficult ones, but we should continue to experiment, right? And then, same as with flights, I don't think to this day, when you look at like the fluid dynamics and simulations of why do planes fly, why are they safe? Are they a hundred percent safe? Some mathematically provable, none of it, it's just like heavily, like all sorts of different turbines, different settings, et cetera. And in practice, and, Today, it's safer to get on a flight than probably to get in a car in most countries. So, I think like with these things, we just have to create the right structure of applications, make sure that they Work well enough and then learn when to exercise common sense around what the things are telling us, right?

NikolaI was at a panel in Paris yesterday and Someone asked the question about bias and I was like, well, I mean you can choose to look at it in a very negative way Where you say, look, there are things, the data is like biased against this group or that group, or it has more data for one or the other. But I open 23andme and the summary of me is that I'm. 100 percent Balkan, actually in 0. 3 percent Korean, which is inexplicable through genealogy

Damienon that 0. 3%. Let's get

Nikolato the root of it. Yeah, it was all

Damienthat

Nikolakimchi. But the fact that it doesn't have a more granular breakdown, it's just because it doesn't have enough data. So I think that as we use these tools, say for medicine or something else, We just need to make sure that in education rather than preventing kids from using, GPT 4 to cheat and, do their homework. Maybe we should be setting them much harder homework where they have to use GPT 4 to do it, right? And if they don't produce something that looks like a PhD dissertation, In those 45 minutes, you failed that class, right? Goodbye. You're done.

NikolaYeah give me another PhD dissertation, right? And, in doing that, maybe that's how we create that hybrid human computer superintelligence. We just gotta push ourselves. And I think we need to educate ourselves a bit about statistics and just understand what it is we're using, right? You know, electricity should be grounded, like rubber gloves are a good idea, or you might get electrocuted, right? That's something we teach children, or at least we ought to. Similarly, with using these models, like A sharp knife can cut you, a stove can burn you, and, using an LLM that's freely speculating over something that hasn't been trained on is also dangerous, right?

NikolaIt seems a little

Damienbit of a safer training curriculum, as opposed to, hot stoves and sharp knives. Just early in grinding with regards to potential and capabilities. Fair,

Nikolavery fair. Cool.

DamienAll right, man. I think that's all we have time for today. I appreciate chatting with you. We'll cover AGI again in six months when it's probably a part of our daily reality. But it was, I

Nikolawas just having a podcast for itself. Well, you should prompt yours and mine and we should clone our voices. Oh, just see

Damienwhat it comes up with. Give us a, give us our next podcast episode. We'll release it. We'll say who did it better.

NikolaWe'll see if we can get internal engineering resources to do this.

DamienYou're the boss. I think they'd listen. All right, Nikola. It was great to chat with you. I'll talk to you soon. Okay.

NikolaSee you later. Bye.

DamienTake care.

Open the transcript file

Hosts and Guests

Nobody is credited on this episode yet.

Credit the people on this episode

Be the first to rate this episode

More episodes

All episodes
  1. What if your contact center became a product team?

    Nov 17, 202635 minAudio

    Three years ago, Simplyhealth's PolyAI deployment was only resolving simple FAQs over the phone. Today, they run voice, in-app, email, chat, and WhatsApp conversations on the platform. Nikola Mrkšić talks with Paul...

    Credit the people on this episode

  2. What happens if your AI agents get 1% better every day?

    Sep 10, 202639 minAudio

    An agent that gets better every day compounds into a very different agent within months. That's the idea behind Wren, PolyAI's new release: an agent whose job is to make your other agents better. Some of what it finds...

    Credit the people on this episode

  3. Who's coordinating your army of AI agents?

    Sep 3, 202644 minAudio

    Companies are pouring money into AI, but not everyone is getting the returns. Ian Jacobs, VP and Lead Analyst at Opus Research, thinks he knows why. In this episode, your host Nikola Mrkšić sits down with Ian to work...

    Credit the people on this episode

  4. Why should CX leaders care about MCP?

    Aug 20, 202636 minAudio

    When PolyAI built its Builder MCP, one client helped shape it: Vixxo Facility Solutions, one of the largest facilities providers in the US. In this episode, guest host Damian Sasso talks with Vixxo's CTO Derek Neighbors...

    Credit the people on this episode

  5. Can AI really hear a call the way a person does?

    Aug 6, 202622 minAudio

    PolyAI just launched Dialog-RSN-1, its first audio-native model, and your host Nikola Mrkšić sat down with its builder, Matt Henderson, to unpack why it’s a game-changer for building voice agents. Most voice AI either...

    Credit the people on this episode