UP529 ChatUP Live Pt2 - The Billion Dollar AI Race: 'Once you give away that level of knowledge, it's gone'

18:07Sean Betts:And we're going to put a cost against it. And guess what? the markets will decide what the value your data is worth over time so imagine fans could build off the back you know your football club give all your fans access to apis because the next generation of kids my 16 year old now is literally in london tonight with three friends they're all building the next generation of like fantasy apps and stuff and they're building with apis because they can do it now with the ai so you're kind of building on this stack but every day i hear stories of someone

18:34Richard Ayers:has you know a new release today google has just released yesterday which has wiped out several companies that whose market value was considerable in the tens and hundreds of millions just purely because it's only only the top layer though i think that so the risks of building if i go and build something on top of no no but that's not the point so the models that no one

18:58Sean Betts:else can build a model like the models are billion multi-billion dollar like no one's getting there uh the point is in which they can operate and survive the oxygen is information and data that's what they need in order for them to keep getting better they may need more information google's done well because it's got huge amounts of data what i'm saying is the oxygen protect your own oxygen and put a commercial ring around it and sell it so that you don't give it away so so that model is amazing i'll be playing with it for the last 24 hours built loads of apps it's beautiful design but it's basically commoditizing the interface layer the ux that the application layer that's that's gone but they still in order for them to be valuable in the future they need super incredible, niche, brilliant amounts of information and context.

19:49Richard Ayers:Okay, so that was the live panel conversation. And thanks again to Craig Hepburn, Richard Ayres, and Sean Betts for their time and enthusiasm and expertise. So after doing the panels, I wanted to follow up on some of the threads that we covered and some of the questions arising. So I sat down, first of all, with Sean Betts again, who is in charge of AI and technology across the Omnicom group. So as you'll hear, he was able to take the conversation and point it towards his world of brands, sponsors and media. Here's me and Sean. So just to start us off, just give me your response from the evening, having sort of looked through the first bit and sat on the second bit.

20:33Richard Ayers:What was your just your general sense?

20:35Craig Hepburn:Yeah, I think it was interesting trying to meld the technology with live conversation. And that, because I'm a tech guy, that was the bit that really kind of grabbed me. Because I do think that there is going to be a bit of a shift in how we think about using AI and it will be alongside what you were trying to do, which is how does it help us throughout our day and listening to our live conversations and helping capture and process and summarize and use that as context for everything else it does for us. So we're nowhere near that point yet, and the hardware needs to be developed, and culturally we need to accept that and all sorts of other things.

21:14Craig Hepburn:But I do think that that idea of having a kind of ever-present AI companion is probably a part of our futures. And so what you were attempting to do at the event was kind of similar to that in a bit of a microcosm, and it was quite interesting because of that.

21:29Richard Ayers:It felt like there was a part of the impulse was to, first of all, it was out of boredom of conferences and panels and trying to sort of do something else. There's a theatre element to it, which I quite liked. But when you get to that sort of how we're interfacing with the chatbot and large language model prompting, essentially. and Andy you know whipped this thing up and it was for me as a complete AI dullard I was wowed by it but I want to know what the is that are we just learning via that method that those in you know my interaction with it is all about okay prompting them being told that I'm not prompting in the right way and you know the language model is hallucinates and lies and doesn't get it right and then it compounds the errors.

22:22Richard Ayers:And I'm just wondering, my little world, my little window into this is pretty much that describes it. I can see the opportunity without actually, you know, manifestly getting much better at anything. Yeah. Is that where we are and is that just structural and that bigger things need to happen? Or are other people further along the line than that?

22:44Craig Hepburn:So I think what you've captured is probably the experience that most people have at the moment with the technology. But I don't think that's reflective of the current state of the technology, if that makes sense. And the reason for that is because 90 to 95 % of people who use ChatGPT, as an example, are free users. And the difference between what you get as a free user and what you get if you're a paid subscriber is quite stark. Not just in terms of the knowledge and the capability of the models that you have access to, but also the features that OpenAI wrap around it as well. so that's something that I think has got to change for most people to be able to experience what the technology is currently capable of because I think free users are probably experiencing what the technology was capable of at least 12 months ago maybe 18 months ago and given the speed that this technology is evolving that's quite a big gap and I think that that will come next year.

23:45Craig Hepburn:I think that will come when OpenAI come to market with an advertising model in ChatGPT, because as soon as that happens, they can open up all the features to free users. And essentially, the difference between what a free user gets and what a paid user gets is exposure to advertising rather than a deprecation of features. And I, because it's part of my job, I am a paid user of a lot of these platforms. And the experience that I have is quite different. I don't feel like I have to think about my prompts too much and I don't see anywhere near as much hallucination. And I can trust and rely on those models to do the things I want them to do a lot more.

24:27Craig Hepburn:And I'm just looking forward to hopefully a point next year where that is everyone's experience rather than just the 5 % to 10 % of people who are actually paying for the premium access. Right.

24:36Richard Ayers:That's interesting. So that there's a, are you, are you sort of getting to, um, more sophisticated systems? So sort of complex orchestrated workflows. So this thing, rather than it making things, getting research faster, rapid summarizations, you know, all of that. And we're talking here from an organizational perspective that you're saying, okay, I can now see this, then that, and that, and that, and these things are going to, you know are working and i i don't know what to call that because obviously i don't know if that's

25:11Craig Hepburn:agentic or or not but so yeah it i i have a bit of a bugbear around the word agentic and agents because i think it's used very broadly to mean lots of different things but the the pattern that you're describing is is absolutely playing out um but it's probably worth just describing it a little bit so you know if you go to a chat gpt or a or an anthropics clawed model or a google Gemini or perplexity or any of these kind of platforms you're faced with a very blank text box that is a chat interface you know it's it's very simplistic it's not very interactive it's very basic and it's not obvious what you should do with it for most people so anyone who starts off there will start just putting stuff into the text box and kind of seeing what happens really and that's that's that's kind of how we learned how to search 25 years ago with google and we all kind of found a way to write our search queries more effectively so we get the results that we want and so i think people have been going through that and are still going through that with with these ai platforms right now what i have found over the last probably year is that i've evolved how i use them from that to something that's slightly more sophisticated because i just spent so much time in this technology because it's it's a part of my job and the way that i use these platforms and these models now is more akin to a work companion I don't want to use the word intern because it's not like I'm briefing the models and the platforms to go off and do stuff for me and they'll come back to me a day later with something it's it's not quite there yet but the way that I work with the models right now is very iterative so if for example I'm exploring you know some new thought leadership that I want to do around around AI as a technology I will start off by talking to the model about the ideas that I've got and the general direction of what I think I'd ask it to give me feedback to give me ideas I will then kind of shape it with the AI to get it to a place where I'm comfortable with it and I think it makes sense then I'd move to if I'm my ultimate endpoint is normally like a written article or a presentation I'll then move on to kind of writing an outline of what that would cover that would then iterate into writing the first draft i'd then be editing that draft going back and forwards with the ai model on it to make sure it's in my tone of voice and my language and it's making all the points that i wanted to make then when i've got a final draft i'll get the ai to review it for polish and any feedback and any kind of typos and things like that and i'll get it to a kind of final state that i'm then happy with So it's it the way I work with it evolves through that piece of work.

27:52Craig Hepburn:And it is, like I say, more like a work companion in that sense. So the.

28:00Richard Ayers:Extrapolating from that, again, it's and it's difficult because if you know, it's we have to make it abstract. but we started off by saying that there's a, there will be organizations in sport and in the round sports, the sports business who will rapidly benefit from this. And I can, and you hear it day in, day out. And again, it's, there's a marketing thing, you know, people are just claiming and over claiming, but then there is reality to that. And one of the bits of the conversation I quite enjoyed from the stage was the, okay well what does that mean trying to put some you know reality to that to that point so that you know the billion dollar sports business idea the conceit one of the things that you said which has sort of remained with me and i've wanted to talk to you about was um the definition of data and you then made the point that it's it's moving from what we you know spreadsheets and personal data and information to to everything or anything yes can you just sort of yeah because i thought that from a sports perspective is incredibly important because we're at this stage and again this was part of the conversation was that how we work with the big foundational models and do we give them everything in and normally in sport it's framed as a sponsorship relationship so we then say right we'll get into bed with anthropic or open ai or microsoft or google it will be framed as a sponsorship deal that has a term limit has three to five years whatever it is but this is something that is really really important about what the share and the value exchange is and if i'm the premier league or if i'm the fa or if i'm the ioc or the nfl i've got a lot of value which traditionally i've traded and given to third parties over the years by tv you know monetized it by licensing it to third parties if i do that this time someone i think craig said this is the last frontier of all your intelligence you know once you do this all your intelligence and knowledge as an organization has gone yeah yeah yeah just give me a sense of that because that's i came away with that as my quite a big takeaway and it's something i want to talk about in the

30:30Craig Hepburn:you know in the follow-up yeah absolutely so let's start with the data point um that i made and i think data has always been anything that has been digitized but because most people's exposure to data is an Excel spreadsheet or a bank statement. Everyone always thought about it as it being numerical and it being tabular and structured and easy to get your head around. And to be honest, that has been the story of AI development for the last 70, 80 years. It's all been about structured data. But the change that we've seen in the last 10, 15 years that's led us to this point now with generative AI is that AI has got really good at working with unstructured data, which is anything that isn't in the table and it could still be numerical but often it's text often it's images often it's video often it's audio it's basically anything that's digitized now so everyone's kind of understanding of what we mean when we say data needs to needs to kind of broaden out and there's kind of two bits to that so if you if you think about the context that we're talking about this in around the sport of football there is a huge amount of data around football not just in terms of you know tracking players and their performance and their health and all of that but the raw footage of the games and the highlights and the broadcasts and the commentary and analysis that goes all around that is a very valuable form of data and just to put this into perspective in terms of why that data might be important we're certainly not there yet but within the next 10 years I would say there are going to be a lot of companies trying to build out robotics platforms that would be very interested in seeing premier league footage to understand how people move so that they can use that to help them understand how robots can move and that will be a very valuable data set for a robotics firm and that might be a general robotics platform or it might even be a robotics platform that's aiming to create a robotic version of football I mean who Who knows?

32:31Craig Hepburn:The world can be crazy. So, you know, that's a very simple example of why that data would be incredibly valuable. And one of the challenges that a company like OpenAI has with the future of training their models is that they don't have access to the same kind of video data that Google does, who owns YouTube and has got billions of hours of footage of all sorts of different things. so open ai if they want to progress further and start training their models on video data in the same way that google will be it's going to have to be looking around for where they can buy that video data from and they'll want that from sports companies from entertainment companies from lots of different kind of verticals because it will have different uses so as as an example that is why you have to start thinking about data differently and you've really got to think about the value that you're sitting on and i think one of the points i made on the panel that we did is that the best value from that data isn't always holding on to it sometimes data can be very valuable outside of your organization because it can be used effectively as a marketing tool but sometimes you want to keep that internal because you can see more value from it internally as well so those those are the kind of decisions i think a lot of big sporting organizations are going to be facing over the coming years.

33:53Richard Ayers:So a deal is fascinating. So a deal with open or open AI and or anthropic or anyone who isn't Google basically is going to be different than a Google relationship because actually Google, there's a need on the other side, which is much more burning, much more urgent.

34:18Craig Hepburn:Yes. But I think the challenge is potentially more nuanced than that as well because you know the premier league sells the rights and it sells the rights for broadcast it sells the rights for highlights those highlight rights are able to go onto youtube the second they're onto youtube google owns them because of the terms and conditions of youtube so in effect they've already sold the rights to google but they've done it via the broadcaster that owns the highlight rights that publishes on youtube so that's something that is already in existence and and probably needs to be looked at more thoroughly because the second highlights get posted to youtube google owns it and i i had a sort of almost in the same way of

35:01Richard Ayers:the week before we did our thing i had a i did a panel i'm not always on panels but um it was with sport radar who are you know the betting sort of infrastructure yeah and one of the conversations there was you know for your robotics um execution of the data they're looking at okay well what does this mean for prediction and what does this mean for you know in terms of reinterpreting what a football match is and again trying to second guess if i am the premier league and we always land on premier league because they're the biggest and the best but every sports organization trying to second guess the future market for their data is a sort of impossible task i would suggest at this point you know it's just beyond their capability yeah i don't know how yeah and it's a lawyer question i guess in terms of future proofing what your data is then used for and whatever but it just feels like there is a there is jeopardy here because in the moment we're in at the you know this is the battle and i'm not sure that many organizations know they're in that fight

36:16Craig Hepburn:yeah agreed and i think i think there's two things that i would be thinking about if i was on on that side of the fence one is any rights or data licensing agreements i put in place right now i wouldn't agree to anything longer than two years just because i don't know anyone can reliably predict what the world's going to look like in two years around this technology right now. I think one-year deals are probably not practical, but two years feels about the right kind of balance. The second thing is, to my point about the Premier League highlights and the broadcasters and YouTube and Google, I think you have to think through a lot more carefully now what the downstream effects of that licensing deal are and what you're potentially exposing yourself to.

37:02Craig Hepburn:because I think that there is unfortunately now very little chance that the Premier League could extract value out of their footage and their licensing from Google because Google have effectively already got years of it on YouTube so why would they need any more whereas if that stuff hadn't ended up on YouTube or at least there'd been some extra remuneration because it was ending up on YouTube and therefore in Google's training data there was some other form of

37:30Richard Ayers:commercial arrangement around that yeah no i get that because then so strategically youtube is incredibly important to google hugely important i'd say it's probably their most important asset right now without a doubt if you are not a rights holder so you know a lot of the focus is on the premier league or if you are generating the source material are you the sports rights holders what are the questions on your side of the fence from a marketing perspective it's because again it sort of feels like a similar conversation in terms of what are we now not talking about your own company but i'm talking about maybe you know some of your clients everyone is then saying well i don't quite i've lost my sense of gravity in terms of what this organization is going to be do you know i mean do you get that sense from clients and talking to people i think well if everything's content and therefore everything is data what are we yes yeah and i think i think

38:24Craig Hepburn:you know sticking with the world of football you can already see that there's there's different clubs thinking about this very differently you know you've got some clubs who who take that context and think okay our role is all about the fans and it's all about the community and it's all about the audience it's all about making sure that we're the most loved football team in our locality and yeah everyone loves coming to watch the games and it's really accessible and all of that then you've got other clubs that are all about for want of a better expression global dominance like they want to be the biggest team in the world they want the biggest revenues they want the best infrastructure the best training grounds the best players etc and that's a very different approach so i i i do think that everyone is trying to figure this out for themselves right now and there is no kind of right or wrong answer i think it is just asking a lot of questions about you know what a football club is and what does it mean as a business and what does it mean as a kind of cultural touch point in society as well and there's different ways into that and there's

39:26Richard Ayers:different ways out of that as well yeah yeah yeah one bit that craig mentioned the api economy essentially what my headline is yeah and one of the points he made which i'll bounce this off which is that sports organizations need to think you know there's a cliche about thinking about like a tech company and build your own api apis and tech stack so others can build on your data was one point so there's that sort of the tech bit and you'll you'll gather that i'm reaching the sort of ceiling of my uh understanding here but the quote was the world is moving to an api layer api is becoming transactional layers and his it went on to say right okay well there's a future there where if you give fans or developers api access to build applications around and on but because again it's very technical from my perspective but i don't know what the real world application of that would be i understand it and it's something i'll drop at dinner parties to make myself sound clever but i don't want anyone questioning it's like speaking you know saying something in Spanish and then having to answer a question but could you explain what that means

40:37Craig Hepburn:really I suppose yeah absolutely so I think this has been a bit of a slow burn in the kind of development community for quite a few years now but it's now becoming I think a little bit more important to every organization and therefore a little bit more kind of visible and talked about so essentially the way I would try and explain this in kind of layperson's terms is that we've been used to being able to find out more and to be able to access information through a website which has been very visual and easy to use and increasingly when we look at how AI is evolving and the fact that AI will be doing things on our behalf in the future online we want AI to be able to access that information and those services and that means it has to be built not just for people but for technology and the way that you build technology access is via APIs that's just the the infrastructure that is being built out by us to to enable that to happen and that's a really important change i think because it essentially does two things one one it helps with the future ai companions assistants whatever you want to call them to be able to do the things that you would do online because they can then access it via an api and do all of that stuff in the same way but it also immediately does another thing which is it opens up your data potentially to other people to allow them to do stuff with that and an example national rail they sit on all of the data of what trains are running and when whether they're delayed or not ticket prices train routes etc etc they built apis out to allow developers access to all of that data and then you saw lots of different mobile applications being built that allows people to see what times their training running, whether there are any delays, book your tickets, all of that stuff.

42:24Craig Hepburn:That wasn't possible until National Rail started building out those APIs to give that data to other people and to be able to build on top of it. So I think we'll increasingly see people building out applications on top of these APIs as they become more readily available, because I think there's just a huge amount of interesting use cases for consumers for a lot of the data that is currently not available because it's just not that infrastructure hasn't been built out yet yeah that's really

42:51Richard Ayers:interesting because there's a again one of the running themes of what we talk about a lot is where the sports body starts and stops and what they view as a leak from their own economy versus what they can capture and put arms around and this requires a very different mindset it feels like they have to have an open mindset which is really hard because they've been trained to have a closed mindset and everything is category exclusivity and you know lawyered up and if you if you sort of advertise on one side of that line away you go and you get a letter so that's it feels it's technical but it's also incredibly difficult cultural shift and a very important strategic

43:33Craig Hepburn:shift as well and it comes back to the point i was making earlier about how rights holders have to be i think really clear about what data they want to keep and charge for versus what data they want to let out for free because it has some inherent value to it as a marketing vehicle so the example i use when i talk about this is from the music industry so obviously you don't want to allow ai models to train on the music of your artists because that is the artist's ip and the record label's ip and it should be protected because that's how you know the artists and the labels make their revenue make their money right but there is a huge amount of value in the information about an album like the track names the lyrics notes from the artist about what the meanings of the lyrics are and what mindset they're in there when they were writing it and information about the musicians and the recording process all of that kind of metadata isn't valuable from an ip perspective but it could be very valuable from a marketing perspective and this this is quite similar to how the music industry evolved in in asia where they effectively 10 15 years ago decided that albums should be for free and they'd make all of their money in live performances so they give albums away for free because it was a marketing ploy to get more fans and they monetize those fans through live performances and merchandise and and all of that and it's it's a similar thing for i think sports right holders there'll be some data that they have that they can use as a marketing vehicle and there's other data that they could monetize directly and they need to think about that strategically and and figure that out okay so that was sean betts and myself

45:18Richard Ayers:and now just to finish off you'll recall if you heard our previous episode where we created a co-pilot type language model that ran in parallel with the on-stage conversation and then fed back and built on the topics and questions that we covered. That's available in a link in a Substack newsletter that went out this week. That model was created from scratch by TFG Labs' Andy Shorer. And I wanted a quick debrief with Andy and Chris Woodcock, 21st Group's CTO, to see how it was from their point of view. Have we learned anything, Chris, about who's going to win, who's going to lose? Just give us a sort of your summary of what we've seen.

45:56Andy Shora:Yeah, what I hear and what I've heard all night is you've got this kind of dichotomy right between everything's becoming really easy. You've got that 10x advantage of stuff that you want to do, whether it's coding or getting an answer. But also, it's really hard. It's hard to build this system. I've seen behind the scenes and the number of nodes on there and the way in which you can build value out of it is really hard to do. You put this stuff into Gemini and you don't get it out. So it's like that's a real interesting thing to tug on, right? It's like, so where have you got, where can you make things really easy and get those 10x advantages?

46:43Andy Shora:But then also where have you got to invest time and effort into building complex systems that actually do what you want to do and deliver value for the clients, which they couldn't otherwise do?

46:55Richard Ayers:So also it's sort of trust. I have a problem with trust in terms of with the machines machines I sound like a grand but with the there's trust in terms of you know good actors bad actors but there's also right and wrong I spend a lot of time going in and correcting at quite a deep level mistakes that they've made and then extrapolated from the mistakes so it's the classic hallucination problem and whether or not we've actually this is gonna get better or are we just gonna have to live with this.

47:26Chris Woodcock:We saw just this week in the news, I think I saw on the metro on the bus down here, CEO of Google has told the world to not trust AI. And to be honest, it's probably the right time for him to project a message like this, because people are blindly trusting LLMs to make and actually execute quite big decisions, which affect their lives, affects their performance at work. And there is a threshold you must pass in terms of competence and understanding what these LLMs are actually doing. If you do not pass this threshold, you'll not be able to amplify your productivity. You're just going to get lazier.

48:05Chris Woodcock:And, you know, the truth is, unless you point these systems at the right data, the advancement in the intelligence of these foundational models means nothing if you're not giving it the right inputs.

48:18Richard Ayers:Yeah, that question of who's going to win, who's going to lose are there any signs in terms of the characteristics of companies organizations who are going to prosper or are in a position to be able to prosper i think you know something else

48:32Andy Shora:like we heard you know over the evening is like there's that cultural element right and you look at the role that sean has got for example you look at the role andy's got in our organization and i think there's a huge cultural element to it and you know i know you're a fan of the faster horses analogy which i love as well and i think there's a there's a big danger here that people take this you know these developments and try build faster horses and what we need is you know the forward model t like um that's what you've got to be thinking about and that doesn't that's a mindset shift but it's also it's really helped when you've got people like sean engaging with you and you've got people like andy in the organization and i think if you if you don't have that that's the risk for me is like you you miss you miss that cultural shift um and then and

49:18Richard Ayers:that's where the danger lies i think the the quality of i mean there's a sort of shit in shit out question here in terms of if your data isn't in any position to be of use this is not going to help just putting a chatbot on top of crappy data is just going to be a waste of time or it's true

49:36Chris Woodcock:And at TFG, we face this exact situation. You know, every company has modern sources of new data that someone has wrestled into a good shape to build applications on top of and probably some legacy data that relies on a human to tell you, you know, which columns in a database not to trust. And we tried to initially build on a larger data set with less reliable data. And it just proved to be wasted time for very simple things pointed at, you know, by chance, the right data sets we got good results but we very quickly plateaued and and the same for every organization right now that's building with ai tools are realizing that they need to get their

50:15Richard Ayers:house in order so one of the one of the questions i've got is and you mentioned it you touched on it earlier which is the sort of value of the data what is the value of what we've got now and i was talking to sean betts about google and why google is very well placed to take advantage because it has youtube and it has already got training data of millions and millions and millions and hours of video so this question which got discussed in the ai panel which was about what is data now because we've all got people have got an idea in their head about what it is but then when everything is content when everything is data we are essentially feeding models it needs to be thought of in that way what do you think about that question yeah it's a really interesting one

51:04Chris Woodcock:And at TFG, data for us means truth. And it's really important that everything we do and decisions that are made, big decisions that are made, are grounded in truth. And it's very easy, as we've probably all experienced, to get an LLM to hallucinate. It will invent information in order to satisfy you, the user, unless you give it really strict instructions to tell you when it doesn't know the answer. So it's really important for us internally to be trained up on the way LLMs work, to be aware of the dangers, and certainly by building a layer of infrastructure that allows us to gain alignment, ask questions to large language models, which then go and interrogate the right sources of data.

51:59Chris Woodcock:it unlocks amazing new avenues for us to experiment and and it increases the amount of kind of access the things we can access per hour from a resourcing perspective in order to achieve greater results now yeah as you mentioned google are very well placed i agree they're they're making massive strides at the moment um but the data that they're training foundational models on is running out and progress is plateauing. There are many ways to invent data to gain small advancements now. But in my opinion, these models have reached the level of intelligence, which will see us for a lifetime of doing kind of various office jobs.

52:49Chris Woodcock:And one of the problems, I think is the things we're asking LLMs to do are not complex relative to the challenges these models are being given to verify their intelligence and this there seems to be this obsession with trying to one-shot mathematical proofs in order to achieve the top of the leader board on all of the comparison of the new new life language models and and none of us are trying to provide proofs to mathematical theorems as part of our jobs at least we're not quite at that level anyway we're doing lots of complex analysis but they don't involve you know achieving a

53:31Richard Ayers:fields medal or a gold olympiad in a couple of times in the run-in and we had group calls in the run-in to the event and one of the things that you mentioned a few times was the one shot trying to do everything in one shot which is sort of something that stayed with me i again i don't know where we are one of the bits of the conversation on stage which was about and i think it was craig who said about the api question in terms of you know whether sports organizations are going to then sort of shift into reinventing themselves and this is going to be a heck of a challenge but they're saying well okay there's an api layer that we that that is going to then grow and one of the phrases that you use quite often this shift from sort of simple large language model prompting which is what we're doing at the moment and we're sort of being told you're not prompting it right and it's not doing it's hallucinating and so you get into that loop versus complex orchestrated workflows and building whether we call it agentic or not i don't know but that feels like that the promise of this thing is that and i don't know how far we are from that and to the initial question that we posed on the night which is the billion dollar sports business what that would involve and the characteristics of which we discussed where are we in that question and it's that sort of orchestrated question i think what's your view on that

55:03Chris Woodcock:so orchestration has been a massive unlocker for us and if i may explain what that means for a moment You can think of an orchestrator as a project manager that's an expert in planning and is available all the time, has a constant view of all the resources available to them. Let's pivot to an example of something I'm also passionate about cooking. The way things used to work when we were playing around with early versions of ChatGPT. Let's say I'm a chef and I'm given the instructions to make a bolognese. I might initially look at a carrot and start chopping it up. And then I might look at an onion and start chopping it up.

55:46Chris Woodcock:I might decide to boil my pasta at that point, but then I'd realize it's probably ready way too early. And you can see by looking at these tasks sequentially and just making decisions based on a simple probability distribution of what should I do next can yield some very unexpected results, but also expected results because we know large language models are just spitting out the next word and so you can do that for a lot of tasks and get acceptable results but if you have a system that can break down all of those tasks chop a vegetable put a stove on boil pasta and if you have an orchestrator they can look at the user's intentions they can look at all of the tools available and they can form a plan and they will often form the same plan given the same ingredients and tools and then they can both outsource parts of the plan to be executed by the right kinds of agents in this case it's just one agent the chef but it would still work and they can also keep a ledger of the work that's been done and the work left to do and just to throw in another amazing part of these systems they're able to course correct if you have a manager and a chef underneath based on what's happening on this ledger or whether the chef's experienced any problems you can enter a new kind of workflow which is something's gone wrong i've chopped it open the onion and it's gone bad for example and enter a new kind of problem solving workflow that could involve other agents it could involve an agent to go out and buy another onion for example and ultimately these these systems are problem solving algorithms in the hands of an orchestrator makes the work of the chef really easy and yields more deterministic and expected results And that's possible now?

57:52Chris Woodcock:It's possible now. And if I told you it's actually quite simple, I don't know if you'd believe me, but this is just classical programming. It's functions being called with the use of LLMs to interpret what's happening.

58:09Richard Ayers:so that question of who's going to win and who's going to lose in the sports business race what do you think the answers are to that because again we touched on it on the night and we are in a race i think and i think you know it will go sector by sector and you you know i listen to podcasts and read a lot of around things like health care or finance or education and so it does start to then splinter and i'm interested in the sports bit for obvious reasons what do you think about in terms of those that you know the characteristics of the winners well from what

58:45Chris Woodcock:we've seen you know it's very easy to go and browse tech news and read about success stories and and they all seem to feature the words ai right now and that that may be a bit of um how the the search, the news algorithms are working. But when we provide solutions to clients at TFG, AI is sometimes part of that solution. The rest of it is technology. You know, we apply the right kinds of technologies to solve different tasks. I think the winners are the ones who acknowledge that AI is part of a solution and technology should adapt to the way your business works. Some people are extremely fluent.

59:27Chris Woodcock:Let's say I'm an analyst and I'm scouting football players and I'm using some tools already and I can do a bit of programming. They're going to use AI in a very different way to someone that's just graduated, doesn't have much domain experience and needs to learn more about how clubs work, how players develop. so the adaptability of these tools is great that is where we have to put the work in and it's not work that really big tech companies are doing because they don't really understand how things work in our industry where decisions are often emotional backed by data data is sometimes wrong or not representative of a situation.

1:00:14Chris Woodcock:And so it's easy to see who the losers are, the ones who aren't experimenting, who are resisting change. And right now, everybody should have kind of adopted these tools to improve their personal productivity because that is absolutely critical just to know your competitors are doing something. You know your competitors have become twice as productive. if you don't you're going to have to punch above your weight somewhere else and for what we've seen the winners are the people who are not necessarily aware of what the solutions are what the next product they should build or what tools they should build but they're the ones who are having the conversations now and are starting to codify how their processes work identify their proprietary data and what value might exist there and the ones that are having the open conversations

1:01:09Richard Ayers:with us within sport again we i did a we did a betting event a few weeks actually the week before we did our own event and you got betting or you've got ticketing or you've got media you've got the football club as a as an entity you've got the sort of premier league uafer level governing body what their relationship with these things is so each of those poses different similar different questions trying to work out what they are one of the sort of interesting bits i think is again this comes to from my conversation with sean was that he was putting forward this idea that okay we don't we don't know what the executional layer of the data is that we possess so it could be like the premier league has got fantastic training data for robotics because for obvious reasons it's got a load of players running around and you know in terms of there are so many iterations of what that could be used for just on that on a story level i can see exactly what he's talking about and so trying to second guess what to sell now what to how to package it even how to make it structured or to you know that there's the sort of problem of structured and unstructured data in terms of just the the level of sophistication needed to turn something into something useful i think it's beyond the capability of any one organization unless they're incredibly forward thinking.

1:02:34Chris Woodcock:Yeah, I mean, if you're in a position to bring structure to your data without being aware of the end use, then you are going to be prepared for the future. Someone will find value in that data and the value of your company will increase. But that is a kind of luxury position to be in. even we sometimes at TFG are pursuing experiments that are probably too far in the future and aren't going to be applicable to any client in the next year and may go down as wasted R &D effort you know a lot of experiments that I've done over the years I've found the timing just isn't right you know and it might mean that something doesn't sell to a client but it also might mean that something's just technically not feasible.

1:03:25Chris Woodcock:It's not going to run on mobile devices right now. It's not going to provide value to the user in solving a problem. So we're at a really interesting point where sometimes we park ideas. When we fail, we don't bin something. We just know the timing isn't right. And I've experienced reactivating parked ideas because suddenly there has been a massive, say, technological advance in hardware and something has become viable so um that is a kind of node i'd encourage every company to adopt right now yeah yeah yeah

1:03:57Richard Ayers:no fascinating well listen i want to just finish off i'd say thank you for that because it was really great working with you guys and it was really interesting seeing how you took the initial brief and turned it into something and again that wowed me and i thought it was it was a really great experiment and has created a whole load of forward-looking questions now which i'm very keen to go and sort of pursue. But thanks so much for all the work you did on it because I really appreciate it.

1:04:24Chris Woodcock:It was really enjoyable and I can barely remember the long hours. I just remember the feeling after the event. But, you know, I'm looking forward to doing the next one where I can feed in all of the code used to provide the infrastructure behind that event. I can feed in the feedback and I can have AI generate the next version of it.

1:04:46Richard Ayers:you can sit at the side just drinking a martini now you don't you've done all the done all the work that's always the plan yeah

1:05:16you