Good morning, everyone. And thank you so much for tuning into this webinar on point that we're calling Generative AI Beyond the Hype. We're really, really excited to share with you what we have been looking into for, I mean, every waking hour for the past three months. And hopefully you will get some new insights in this session today. We're actually a little bit curious about who has joined us today. So we would ask you to just briefly in the chat, write your name, your company and where you're joining in from, because I think we might have a bit of an international presence this morning. So let's see who's out there. And maybe while you do that, I can introduce who's in here. My name is Marie Gelber. And I'm Adam. And we are part of Implement's data analytics team working specifically with machine learning and artificial intelligence. And as hinted for the last few months, we're going to be able to do that. And we're going to be able to do that. And we're going to be the ones to take you through this session today. But let's see who's in the chat over here. I'm seeing a lot of Copenhagen. David from Zurich. Hello. Kevin Copenhagen. Amazing. Quite a lot joining in here. I see sweet South Africa. Hey, okay. Then it wasn't completely off saying it's an international presence. We also have Veil. Veil there. So we're far ahead. Exactly. Around in the landscape. But I think with that formal introduction to us and to you guys, let's just look a little bit at what we will cover today. So for those of you that signed up prior to Monday of this week, we sent out a survey because we figured this was a quite good opportunity for us to also figure out what is the current use of generative AI in the business landscape today. And we promised we would share those survey results with you. And that would be the first thing that we will cover. We will then take it back to the basics and explain what is generative AI and how does it work. And then talk a little bit about why are we talking so much about it? What is the hype really about? We will look into what is the value that it can bring to organizations? How can you actually apply this? And then we will try and look at some of the risks and challenges that you need to be aware of, but also how to mitigate them in the best way. And then we will look into the best degree. That's the agenda today. And I think let's just get started with the survey results. Thank you. So we sent out this survey. We got 57 respondents throughout the survey, which we're really, really pleased with. And the number one, the headline number for us was that 91% of you guys see this as an opportunity. And we are really, really happy about this because we absolutely agree that this is a massive opportunity for businesses. And this is a really, really high number. We have a lot of these similar surveys inside of implement with different fields. And I haven't actually seen anything like this. So seeing this many of you guys see it as opportunity and not as a threat or not as something meaningless to you guys is really, really powerful. And we are super keen to work with you guys on this today. The second headline number here is how many of you guys are actually using it. And we see this sort of bell curve shape. And a really funny thing is, as the response has gone in, this curve has actually been used. It's actually shifted a bit. The number of people who don't use it has dropped. And we are sort of also seen this with our work. So this is a change that's actually happening. But it's still about 40% of people seem to be using it to some degree in their work. And this sort of maps pretty nicely with what we're seeing. And it's pretty impressive. For a tool that's existed for just a couple of months across different industries, about 40% of employees are already using it in some kind of fashion. And maybe a little surprisingly considering that, the amount of leadership involvement and strategic work around it is very, very low. This is very much a bottom-up movement. A lot of people are finding it useful. Half of companies and organizations, everybody is self-taught. But also a significant portion are sharing tips and tricks. But formal training and guidelines are very, very rare. And of course, there's the question of jobs. That's really a big one in this field. It's displacing quite a bit. And we see a lot of people are seeing that. We see a small group who's concerned about a reduction in jobs, which is also something that is definitely worth considering. And finally, just 35% says that it won't have any effect. I think that's an interesting number. Because it means that about two-thirds of you guys see this will actually have effects on the labor market sort of at a grander scale. And then finally, sort of the major strategic outlook. About half of organizations have not considered AI strategies at this point. We're really, really impressed by the 38% that are right now considering an AI strategy. And there's sort of some spread out on the tail end. This totals to just over 10% who has various degrees of work on an AI strategy or maybe actually already have things implemented. So we're really, really impressed by these results. We think it's awesome to see this bottom-up movement. So people see a lot of potentials. And to sort of see leadership involvement start. But it's in the very, very early days of these strategic and guidelines that we're working a lot with. But if we take one step back from this and sort of ask, what is generative AI actually? Generative AI is this field that has existed for quite a while. But it's when algorithms and machine learning starts to create new content. You feed these massive algorithms huge volumes of audio, code, images, text, and videos. And they sort of learn the patterns of it. And the nature of these algorithms is that as they see the patterns and the data, they're able to produce more of it. The most common example right now is text. And you guys probably, a lot of you know ChatGPT. And the way ChatGPT has been developed is that you take large amounts of text and then you start removing words from it. And you start to ask an absolutely super, super powerful model to predict the next word. So it might look a bit like this. You have a sentence like, action speaks louder than words. Sorry, action speaks louder than house, feelings. The model will sort of try different things. But we all sort of intuitively know that the right word here is words. And sort of, and that way, when you scale this process, as you get a more complex model able to comprehend more and more language, and you have more and more sentences, it starts to learn grammatical structure. It starts to learn sort of the nature of language. But eventually, in order to get really, really good at this, it needs to learn facts about the world. It needs to learn sort of common sense and reasoning paths. And it turns out that this solution scaled, and it scaled massively. So things have really, really picked up in recent years. You guys have probably seen a lot of these models around. We said the timeline actually goes a little bit further back. So we had ChatGPT2 in early 2019. But really, after ChatGPT, we sort of saw this explosion of models arrive. We have access to things from Google now. And we have various startups coming online very, very quickly. The technology has sort of been in the lab for quite a while. But this was sort of when it entered into the mainstream. And we're seeing this plethora of models coming up. And we expect a lot of this to continue and a lot of different styles and types of models to come up. Yeah, it is quite impressive that we're seeing almost every other week now that some big model is being released to the public. And they're just getting bigger and better. And in some cases, smaller and better, which is also a completely different development field. So why is it that we're seeing this? Why is it that we're seeing more and more of these models? And what is really all this hype about? And as we put it here, it is because it is a new computing paradigm. This is a new computing paradigm. This is a new computing paradigm. And with the release of ChatGPT late last year. I think a lot of people they opened their eyes to the fact that you can do so many different things with having models that can understand language to the degree that we're seeing them being able to do right now. And I think just to understand why this feels so different than what we're used to, we borrowed a page out of a psychology book, actually, to try and classify intelligence. So in this particular framework, we're talking about crystallized intelligence versus fluid intelligence. And here crystallized intelligence would be stuff like facts, it's logic, it's encyclopedic knowledge, it's your experience based and it's memories. On the other hand, you have fluid intelligence. That is more stuff like reasoning, comprehension, problem solving, abstraction and creativity. And what we have been used to with the computers that we've had up until now is that they're very, very good at this crystallized intelligence. I mean, your Excel sheet will perform to the tiniest decimal point exactly what calculations you ask it to. Your ERP system can record all transactions. The crystallized intelligence has been very, very good in our computers so far. But you also know that these computers, they require exact instructions. This is the whole reason why we have invented code. It's so that we can give exact instructions to computers for the output that we want them to give us. What we're seeing now with generative AI is that we're moving a lot more into this fluid intelligence space. We no longer have to give exact instructions. We can ask ChatGPT, how do I calculate this in Excel? And it will just understand from your natural language. I think one of the punchlines or jokes that's going around is that the most hyped coding language of last year was English because everyone can interact with the computers now in a completely different way. And then, I mean, some of you might be sitting out there thinking right now, okay, fine, but is ChatGPT really reasoning? Does it really have a comprehension? And what about creativity? And to that we found a quote that is widely used in the computer science industry where we're saying that the question of whether a computer can think is no more interesting than the question of whether a submarine can swim. That is to say, it might not be actual comprehension, but it looks like comprehension and we can use that for something. This is why it feels so different. This is why we're looking into a completely new computer or digital capability. And why we're also thinking that this is why we're thinking that this is going to take an impasse in a lot of different industries in the years to come. And just so that it's not us only saying this is going to be widely used, we brought some numbers from Gardner. They made some predictions around generative AI. So Gardner is a world leading IT technology research and advisory firm. And I think they're actually kind of famed for not hyping things or having a real really pragmatic approach to new technologies. But when it comes to generative AI, they said that 30% of outbound messages from large organizations will be synthetically generated by 2025. And I think, I mean, the number in itself is quite wild, but this is a two year prediction. We're not talking five years, we're not talking 10 years. This is in two years that 30% of these messages will be synthetically generated. Even wilder, maybe they're saying that 50% of drug discovery and development initiatives will use generative AI will use generative AI also by 2025. And then by 2027, they're saying that 30% of manufacturers will use generative AI to enhance their product development effectiveness. If this doesn't kind of underline that we're seeing into a new era of the use of this technology, then I'm not really sure what would. So, okay, we might be seeing this having an impasse, but will we see a benefit from using these new technologies and tools such as chat GBT. And again, just to throw some statistics at you to underpin that point. We fresh off the press have this scientific study that was conducted by MIT. It was released in the beginning of March and it's called the Experimental Evidence on the Productivity Effects of Generative Artificial Intelligence. And what they tried to look into here is, can productivity increase with the use of tools such as chat GBT. And they took around 450 college educated professionals and then they had them solving two tasks with the first one, just solving it regularly as they would have prior to the release of chat GBT. And then for the second task, they asked half of them to use chat GBT to help them solve it. And they saw a 40% increase in productivity. And we've done our own little self studies with the tasks that we usually have in our day-to-day work and knew that we would probably see some sort of productivity increase. But 40% was way higher than I think any of us had imagined. And this is quite astonishing. More to it, we see a 15% increase in quality. So it's not because it's at the expense of quality that we're just doing the tasks a lot faster. We're actually also seeing an increased quality. So this really has the potential to be used widely and make us work in completely different ways. Yeah. So how might this actually look? You guys are probably all familiar with Siri and some of you out there might still sort of be thinking about traditional chat bots or these natural language interfaces we had for years using our cell phones. And in many ways, they've been the butt of a lot of jokes. So it's really important to know that while this might look a lot like some of the stuff we had previously, it is a fundamentally different technology. And sort of where does this productivity increase come from? We sort of have one example here where we asked chat GPT to write a email based on a set of meeting notes to drive a certain key point home. And of course, we should stress, be careful what kind of meeting notes you're uploading to chat GPT. This is probably more suitable for your Christmas lunch planning than sort of key business decisions. But it gives you an idea of some of the use cases there actually is for chat GPT. And then, and this is really not to drive any kind of hype home, but just an example of what some people are doing when they're thinking a few steps ahead. We want to show you an example of sort of the absolute frontier of the people working with this technology. It's a company called Adapt. It's one of the largest generative AI startups right now. And they made their own model called Act One. It's a massive language model. But this is special in that it's a multimodal model, a style of model that we will very soon get used to having more of, which means that it's able to not just work with text, but it's able to work with multiple types of media. Act One in Adapt's case is able to work with images from a screen, as well as text on the input side. And on the output side, it's able to create text, but also mouse and keyboard output. So what they're using at first, is that they are attaching it to a computer. And I'll just let the clip roll here. So what happens is, if the video starts, there we go. And the request of it to register a lead at a particular individual, and the adept system will look at the screen, and it understands the interface. This is Salesforce, for those of you who are unfamiliar, this hasn't been configured for Salesforce, it just reads the screen, it reads the buttons, it understands the pixels on the screen, and it's going to write the text, find the things and add, the lead. Now it's adding a order for 100 widgets. It navigates the interface, and it's able to do this. If you guys look up adept's website, you can see a lot more examples, where people are requesting it to find a home within a certain price range, and it navigates Google, it finds a website selling houses, start setting up filters, performing searches. This is sort of this natural language interface that allows the sort of configuration-free robot process automation. Adept is currently raising money. And they're raising a lot of money, very, very fast. Act one isn't publicly available, but we expect it to be publicly available pretty soon, probably. And this is the kind of disruptive technologies there is actually happening. This isn't a fancy tech demo somewhere somebody is doing. This is an actual company doing this. So it's really, really impressive. So that probably brings you to the question of how can you then leverage this technology in your particular organization? Because the example that Adam just showed, yeah, that's when you're taking a startup and that's just incorporating this technology into their core product, but you can actually also use this to gain quite a lot of value also in the small tasks around any organization. And I think the study, the productivity study kind of underlined that, but how do you actually interact with these models? So I don't know if this was just my LinkedIn and is currently still my LinkedIn, but I see so many different posts of people using chat GPT. It'll be stuff like six good prompts to how you can use it, or it'll be one specific case that's really mind blowing. But pretty quickly, we thought that we were missing kind of the structure of just categorizing how you can interact with these models. So we made a framework to try and put that into a bit of structure. And what we are saying is that we can divide the use of these language models into four different categories. So the first one is dialogue. This is the case where we're giving a prompt or an input to the model and we're getting some sort of expertise back. This is why we're seeing these called chatbots because it's the way that a lot of people are using stuff like chat GPT these days. You can use this to spar with an expert or brainstorm on a specific topic. And then I think also on top of mind for a lot of people is the generate content. This is also we're calling it generative AI. We're creating something. This is when we give a prompt to the computer and we're getting some sort of content back. And this could be that you're drafting an email or you're writing a LinkedIn post or similar. Now, the fact that these models have been so good now at understanding language means that we can also use them to analyze content. So this is more the case where you're giving it some content and then you're getting some expertise back. So maybe you want to extract the sentiment, the the tone of voice, the tone of voice. You do a lot of different really advanced analysis of content at this point. And then we have the final bucket where we're calling modify content. So you're giving content and then you're receiving content back. This could be the case that you're giving it a very technical text that you want to have simplified. You want to change the tone of voice in a particular email because you're changing the target audience or similar. So this is kind of to structure how and the wide range of possibilities of using these models. It's not just chatting with a chatbot. It's not just generating content, but it's much, much more than that. And when we're looking into how that can then be used in organizations, we put it into these two different pathways. So you can either use this to increase your productivity. This we've talked about, I mean, quite a lot already. Just the fact that you can do tasks better, quicker. You can get sparring in a different way than you've been able to do before. So increasing productivity. Then you can also use this technology to create new products and services. This is where we're seeing that the technology can go in and actually really disrupt different industries because all of a sudden you have the opportunity to just expand your product portfolio with new and different services. And that has not up until now been available to use in your company. So a couple of examples of this. We promised you guys to sort of show what does this mean within your organization. And of course, you guys are coming from a ton of different places. So it's hard to sort of tell you exactly what's going to be useful for. So we try to hit some functions that we know most organizations will have. So for instance, within marketing right now, the productivity boosting aspect might be drafting a new marketing campaign content. We see a lot of people working within social media, quickly changing our tone of voice. On the other end of the spectrum, actual new products and services might be stuff like virtual try-ons or simulations. If any of you guys have seen TikToks as bold glamour filter, that is a quite controversial filter, but it is an impressive showcase of the technology and what is actually possible today when you start taking it out. A secondary is HR. We have had AI-powered analysis of applications for a long time, but this is a completely different style of analysis you can do with these technologies. And on the sort of entirely new products and services, we're expecting to see AI chatbots as part of the recruitment process. Maybe have a half an hour conversation with a particularly tailored chatbot as part of that. Or maybe as a part of onboarding, having new colleagues have something they can ask about questions within the organization. Where do I find things? These are absolutely products that would be coming online. We expect to see from major HR software providers. within IT and engineering, we are already seeing a lot of code generation. We have Microsoft GitHub Copilot coming online that's able to produce large swaths of code. OpenAI, when they released GPT-4, showcased how it could produce entire website or small games outright from imagery or simple prompts. And on the other hand, we just saw ADAPT able to interact completely different with computers. ADAPT has a vision to make computer interfaces obsolete. and make documentation something merely for machines to interact with each other, which is a pretty radical vision of how our computing future might look like. Within the legal space, there's a lot of things where we can accelerate contract development, accelerate contract translation from one jurisdiction into another. And on the more radical side, producing entire AI lawyers able to do questions and answers, it probably isn't going to replace lawyers entirely. anytime soon, but sort of as a first line defense for quick basic questions. So just to get an idea about where you're going or if you might want to escalate this to a different party. It's a really exciting field and we see lots of startups in this field as well. And finally, within R&D, an existing use case we see coming up is getting overviews of huge swaths of data, asking generative AI to sort of look through tons and tons of research results and papers and highlight important stuff, We're going to request it for Seth Harvey antis� dice from Bosco who is your brain. As Irenehorni感覺 is already being transformed by generative AI. When things are moving very, very fast. But there is of course also a list of risks and challenges. This might be a lot of the different ways that you can use this technology, but there's also some things to be aware of. And we kind of try to highlight some of the most important ones here. And we will go through them one by one. So the first one that we're looking at here is what we're calling decouple from reality. These models don't necessarily have a sense of the world around us. So as Adam mentioned in the beginning, they work by predicting the next word in a sentence. That doesn't mean that they understand what is true and what is not true. So it can sometimes just make up stuff and it can make it sound very, very convincing. And to show an example of how that might look, I tried asking ChatGBT to write a short, a positive and engaging review of the book Generative AI Beyond the Hype by Marie Galva and Adam Hiddel. We have not written this book. I kind of wish that we had. But ChatGBT will give you a very convincing review of it. Apparently we have managed to write a must read for everyone where we tackle complex problems with ease. We give a realistic perspective on the potentials and limitations. I mean, now I know that we haven't written this book, but if I wasn't aware, I would think that this was a quite convincing review and a book that I might actually want to read. So being aware of this feature of these models and knowing how to overcome it by asking the right prompts, by making sure that you're fact checking, you're reviewing, is something to be aware of when using these large language models. While this is one of the greatest limitations of the models right now, it should also be stressed, it is also the strength of it. It's what allows the model to be creative and imagine solutions that doesn't exist yet. It isn't grounded in an existing database of knowledge. So it has these creative capabilities, but right now it is uncontrollable. So it is dangerous to use it without oversight. The second consideration is copyright issues. And we can sort of divide that into three sections. There is the copyright issues within the training data. There's the copyright issue within the models themselves. And there's the copyright questions of the outputs of the models. Delving into the nuances of this is complicated. We just had a paper released from the US Copyright Office trying to position this as something akin to a digital camera. But there is ongoing discussions in this area. And if you are within a copyright sensitive industry, it is the most important. So that is definitely something to look for. It's not going to be a deal breaker necessarily, but there is a lot of complexity in this area. And it's something you should be considering. The next part is the fact that these models can have a tendency to enhance biases. So the model is a product of the data that it was trained on. And quite a lot of the data that we have available out there right now, it can be quite biased. And sometimes these models can enhance these biases either quite obviously, and other times it's more subtle. So being aware of the fact that this is a risk is also quite important. And we brought another example in this particular risk. Because Adam and I, for another presentation, we wanted to create a beautiful illustration of leadership and used one of the image generating models that was available at that time. And this is the result that we got. Now, I know this is virtual and we've showed this a couple of times. So I have a pretty good idea that a lot of you are probably chuckling a little bit behind the screen right now, because this is just outright ridiculous, at least in our view. But it illustrates the point very, very well. All that we asked for was a beautiful illustration of leadership. And besides the kind of dictator vibe, we have a beautiful illustration of leadership. And this is the kind of a mafia. And there's the kind of a mafia. And there's the kind of dictator vibe that we're getting, maybe a bit of a mafia. And there's the quite obvious problem that in this case, leadership equals old white men. And I think we generated around 100 pictures or so. And it was more likely to create a male lion to depict leadership than it was to produce a woman. This is one of the cases where it might be right in your face, the biases that it's enhancing. But it can also be the kind of dictator vibe that we're getting. And it can also be a lot more subtle than that. So that's something to be aware of. And this is also something that is being worked on quite a lot in improving these models. I know you tried the newest or the upgraded version of this particular model, where we're not seeing quite the same horrible picture. But it is something to be aware of. It's most definitely an ongoing area of research and development. And it's not a fixed issue yet. Of course, there is the question of privacy and security. Right now, the terms and conditions of many of these models are complicated. And while there are open source alternatives that you can run on premises and your own hardware, they might not be as powerful quite yet. And just the security side of it, there's a massive, ongoing, super fascinating field of people trying to jailbreak these models or use them for things they weren't supposed to. We have our own cybertech services team inside of Implement working with many of these cases. And it's a fascinating world. But it is something to be aware of, especially if you're doing customer facing models. Where you let the general public access some of these things within your systems. Your attack surface looks quite different. And for internal cases, this privacy question, what kind of data goes where, is definitely also something to be aware of. Then we brought this and wanted to highlight it. And it's the risk, at least to be aware of the dependencies on third party providers. And these models right now are being trained. If you have enough capital, you can train your own model. But that is quite a heavy financial investment. But the models behind the tools such as ChatGPT is being exposed or being published through publicly available APIs. And you can use that and you can generate your own solutions on top of it. That just means that you need to be aware that now you're dependent on a third party provider. As Adam mentioned with the summarization of the meeting notes, you need to be aware of what data you're sending. And they will be using. They will be used for further training, at least if you're using ChatGPT. They're promising right now that through the API solution they won't. But there is also the case that they can just shut down their services. They can change the pricing structure. You will have a sort of a dependency. So just to be aware of this and how big a risk that plays into the solution that you're trying to create. And of course there's a risk of losing uniqueness. These models are statistical by nature. So if you don't know what you're doing, if you're using this sort of headlessly, you might see some really, really odd patterns. And actually we brought an example of this. So if you take the word robot and just add it into the model DALI2, it's one of the more powerful image generation models, you get this image out. And while this is a cute robot, if you just did this to get a robot, you would get something very, very similar looking. So we're sort of, one risk is that we'll see the reoccurrence of the white stick man that we all know from Google image searches for various corporate things. Of course, this is something you can get around in various ways and we'll get to mitigation solutions in just a bit. But be aware that the statistical nature of these models might actually provide a challenge if you don't know what you're doing. Yeah. And also just, you can get around this. I mean, it's, they're trained on so much data, these models that if you give a very generic input, you have the risk of getting a very generic output back. So it's just as, it's a new skill to get, knowing how to interact with these models to spark the creativity and not just get into a more mundane and similar looking outputs. We brought ethics and transparency as well. It's important to mention it. When we mentioned some of the use cases, for example, in HR, using it in your recruitment process, introducing AI might be actually more objective, but we were also just talking about enhancing biases. So is this really a process that you can use your AI solution for? And then there's the part about transparency, because if you're using it, to create or to create or to make a lot of decisions, the transparency and exactly why these models choose the way that they do might be lost in the process. So if this is an important part of the solution that you're trying to develop, then just have some considerations around the effectiveness of using generative AI. And finally, there's disruption. Disruption happens in many different layers, but we just talked about a potential 40% productivity increase. And with the good old Clayton Christensen book, The Innovator's Dilemma in Hand, we all sort of know that everybody loves a 2, 5, 6, maybe 12% productivity increase. But if you're looking into very large productivity increases, 100 or 200%, you're looking at a disruptive new technology that is going to change the conditions of the conditions of a certain market. And there is a risk, and we don't know yet which markets are impacted how, that markets will change dramatically because productivity output of employees changes so radically. If we're looking into actual 40% productivity increases across a wide swath of professional services, that is going to be quite impressive. So knowing how to handle this disruption, how to adapt a business model if required to contain some of these new opportunities, and not get disrupted and displaced by either startups or competitors leveraging technologies faster is a real, real serious risk. So that was just a collection of the different things to be aware of. But we have of course also thought a little bit about how you can mitigate quite a few of them, if not all, of course, depending on what you're trying to develop. So when we're looking at initiatives and we want to reduce risk, accelerate business impact, the first thing right now that we're kind of recommending everyone to do is establish some guidelines, some corporate guidelines about how you want to be able to use these different tools. It might not be that you're developing your, the next new big service, but the availability of the models such as ChatGPT just makes it unlikely that your employees or your colleagues are not using this already. And if they're not aware of some of these challenges that we're looking into before, you need to establish some guidelines. What kind of data do you want to send through these models? In what different processes of your business would you allow them to use it? I think we're seeing some companies just doing outright ban because they can't really get their head around how they can establish these. This is very much not our recommendation because these tools will just get more and more integrated in our society. So just getting some clear guidelines from the get go is a much better start. And then the next part is specific training. We asked in the survey how many of you have received training or that your company is doing something structured around it. And there wasn't that many that is saying that there is some structure to this right now, but this is a new digital capability. And as we mentioned before, you can, you can get really, really far by knowing exactly how you interact with these models, but also just the inspiration of the different tasks that you can use it for so that you can actually leverage this productivity increase that we are, that we're seeing. You might say some of the best ways to get around a lot of limitations is to be aware of them and to work actively with them. And training is a really, really core part of that in the short term. Yeah. And this is what we're saying is kind of the foundation. We would recommend this to just about anyone right now. And then there's the part where these models can bring some accelerated value. And here we're saying get an AI strategy, get some clear, yeah, goal, image goals to, to where you want to take this and the, and what kind of solutions will fit into your company. We, we have, we've seen some of these risks before. I think when Adam talked about disruption, you can get really far by just having a structured approach to figuring out how is this going to affect our company and what is the direction that we want to be taking? What kind of solutions do we want to take? What kind of solutions do we want to bet on and which don't we? And then there's a mindfulness around the AI architecture. As we also mentioned before, we have this dependency to third parties. If we're just using the openly available APIs, the technology landscape around how we're using these AI tools and being mindful of that and putting that into a structured approach is kind of the next step. And then you can build your AI. So you can build your AI solutions. You can start having these solutions natively incorporated in the, in your business, employees using them on a, on a daily basis and just developing for continuous improved value. So this is the, the way to mitigate quite a lot of the risks that we saw before. Cool. So while Marie gave you sort of the roadmap, what, what are sort of five good steps to begin taking? We just want to leave you here at the end. We're sort of what are we actually doing? We're seeing an explosive growth in this area right now. And these are some of the things we're working with clients on right now. Of course, we have sort of the use case development aspect of it. A lot of companies are asking, what can we use this for? This is a tricky question. And while the productivity side of things are somewhat straightforward, the new products and services and understanding the limitations and opportunities of technology is something we're really, really keen to, to help companies. And we're doing a lot of that. Of course, there's the strategic approach. figuring out where to position yourself in the market. There's a complex field of build or buy decisions, especially considering the avalanche of startups coming in, leveraging and disrupting with these technologies. AI architecture questions where modularity is more important than ever due to the rapid clip of, of this technology's development and just pricing questions within API usage and what's acceptable for different businesses. We're working on productivity increases sort of more directly. This training bit can't be stressed enough and it can take many forms. Sometimes it's more focused on guidelines and ensuring quality. Other times it's more directly focused on how do we leverage this to accelerate our employees the most. AI operating models is within the same field. What's sort of our way of working? What are the rules about sort of how much review does a piece of content go through after it's been produced from the model? And these are sort of a lot of the fields we're working within. And it's really exciting to see this plethora of different opportunities. And we are inspired a lot about from our clients as well at some of the ideas they're coming with. So this is just to give you an idea about this is sort of, I would argue the bleeding edge of what's happening out there in the development aspects of it. That was it. That was what we have for you. And I think we have about five minutes left. And we actually want to open the floor to some questions. I see that there have been there's been posting some different things in the chat along the way. But now is your chance to ask that burning question that you're sitting with and then we will try and answer the best we can. So let's see. What is implement helping customers with in practice considering with in practice considering generative AI? There are obviously great limitations to what we can actually say there with regards to specific clients. But right now there is a lot of considering different solutions and prototypes drafting up how might that look. There's a lot of in practice setting up these guidelines figuring out the limitations of the technology sort of not containing this bottom up movement that's coming but sort of getting that bottom up movement in line and leveraging the power of it. And of course increasing also the strategic questions. Large companies are sort of seeing the need to get a structured direction of this and not sort of to be taken by surprise. Then we have one saying how has your productivity increased? And I think it's a matter of figuring out how to use these models that fits best into your working life. For me it's a lot about getting that first initial draft of something not starting blank page if I have to come up with a list of use cases or if I have to figure out a specific structure for solving a task. Some things to be aware of some things to be aware of in a certain industry. This gives me kind of the first draft and then I can look at that and get inspired from it. It's kind of my brainstorming tool and that has definitely increased my own productivity. And I think just to sort of go in the opposite direction because I'm pretty good at getting that first draft out there but I'm not a perfectionist. So it's been hugely helpful for me to take that first draft and say write this more clearly, write this more concisely and sort of having that sparing partner to help me sort of get to a more more polished product. And it can really honestly do both. It depends a lot on your personal use case. Yeah. And then I guess this is also new. I mean I'm still continuously figuring out how to best use it. One of the other areas is just coding with ChatGPT is extremely helpful. It's a debugging tool like no other and that has also really increased productivity. Then what is your number one AI productivity hack we could implement today? We have such a list of different ways. Do you have your number one? I think the most powerful one for me is sort of a new one. I use it a lot sort of whenever I have something difficult I need to consider. Sort of where, how might I formulate this or how I want this particular thing or to happen or communicate this point. Just to sort of start writing out the question to the machine and figuring and in the process of writing that because it is so believable that it's actually going to produce good output. Sometimes I solve the issue right there just in the process. Just like you would do if you were talking to an actual human being. But having the ability to sort of whenever you stop there and say okay I don't know where to go from here. Just hit enter and have the machine continue right there. I think that is what is currently empowering me the most. But it's hard to cook it down into one thing. It's a plethora of different. Use cases. But I think that kind of stuff I hadn't thought that possible. And it really is. And it really works. And I think maybe the biggest hack or if you can even call it a hack is when we're looking at the numbers from the survey for example and it's saying that I think it was 18% that haven't tried it and then there was around 40% that has tried it just a few times. My biggest hack would just be try and start using it. You will find so many more ways of how this works for you if you just go out there and get your hands on the keyboard and figure out how you can use it. I want to grab the last question because I think it's a very very interesting one about suggesting approaches to evaluating success in generative AI projects. It's a really difficult question. And of course there's a large part of traditional success measures and value realization. I think we're increasingly seeing is that there's a sort of independent value in these generative AI projects in having the innovative approach getting it in the hands of employees and getting the experience just because the tech is so powerful. So I think you should measure the impact on two fields and sort of a traditional value realization approach but also considering the innovative potential that's available in the technology. I think that's available in the technology. I think that was it. Our timer has reached 0000. So that was all the time that we had. We will look at some of the other questions. We will reach out to you guys. And by all means, this is just a field that is moving so quickly and we would love to have some chats about how you can use this. So feel free to reach out. And thank you so much for tuning in. Thank you so much. See you guys out there. Bye. Bye.