Opening Remarks

Yeah, good evening, and thank you for inviting me.

Background and Experience

Yeah, really, I wear, well, a lot of different hats. I'm in, let's say, field of AI, seeing it was not popular yet. It's a little bit more than two decades now. And yeah, I was making my own project business in the field of AI and then worked as an investor as well. Consulting and Advising in AI

These days, mostly my day-by-day job is I'm consulting, advising in this field. Applied Research in AI

Current Focus in AI

And also with my team, we perform applied research, trying to answer question whether something is already able to be done and implemented into real life business, real life business processes. And that's a question that I hear more and more. And we'll try to... Well, touch that a little bit today.

AI Narratives and Trends

And that's very interesting because we were not connected before with the previous speakers. But if you look at it like at all that speeches, on the abstract level you'll see that it's all the same narrative about AI today that sounds very interesting so I did this as you can understand this picture was created by AI and the main idea is that in artificial intelligence now as the let's say family of technologies, just a set of technologies.

It's now ruling us, but it's still a unicycle. It's very easy to fall down when you're trying to invest in it and you're trying to use it in business. Well, because it's not there yet. We don't have a lot of defined solutions, a lot of defined cases that you can use and you can broadly implement, especially in the latest large language models. That is still to be developed.

Understanding the Gartner Hype Cycle

the more or less common business model for that kind tell me who knows that graph who has seen that before not many of you but very good that you did some of them this is Gartner hype cycle they developed it I think in the early 90s and started to use that that graph actually shows the how every particular technology actually being developed mostly in the minds of crowd not from the universities not from technical perspective from the minds of the crowd when some new technology appears in this world it goes on the first slope usually from universities or science group that developed that and media started writing about that and saying about that so people start knowing it more and more things that we're discussing right now been there approximately 15-20 years ago now we're talking about that very loudly But at the same time, the transformer architecture of neural network actually were published in 2017, so not long ago. And so every new technology goes on that graph faster and faster.

So the first slope that you can see, it's actually a place where all the investment starts. grants that goes from special institutions or very early pre-seed investors that are trying to catch something new that will give them huge growth the peak of inflated expectation from investment perspective That's around a position. Everyone talks about that. This is very modern, very popular. Everyone wants to put some dollar in that, expecting 10 times return. What is important and what's the catch? That actually the second slope starts from there as well. And that's the slope where the technology should find cases and implementation in business.

So they're really money acquired on the top level when it's very popular should be spent on testers, on MVPs, on the pivots. And if that money actually spent correctly, the company, the project can go to the slope of enlightenment through the draft. that's a very important idea because when you look at it that graph was developed by Garton not just out of nowhere not out of limbo for nothing not about just media coverage when you look at it you understand is it worth investing or not and if we're talking not from investors perspective but from a business from a big business for corporate perspective actually they approach it the same if you're trying to invest into something that goes on the first slope on the top you need to understand that the risks are high and when you invest there you need to be prepared that it goes nowhere and you just need to like loss unfixed losses but then you need to understand that when during that time you find special case and for corporations it's much easier to understand and much easier to implement ideas and test it in their environment let's say so when we're talking about investment into these days LLMs by corporations that a lot of companies are trying to do it's still over there on the peak and we're just starting to find cases where we can use it in business that really makes economical impact on what we do and we need to be ready that most of money as an investment field, most of money goes in vain but some of them will get you very high return.

Innovation Through Bundling and Unbundling

Now, if we're talking about investment, we're talking about new business and new cases, actually you can find both in either bundling or unbundling functions and features. Well, you put multi-touch and macOS on the phone, you get iPhone, and it's a whole new generation. This is very easy. All the technology business built on that.

Every interesting decision, every interesting thing that was made in the field of computer vision now is being added into Adobe Photoshop. They're adding everything in, they're making like a combined solution, raising the value of the tool that they offer.

At the same time, well, If you look at like a decade ago, you'll see that every Unix function, well you know what a Unix is, every Unix function actually became a separate company, just like every Excel template. Now it's a small, or maybe not small, but SaaS company that you can use. Just compare that. Next time when you have some free time, open Excel, open new file, and see templates they offer. And then just Google it. For every template they have, you'll find several SaaS projects that you can use instead of Excel. So money comes from either bundling or unbundling. And that's still the case. And for AI, it works approximately the same.

The Impact of Platform Shifts

Yeah, that's, again, going into chat form changes a lot how we work with the computers. Because, well, let's start from early 90s. That was web. web led to Google which is aggregation then we came to more media platforms say Facebook which is more recommendation and now we're in a moment where we can just tell computer what we want we don't need pages We need nothing. And how business will be built over there, we still don't know. But that's the idea of that platform shift.

And I think that's why you came today over here to know more about that because we all want to know and nevertheless I spent like 25 years in this business I'm still curious what's going to happen tomorrow so we already know that the first thing when we want to get something to create something to know something we need to go to some LLM chat

Challenges with Large Language Models in Business

but from the business perspective we need to understand what the limitations are for the LLMs large language models that we can use that are actually most popular well first you cannot add the solution based on large language model as it is now into real-time operations for example you cannot plug chat gpt into your phone conversation into customer support you can plug it into chat if you use like chat conversations but for for phone calls you cannot it's not fast enough at more more than that the time of generating an answer is not predictable at all

so you cannot use it like that so if you want to use that possibilities you need to do something else well something that you can do is here but there's more again it's not easy and almost and it's not possible for example to use ChatGPT, Gemini to train on your own data For example, if you have a set of documents and you want your employees to work with the documents, but in the form of dialogue, for example, they can ask questions about something that is within those set of documents. You cannot feed the system, you cannot feed Gemini with those documents. Because the window of contacts has limitations and again, it's available only from the Cloud and there can be sensitive information. So you need to find workarounds to use different solutions that probably not so powerful as the other systems.

okay this is very important the way I'm talking today about business investments so that's from business perspective so what can we do ah hallucinations yes definitely that's something worth mentioning I didn't write here anything I just highlighted there just remember this is very important note over here

Of course, you need data and you need training and you need verification of the result, especially when you use it in the business, in the process, where that should be there. So what can we do with that? Yeah, I'm sorry, the slide's actually a little bit not the way I made it, but you can see. So there are several approaches that we can use.

Strategies for Implementing AI in Business

to make that thing work first of all that's the easiest of them all and that's the easiest because we actually we actually teach kids that way because we tell them we show them task and say I solve this task this way here is another task of this type show me that you know how to solve that like I did that

so chain of thoughts that's a very easy to use just read about it a little bit and you'll see that even your current work with ChargerPT or Gemini that you use right now becomes more powerful you get more interesting results and more of that you understand the reasoning of the network you train in this little task but you train the system to reason so you understand why this answer is like that it becomes more predictable another thing is actually being mostly used by developers but also very useful to like autoGPT for example makes your GPT your colleague prompt engineer it makes the prompt engineering more automatic it helps you to build context it helps you to get better answers I will not be reading this you saw that the next idea that I'm trying to pull here is that it looks like we need some approach

The Role of Autonomous Agents in AI Deployment

that explains how to use that mighty tools into real business and how we surrounded what kind of technology we need to surround that to get the good results and that's not something new there's approach named autonomous agents we all understand that we already know about that a lot we already read about autonomous cars and that's a real example of autonomous agent but now we understand that we can use the same approach into business that is mainly the activities around data, activities around information because most of business this day has become IT business and most of business is just moving information and making decisions

and again the cycle repeats so now we can build autonomous agents around the larger language models we can use several of them it helps us to tools that that helps us to connect verification things and approaches how to train systems to reflect on itself and get the self-critics so I'll be finalizing with this there's a link this pretty interesting article about autonomous agents, so that explains how to build that.

If you want as a business invest and use LLMs, that's probably now the only approach that you can use and be sure that it adds value into your business, into your process. What it gives you, I'll just mention here that long-term memory is actually experience that was learned, fixed as a set of documents, as an experience of people. The system can be trained with it and use it also as part of a context. The short-term memory is actually information and data performing particular tasks that have been performed in the current time being. So the tools that can be can be also implemented, gives more context. The modern LLMs, they have a huge window of context, and it's just a pity not to use it all.

So we get all the information, all the necessary algorithms we can implement over there as well. And the most important part of autonomous agent atability is ability to plan. It also includes that chain of thoughts thing that I also explained, and the reflection and self-critics.

Well, we don't have much time here, so we'll not be expanding on that more, but my... well go here was to just to show the direction where to dig more if you want to go there and again agent inside can have more than one LLM it can be several LLMs working together now when we cycle back and almost the beginning of my speech

When we're thinking about bundling or unbundling thing, or we're thinking about business processes that we want to automate, we probably need to switch into description not the process but roles in terms of autonomous agent because part of what is the main parts of every process is the people employees working and performing that process with or without help of computers so when we're trying to describe and explain the work of every employee of ours as an autonomous agent? thinking about what kind of experience one uses what kind of information for every particular task requires what tools one needs and other things how result is being verified and how is playing and what actions to be done so when you go there and start thinking about employees or your clients as autonomous agent you start understanding how do I jump from that second slope to the next race to the plateau of effectiveness and creativity and that's all for me thank you