Vojtěch Komenda: ChatGPT is not enough anymore: Discover what AI can do today | 184
- Martin Hurych
- 18. 3.
- Minut čtení: 25
AI is no longer the future. It is the present. And those who ignore it will soon have nothing to deal with.
We all hear about artificial intelligence at every turn. Almost everyone is already paying for ChatGPT, but the true potential of AI in businesses is still only realised by a few. Companies often end up with experiments that dont pay off. They then walk away from AI unnecessarily. Yet right now we are on the cusp of a fundamental change. AI is changing the game as fundamentally as the internet or electricity once did.
That's why for this episode I invited Vojtěch Komenda, owner of AI Excellence s.r.o., who deals with AI deployment every day. Vojta not only knows how AI works, but more importantly, he can clearly demonstrate how to practically benefit from it in your business. According to him, AI is not just a "new toy" for IT enthusiasts, but a strategic necessity that companies need to understand and master quickly - if they want to be able to keep up with the competition in a year or two.
Therefore, Vojta and I discussed not only general trends, but we went directly into practice. I was interested in where AI is helping companies today, how quickly you can get a return on your investment, and what pitfalls await you. We also discussed whether it makes sense to regulate AI and why Europe desperately needs its own strong AI model if it doesn't want to lose out to America or China in the new era.
In the episode you will find out, among other things:
🔸 How specifically can AI save a company money
🔸 Why are most companies just getting with ChatGPT
🔸 What are multi-agent systems and how can they dramatically increase efficiency
🔸 Where exactly does the deployment of AI make sense - and where is it unnecessary
🔸 How to quickly gain a strategic edge over your competitors
🔸 Why is it important to have our own AI model in Europe
If you've been thinking of AI as just another trend, it's time to change your perspective. Believe , this episode could be one of those that will significantly impact the future of your business.
"The AI is there and by ignoring it you are treading on your luck. It can help you and it's not as hard as you think. Just start using what's available. Implementation is neither long nor difficult.“
Vojtěch Komenda | Managing Director@ AI Excellence s.r.o.
Today we're going to talk about artificial intelligence. I'm almost certain that this topic couldn't be erased from your bubble, so I've brought in a guest today who is involved in the deployment of artificial intelligence in companies like yours.
Who is Vojtěch Komenda and what does he do when he is not working on AI?
Martin Hurych
Vojta is the owner and managing director of AIExcellence. You are in an AI environment from morning to night, what do you do when you turn off the AI and just get out into normal life?
Vojtěch Komenda
I have 3 kids, so I try to spend all my free time with them. Of course, I have some hobbies that I have to put on the back burner a bit now, but as the kids grow up, they each like a different one of my hobbies. My oldest daughter and I are doing astronomy together, my middle one is singing and into music, and my youngest is getting very excited about catching a basketball. I'm looking forward to when she's older and starts school, that we can start playing together.
Martin Hurych
It almost seems like only the youngest basketball player can't be helped by artificial intelligence yet.
Vojtěch Komenda
As for , of course the oldest one is already playing with it. When Daddy gets tired and lets the youngest tell a story with the AI, he likes that, he can influence where the story goes and he makes it up.
What does AIExcellence do?
Martin Hurych
If there was anyone among our viewers and listeners who, for some mysterious reason, hadn't heard of artificial intelligence, what does AIExcellence do?
Vojtěch Komenda
AIExcellence is trying to help companies make the transition to the new AI economy that awaits us all. It's a thing to come and it's only a matter of time before it comes to your industry. It's a change similar to when the internet came along, a change similar to when electricity came along or when the steam engine came along. It's a change in the way that business is done and things that didn't go before suddenly go and things that used to cost an awful lot of money now cost a fifth. It's a change in that business environment so big that they're going to be dealing cards again now like they did in 1850, 1920 or 2000. These are the changes that are coming and they need to be dealt with.
Why aren't companies deploying AI en masse yet?
Martin Hurych
They talk about incredible possibilities, they talk about making a lot of things cheaper, they talk about the possibility of replacing people we don't have. Yet it seems to me that a lot of business owners often see the deployment of as a sprint rather than a marathon. We try something, we dive in, then we find we don't really know which way to go, we shut it down, and I don't really see any focused long-term experiments. How do you see it as someone who has a company to do it?
Vojtěch Komenda
It depends on what kind of company and how much it is under the influence of other market pressures, because today's companies have a lot of other things to worry about. Czech firms are dealing with ESG type requirements, they are dealing with high energy prices, they are dealing with things that relate to the automotive industry, which is in a bit of a crisis at the moment.
All these influences are hitting these companies and now on top of that there's some artificial intelligence and everyone is telling them they should deal with it. At that point, they say they'll try it, they'll give it priority number 4, 5, try something, then they come to some conclusion that maybe it helps somehow, but that they would imagine something else. The conclusion is then that they tried it and now they wait six months.
Extra AI is not just that you pay for two ChatGPT licenses. ChatGPT is such a basic way of knowing what it can do, but today it can do it six, eight, ten times more, and it's all about first identifying where it could help. So you need to do a little audit internally or externally, go through the company and say what are the specific processes where it could help us now. We can't try it on the first idea that comes to mind, but we'll try it on the idea that ideally is relatively cost acceptable and that has a relatively quick return on that investment. Almost all investments in implementing AI in a company have a return on investment of under one year, which is usually two, three, four times faster compared to any other investment. But just because something works with AI doesn't mean it's a good idea in your company. Out of the 30 ideas you come up with, it's a real good idea to do 15 and start with 3, 4, 5.
What can we practically solve with AI today?
Martin Hurych
I, when I listen to podcasts or read articles about artificial intelligence, a lot of times I'm at that level like you just said. There's actually nothing wrong with it at all, we always have to audit. What I see there is that a lot of mainly more experienced, older owners actually have no idea and can't imagine what it's good for. So they can't say, based on some audit, let's try to deploy it here. Can we say, is there some practical masterstroke that we can already solve with AI today, some real use case?
Vojtěch Komenda
There are such use cases. There are some that are general, and there are some that are specific to that particular company. I'm going to try to pick one across all industries that almost all business owners are familiar with, and that is central purchasing. Most of the larger companies already have some sort of central purchasing, one, two, sometimes 5 people that handle that situation. The standard thing is, when you need to get something, it depends on how much, how much time you have to do it, how much it burns you, how much it hurts you if you can't get it in a given amount of time. Depending on that, you're under pressure to take maybe the first offer, the second offer, the third offer of that option.
For example, try to find a company that will offer you marketing services when you have one person in central purchasing. You give him the task, he basically googles marketing services Prague, Ostrava, Brno, Pilsen and 5, 6, 7 different agencies come up. He clicks on one, looks, takes the email, sends them an email and waits for a response. He does that with 5 and says that's probably enough. He waits a , two days, then two replies come in, three replies come in, and they ask about 50 different things. He may have some tender documentation, he sends it to them, but basically he reaches out to 5. But why doesn't he reach out to 20, why doesn't he reach out to 25? It's because it's administratively demanding, they're going to respond to him, he's going to have to do some mega spreadsheet and suddenly an action that's supposed to take him an hour or two becomes an action that takes him 3, 4, 5 working days.
This is exactly where one of the possible uses of artificial intelligence comes in. Artificial intelligence can do exactly the same thing as the human in central purchasing and can identify 20, 25 companies that can offer you that service. Based on that, you're able to get double, triple those possible options. First of all, then you have the ability to push the price, and you can get something even faster, because if you need a truck transport from A to , for example, it is not easy to find a company that can do it and can deliver the truck the day after tomorrow. But the AI can do it. It will find you 25 companies, reach out to them by email, if they don't get back to you within 6 hours, it will call them and tell them you have a request and if they have a truck available. At that point, you're able to get what you need in significantly less , at significantly less cost, and the conclusion of that is that maybe you're able to reduce your purchasing costs by 5, 10%. Assuming your margins are in those exact ranges, you can double your profit. That's something . Plus, it also helps you in situations where you're, say, an engineering company and you're making some turbine that you need to ship. Now you get a part from one of your suppliers that doesn't match in quality and you're not able to put it into that particular assembly. At that point, you are faced with the option of calling India, Turkey, Tunisia, El Salvador, telling them that you will deliver the turbine a month later. But when you deploy AI, you can get those inputs in significantly less time, you can reach out to companies that maybe you haven't reached out to before, and you can try to get that part significantly faster.
Martin Hurych
So am I already using a repetitive process that I have somehow deployed in the company, or am I using AI as a one-off tool?
Vojtěch Komenda
This use case I've listed here is a set of individual agents where each gets a specific task. You basically make a way that you first identify what you would like and some first AI boss assigns the task to an agent that can search the internet. I'll give a task, I need marketing agencies in Brno or trucking in the area of North Bohemia, Germany and Poland. He will generate search criteria, 6 search criteria, two in German, two in Czech, two in Polish and search for companies. It puts together some 30 trucking companies and then it contacts them and puts together some 30 trucking companies.
Then you reach out to a second set of agents who basically crawl their website. They'll see if website exists, if the first one hasn't made up what they're offering, how wide their fleet is. You call in another set of agents and the next set of agents prepares 30 email outreach to those individual companies, some in Czech, some in German, some in Polish, and sends that out. They'll wait for the result, and as soon as a particular response comes in, they'll put it back on that list, and you'll come back to it maybe two days later and ask how you're doing. He will tell you that out of the 30 contacted, you have received 10 responses from these companies and this is the response. You are in a hurry, so you select these three and request quotes from them. He basically prepares the emails again, sends them out again, and that's how he helps you manage the process on a significantly larger scale. This is one of the uses of multi-agent systems that we put together in our company.
Martin Hurych
Is based on ChatGPT?
Vojtěch Komenda
This is based on the APIs offered by each model. Most often it's ChatGPT, but it can be any other model. We can that another model gives us better results, for example in search we can use Google's Gemini, technically we can also use DeepSeek and possibly other tools that are . If the company is concerned about its data, which is a very common scenario, then in that case, the internal local system that the company has, for example, in its infrastructure, can be involved in exactly the same way. I always say that the model is the outlet and we're putting the appliances together. At the moment we have 10 different types of sockets from different vendors and it's even possible that the company will make its own socket because it's afraid to use a socket that's provided freely on some shared storage.
Can I make multi-agents myself?
Martin Hurych
I try to make the Ignition as practical as possible. Since Ignition, like any other podcast, is primarily listened to in the car or while running, when I finish or run home, what can I get done by myself between now and tonight in terms of agents and multi-agents?
Vojtěch Komenda
As for agents, it basically depends on how far along the audience is in using AI. A large number of people today have already tried that by opening up OpenAI, opening up that Chat.com site, and asking a question there. That's the basic step. The next step is that you start structuring that prompt, which means your prompt doesn't have one line, but it has maybe 6 or 7 lines. You start using that
"Act as", basically you put him in a role, then you tell him what he should do and you start programming him, you start telling him what to do. Then you take it to the next level and that is you create what OpenAI calls GPT models, it's also called assistants, it's called copilots, each company calls it something different. It's basically that you create somebody who already knows a prompt and behaves in a certain way. You can add one last thing to that, which you have the ability to add directly in ChatGPT, for example, and that's to add what's called actions to it, in the API it's called functions. Basically, you teach it to call some additional service, some additional technical API. It's not super complicated to build a technical API because you just have to find somewhere on the internet how it works and then there's an assistant inside of ChatGPT that pulls that data from that web page and connects it. So you can make a very simple agent that can, for example, look into your gmail account and pull up what appointments you have. It can call a Google search, it can send an email. That's still a level that can be inside of that OpenAI, even that solution is nice, but you're still in control.
You basically have to combine those 8 agents together or you take the next step and we'll be happy to make that call to AIExcellence. It's a combination of those agents, dividing it up between them, like dividing up the work in an office. The boss gets some work from a client, in this case the client is a human being, and they divide it up between those individual subordinates. Those subordinates who can search, search, those subordinates who can compose emails and mail, mail and divide it into individual tasks. Basically, you create 60, 70 agents at a time, and those agents fulfill the goal that the person has set.
Martin Hurych
Do I understand correctly that the level I'm going to take myself to is one agent? I that I've been paying for ChatGPT for so long that I don't know what the unpaid version looks like. Are there agents in the unpaid version?
Vojtěch Komenda
In the unpaid version you can use some agents, they call it GPT assistants. The moment you add some additional functionality to it, even a simple one, it already meets the definition of an agent. In the non-paid version you can use them, but you can't create them yet.
What is pair prompting?
Martin Hurych
One terminus technicus in your preparation caught my attention, pair prompting.
Vojtěch Komenda
Pair prompting is a thing that is for people who are just starting out at that point. I did some first prompt and now it's still not doing I want. Very often people start prompting and after a while, ChatGPT or other tools don't give them what they want. At that point, it's very important in that company and in process to have somebody there to help you, to help you. Because if that help doesn't come within a day, then you shut it down and you go right back to the way it was and you start to figure it out on your own.
That's the thing that we see around us and that's why we basically offer the option, if you want it and you don't have that person in the company, we're happy to be that person and we call it pair prompting. That means that you try something and if it's not doing what you want, you come to us and we'll say how do you adjust it or we'll say you want something that that level doesn't give you yet and you need to move a level. You need to get an assistant, for example, or
then an agent or a multi-agent system, because the basic model doesn't do it yet. Alternatively, we'll tell you not to do it in GPT-4o, but to go to Claude or start doing it on OpenAI's o1 model and you'll get the answers and conclusions there.
Martin Hurych
What I've gotten for pair prompting from some courses is that I put one prompt, it returns intermediate results, and I put a second question right in the same prompt that references those results. In the definitions that they told me, the paired prompting was, search the internet, find 30 companies, and then based on those results, pick the best one for me.
Vojtěch Komenda
That means you give it to him directly in one prompt. This can be implemented, of course, I personally would have doubts about how the result will look. The fact that you are giving him these sequential tasks is basically some workflow, some stepping. But then even if he completes the first step relatively successfully, he then has to follow up on the second step, and if you do that within one agent, you can very easily overload him. That's a thing that happens very often, that if you keep writing more and more complex prompts, after a while you find that it starts to ignore your other instructions. It's already full of it, it's not giving it, and it's basically behaving very much like a human. If you give someone the 60th task, if they have 50 more important ones before that, it goes out of their head. That's exactly the solution, where you have the option of either improving the intelligence of that model to do that, which is what's happening, and that's what OpenAI and other companies are doing, they're trying to add 4o, o1, o3. That's great, but even so, a lot of those practical tasks that you want to do from A to Z, you need to divide up among more of those agents and do it in incremental steps. That's the solution that we do within AIExcellence in our multi-agent system.
When and what to choose the GPT o3 model for?
Martin Hurych
There are tons of options, you've listed a bunch of them here, we don't just have ChatGPT, we have Gemini, we have NotebookLM, we have Claude, however, before the shoot we were talking about landing paid versions of o3 in ChatGPT. When and what do we pick it up for?
Vojtěch Komenda
It always depends on the complexity of the task. Personally, I'm not a fan of these o1, o3 solutions for one simple reason. She's basically breaking down the task gradually and looking for specific steps, but when you're solving most common situations, that breakdown is very important there. I'm going to give you a task, search these 50 data protection sites for me, because I want to know how the Czech authorities feel about data protection. You're basically asking him to open up a particular page and search that page sequentially inside, and if you use the classic GPT-4o to do that, that the result is not the brand ideal. He'll give you some appropriate answer there, but that answer may not be true, and in many cases it's not true, it's the classic hallucination. It's because he doesn't have enough information at that moment to process it. If you drop o1, o3 on it, she'll think about it, correctly determine how she should approach it, but at that point she runs into the other thing again, that she needs some additional information in the course of that thinking.
If he can open those pages and search through those pages and pull up that information, that's when it starts to give some value. That's functionality that was deployed over the course of this week in the paid version of ChatGPT Pro, where they do what's called deep researche, which is something that maybe Gemini already does as well. The main problem with those o1 and o3 models is that they do break down the task correctly, but then they don't have the tools to look at it. Part of that might be that I need to look at our internal system, our SAP, which is something that one can do, but the o3 doesn't have a chance of achieving something like that. Once there's the ability for those o1s and o3s to pull that information from those non-public sources as well, that would be great. The o1 and o3 is great, for example, if you want to calculate what the irradiance of your solar panels will be, because you can already pull all the information about how the sun moves there. Assuming you're solving a more realistic problem and all that information isn't publicly available, at that point o1 and o3 don't provide as much additional input.
When to choose Claude, GPT and Gemini?
Martin Hurych
We're in the Chat framework, but if I simplify it down to just the three most common ones, ChatGPT, Claude and Gemini, what to use when for what?
Vojtěch Komenda
Claude is very good at programming. For people who program, Claude will be the best solution. As for Gemini, it's a little bit better in the search area because of all the power of Google behind it. ChatGPT is for everything else.
When does it no longer make sense to implement AI?
Martin Hurych
Now let's move a little bit to the philosophical level. When does it no longer make sense to implement AI, where is line beyond which it doesn't make sense?
Vojtěch Komenda
There it comes down to whether it's a technical sense or an interpersonal sense. I'm an advocate that OpenAI should never replace interpersonal communication. If you send an email to a company and ask if they can transfer two trucks from Liberec to Wroclaw, I think that's perfectly fine. Assuming that you telephone a restaurant to ask if they have space because you are looking for space for 30 people, I also think that is perfectly fine. Assuming that AI is going to be calling clients and offering them your services, I don't think that's okay. It's all about finding a way where it's beneficial to that user and where it's more of a benefit to the person who's not on the call. Where AI shouldn't go at all, for me, is in the area of some sales.
Martin Hurych
A lot of companies will disagree you on that, because I can totally see the enthusiastic legions of salespeople hoping to push a button one morning and get called and written up.
Vojtěch Komenda
They can have it subpoenaed and signed, that's not a problem, technically it can be done, but I believe that if I'm a customer, I want the other side to take the time to convince me of their product. All of that will be true as long as there is a human being on one of those two sides. Of course, it's a very real situation where you have artificial intelligence answering the phone for you, and it's technically possible to do that today, but the discussion between one voicebot and the other voicebot at point is going to be about factual arguments and not so much about emotion. Once we get to stage, even scenario makes sense, but what I am saying is that those people will only be willing to talk to the AI if they are immediately sure of what it brings to them. As long as the AI is buying, that's when it will work. If the AI is going to sell to a human, I think the success rate is going to be even lower than the classic human phone operators that are flying around.
I am currently still trying to convince myself about a nice use of time, which is related to finding a craftsman. It's not a good idea to use AI to summon craftsmen, because the people who have their hands in the fuses or are running water at the time are not going to talk to a robot, it's absolute nonsense. But how many times have you finally gotten through to that electrician and he said he can't do it right now, maybe next week and to call him tomorrow? If those people had the option that if they hung up the phone, their voicebot would take over, those people would be willing to talk to their electrician voicebot. They would describe to them what their problem is and he would be able to schedule a specific thing on that calendar. Then when that electrician would be leaving that job that he'd been doing all day, he'd just go and ask who called him that day and what work he might have for the next week. The AI will explain it to him, maybe even give him a straight quote, and if he agrees, he'll send
E-mail the client with some pricing and by the time the electrician gets , he's got a job ready for the next two days.
How does the younger generation view AI?
Martin Hurych
To what extent is this the world view of our age generation, even though you are much younger than me, and to what extent do people between 25 and 30 not care and see the world in a completely different way? I've had a very interesting debate this week about how these people potentially look at the world, and maybe the first version of virtual marketers, where it's admittedly artificial intelligence being tested on young people, is not received badly at all.
Vojtěch Komenda
It's possible. Let's take it that still for at least the next 10 years the main area of that business will still be for our age generation, in the B2B business 100%, in the retail business it will gradually increase. I think it's all to do with the amount of time you have for these situations. If you are young, you have more time and that means you are not offended by these situations. It doesn't offend you that you get a call from an assistant or that you get a call from a person offering you something and you are basically willing to listen to them now and then. The older you get, the less time you have, and in those situations you're quicker to put the phone down. That's been my experience in life, and I think when those people are 40, they're going to function very much like we do.
Is it a good idea to regulate AI?
Martin Hurych
Looking at your area, I was , how does a company deploys AI into companies look at the European Union's efforts to regulate AI?
Vojtěch Komenda
Of course there's a complication, all entrepreneurs are not happy about regulation until that regulation starts to protect them. I think we all feel, and we've all experienced it, that artificial intelligence, for example, can implement telephony in such a way that you wouldn't know it at first sight. You will know the you deviate from the standard scenario and suddenly put it in a situation for which it was not fully prepared. That's when you'll know, but otherwise the voice is fine, you basically can't tell if you're talking to a human or not. One of those European Union regulatory things is that you always have to know that you're talking to an assistant. In the case of chatting to someone on a website, that is basically irrelevant and not necessary, but in the case of making a phone call, I personally would want to know. It seems to me like a very reasonable requirement that those AI assistants introduce themselves and that at that point you clearly know that you are talking to a robot and that there is no human on the other side. That's a requirement that seems very rational and very correct to me. It also protects those users so that they know unambiguously who they are talking to and who is on the other side.
Another thing is that over-regulation stifles business. The way the AI Act is set up now, I think it does dampen some business, but I don't think it's terrible. Of course, we'll see what comes next, because the European Union can make a nice catcall out of a rational requirement, but so far the requirements are not completely stupid in their idea. One of the requirements is that AI cannot restrict people's access to the labour market. This is about the fact that artificial intelligence, by its very , even though it is statistical, naturally tends to put people into certain groups. Assuming that you outsource all of HR to AI, at that point it's not ideal, because those are exactly the patterns that we're trying to eradicate from that company instead. Automated employee selection is not prohibited, but it is regulated. You have to keep logs of what you've made decisions based on and how and why, which is not insurmountable, but I think it's worth thinking about.
We'll see what the next months bring, for me the demands are not completely illogical yet. Let's hope that we stick with it and it doesn't expand further in the form of some directives and other things that could fundamentally limit the business.
Martin Hurych
So don't you see that this is somehow holding us back from potential growth compared to, say, the United States or even China, even though we can think whatever we want about the model?
Vojtěch Komenda
At the moment, I think that the main obstacle is not the action of the European Commission and the European Union, but rather inaction. The OpenAI was created in the United States and similar models are being created here in Europe at a similar time. At the moment, America is moving at an incredible speed and China is trying to keep up. Here in Europe, we have companies that would be capable of doing something similar, but those companies are failing to get into the European market. Look at you and me, we have been talking about the American solution all day. Why are we not talking about the Euro LLM model that we could use?
When will we have our own LLM model in Europe?
Martin Hurych
I'm glad brought that up, because I was when we would have our own language model in Europe.
Vojtěch Komenda
I think that these foundation models, which are now springing up like mushrooms all over the world, to be local. It is also for the sake of having some strategic security. This foundation model will very soon become a commodity like uranium, oil or gas, and if we have the possibility of having such a commodity within the European Union, that is only a good thing. Of course, it depends where you draw the line, whether we have to have it in the Czech Republic or in the Central European region. I think that the European Union is a nice enough area and that we should have
at least one good competitive model. If we don't have one, then we should do everything we can to sure that there is one, in case there is limited access to those foreign models, which could very realistically happen to us, or it becomes too expensive as a result of some trade wars.
At the moment, of the European models that look relatively reasonable, it's the French Mistral model and the European Union-backed project, the Euro LLM model. It's a consortium led by Charles University, and in addition to a lot of good universities, there are mainly representatives of two companies that ensure that it's not just an academic exercise. There's a company called Aleph from Germany and then there's a very high quality company called AMD in Gen AI from Finland. These are companies that if I had to name three, apart from Mistral, these are the two in Europe that can create such an LLM model. So I think that we will have such a model, that it will be an open sourced model, that it will be a model that is realistic to implement locally and that we will be able to trust. At the moment we trust OpenAI as well, but we have to make our European socket.
Martin Hurych
How quickly do you see us having some Czech AI based on this model?
Vojtěch Komenda
It is technically realistic within a . Given that it is an EU project, will be a little slower, but I expect that we will get a solution within a year or two years. But I have similar input on this as you do. If you want more precise information, just call the Faculty of Mathematics and Physics, where colleagues are leading this project and where I'm sure they will have more precise information. I will be waiting with great enthusiasm, and the moment they make some outputs available, we will implement them very quickly in our multi-agent system and offer this as one of the other alternatives that will work.
Martin Hurych
I'm very glad you mentioned it because it's especially visible on the Chinese side, how we're all defending it and China is de facto admitting it de facto, not openly, for me these models are like a modern crusade. It makes you spread your world view. That is also why it would be nice to have a European model and to have the chance to tell the world how see it, however we see it.
Vojtěch Komenda
I agree. It's always the model, especially the language one, and it depends on what data you learn it on. That's a question I get very often, for example, for the very reason of whether the fact that we're using a little bit more of the American view today is something dangerous. China decided the moment OpenAI came along that they didn't want to use the American viewpoint and basically banned OpenAI and then made their own model that represents their worldview. I wouldn't use the word, I really think of it as more of a race to the moon like it was in the 60s, it seems a bit more optimistic than the Crusades option.
Martin Hurych
You can see it awfully well on the Chinese side, so maybe I chose too strong a statement. Anyway, if someone just clicked through this podcast and clicked through clicked through, came in at the end now and had to take away 3-5 of the most important sentences from this podcast, what would that be for you?
Vojtěch Komenda
AI is here, and whether you build it on the European, American or Chinese model, it's the kind of thing that can help any company significantly speed up processes, reduce costs and increase revenue. It's not as complicated as you might first think. It's not that you immediately start doing neural networks inside your company, it's that you can take those models that are currently available in the market and implement them into your processes. The implementation is not long, it's not that difficult, we're talking about months and hundreds of thousands of crowns. This is not something that a medium or small company cannot afford and can then recoup the investment within a year.
If anyone had such an investment opportunity in, say, the capital market, we would all rush into it. Artificial intelligence is something new and that means we all have different expectations of it. What I know about it and the way it works allows you to really streamline processes and gain an advantage in your particular market. But you must not delay it for a long your competitors are not sleeping. What will very realistically happen in your market is that within six months, within a year, within two years, someone will come to market with a competing product or one of the competing products will become cheaper. You have to be prepared what you are going to do about it at that point. If you don't do it now, it's going to happen in a year or two and the best thing to do is to be the one who comes up with the cheaper product and takes the market.
