Thirty years ago, the PC revolutionized technology, and fifteen years ago, it was the mobile phone. Now, in 2025, we are witnessing a similar transformation with AI, marking a significant shift in the industry every fifteen years. Without understanding the AI trends, one cannot predict what next year will bring. Planning a business for next year without knowledge of these trends is pointless. Therefore, we have prepared the 2025 AI Trends course. It clearly outlines the technological evolution of AI and the changes in the industrial landscape that AI will bring, as well as the key points that corporate members need to consider. It is not faster than before, and if you ask for a summary like an email, it won't help. This is why you cannot use general service frameworks. It is essential to recognize that the markets are different. The AI service market and AI application market are poised for explosive growth, leading to a reorganization of industrial order. The AI agent market is opening up new opportunities, and it is crucial to set goals using the right tools in this new era.

It's about realizing this; we need to be able to use tools like this for the AI economy to thrive. OpenAI has released GPT-5.1. There are meanings behind the name, and it can handle multiple inputs such as sound and images. Previously, there was GPT-4, and then GPT-4 Mini was released. Before that, we had GPT-4 Turbo and GPT-4, as well as GPT-3.5. Now, we have GPT-5.1, which is not just the next version of 4.5 but a completely different version. This is a significant departure, and I believe it marks a historic turning point in the development of large language models. For the average user, the maximum usage limit has reportedly increased from 30 to 50 times a week, but many people still find it challenging to use that much, unlike before.

There have been dramatic changes, like when images were not processed and then suddenly they were, or when calculations were not functioning and then they started working. From the perspective of an average person, it seems like AI is thinking and responding, but it also feels like not much has improved. So, what exactly has changed? 

First of all, the target market for this technology is not general users; it's businesses. Currently, OpenAI's GPT has revealed that there are ten million paid users, which includes individual paid users. There are also enterprise accounts, which are more expensive and offer better features, such as information protection for corporate data. There are now one hundred thousand enterprise accounts. It's important to note that the target market is different. 

It's not necessarily faster than before, and if you ask it to summarize an email, it may not perform well. Therefore, the improvements are not as significant as one might expect.

To give an example, it can be interesting. In what I’m introducing now, there are terms that only Koreans can read, commonly referred to as Air BM Beach. Honestly, I couldn't understand them at first. There are phrases like never stay here. The reason for this is that the owner of the house rented out on Airbnb might dislike receiving negative reviews about their property, and they might not understand at all. Even the most advanced AI translators in the world cannot translate these phrases. However, Koreans know these expressions and use them among themselves. By doing so, they can avoid such places. This is something that only people who speak Korean can do. However, in 2023, it has become possible to translate even those phrases accurately.

It means that when a human user asks a question, the AI itself poses a question before answering. To answer this, it breaks down the logic into several parts, analyzes them, and then asks itself relevant questions. For example, if three apples cost 600 won, how much would five apples cost? Up to now, it absolutely does not perform calculations. Many artificial intelligences do not possess mathematical abilities. Therefore, large models can only make inferences. When a token, which is a word or a group of words, appears, it simply generates the next word. For instance, when asked about the price of five apples, it might reference statistical data on what the usual answer tends to be. There would be statistics indicating that one apple costs 300 won, leading to a total of 1,500 won. This is a statistical figure. So, stating that the price is 1,500 won is not necessarily the correct answer. This phenomenon is referred to as hallucination.

So, up until now, we have been talking about the probabilistic answer that has come up the most in the prices of five apples. But what it asks itself is what the price of one apple is. If we think about it, three apples cost 600 won, so the price per apple comes out to 200 won. Remember that one apple costs 200 won. Now, what would be the price for five apples? Well, since we already have that answer, we multiply it by 5, which gives us 1,000 won. That’s how it can arrive at the correct answer. It’s engaging in logical thinking, right? This demonstrates the logical step-by-step thinking process that humans have, and we can say that this is the biggest feature of 51. The most important thing is that the training data it received from the beginning is structured in a step-by-step manner. According to Semaphore, they employed about 1,000 developers, and while OpenAI didn’t hire them directly, they outsourced the work.

After analyzing all the development codes that exist for humanity, we broke them down into a step-by-step structure. For example, when performing a calculation, there can be a division function and a multiplication function. We have separated everything by function. Coding is still being completed effectively, but hallucinations are inevitably present when developing code. Therefore, a senior developer needs to review it at least once. Of course, the review function will not disappear in the future, but the frequency will significantly decrease because development can now be done step by step. In the fields of mathematics, physics, and development, step-by-step learning data has been created, and we have used this for training. This report has already been released in January of this year, and related research work was published last year in 2023. Therefore, in this industry, we have Entropic.

I believe that companies like Google DeepMind and others were aware that this was something they would be pursuing. I also think they have just begun their journey. The reason I consider this a historical turning point is that until now, a lot of effort has been put into learning. The developments such as GPT 3.5, GPT 4.0, and GPT 4 Turbo are all about increasing the size of large language models. While increasing size doesn’t directly correlate with performance, it does generally improve performance in a relational sense. You may have noticed that version 3.5 performs better than 3.0, and similarly, 4.0 is superior as well. As the versions are updated, the size of the language models increases, and the number of parameters they can handle grows exponentially. Consequently, GPU computing power has also increased tremendously.

As needs arise and updates are made, costs continue to increase, leading to a situation where the competition among large language models has narrowed down to about six or seven contenders. Until now, the competition has focused on how to significantly expand the training data. Another aspect of this competition has been reducing inference costs. In other words, the cost of providing answers has become much cheaper recently. Companies using API Pro are aware of this; the API costs have indeed decreased significantly. They have been reduced from one to one, which is a massive drop in API expenses. In the B2B sector, this has enabled much greater use of generative AI due to the competition to lower inference costs, while at the same time, the costs of training have increased to create much better large owner models. This has been the underlying fact of the competition among large owner models in the market, and now, inference costs are...

The costs have become extremely expensive. The API costs are very high, around 5 won, which is quite a lot. Therefore, typical services cannot afford to use 5 won. Instead, we focused more on enhancing the inference this time. It's not just about making inference cheaper; what's more important is improving the accuracy of the inference process. This also involves computing resources, which inevitably increases costs. So, we raised the pricing accordingly. However, this is meant for those who can generate added value even if the inference costs increase by about 20 times compared to now. We have essentially defined it this way. That's why paid users also have a limit on the number of questions they can ask. The reason for this is that inference costs are significantly higher, so we encourage selective usage.

You were informed from the beginning, right? Yes, yes, yes. If we say this is a step-by-step learning process, for example, until now, everything was learned all at once, but now we can think of it as being divided into stages. That is important too. Nowadays, some people divide it into six stages, while others divide it into seven stages. Usually, the three stages refer to domain expertise, which is the specialization in a specific field. In this case, with O1, it has clearly acquired expertise in mathematics, physics, and computer science right from the learning stage, and it can be seen that it has been developed appropriately for this. Domain expertise has definitely begun to be secured this time, and we can see that these aspects can be expanded in the future. OpenAI is currently receiving investments, and there are talks about collaborating with companies like Apple as well.

Right now, we are lined up like this, and this time the value of memory is about 150 billion dollars, which is around 200 trillion. We are in a situation where we are trying to surpass this nonsensical space. This is significant in itself. The ability to infer step by step or through a chain of thoughts is part of the Strawberry project, which continues to develop. Now, in the next phase, with GPT-4 and 5, my service will be Onion. I’m not sure if it will become 5. It remains to be seen whether this part will adopt any current trends or if 5 will take a separate path, and whether Onion will emerge as the next version of 5. I think we need to observe this a bit longer. Now, the development trajectory of large owner models has two paths. One is that until now, large owner models have developed significantly with each version upgrade, and their capabilities have also improved.

여기에는 학습 단계에 어마어마한 투자가 들어가는 천문학적인 수가 들어가는 경향이 있었다. 뛰어난 추론도 중요해. 모드를 만족시켜 주는 추론의 완성도가 아니라, 컴퓨터 공학과 수학, 물리학 같은 분야에서 리서치에 일하는 사람들에게 도움을 줄 것이다. 하지만 더 중요한 것은 컴퓨터 공학과 컴퓨터 프로그래밍에 절대적인 영향을 줄 것이라는 점이다. 그래야 AI 응용 애플리케이션의 시대가 열릴 수 있는 것이다. AI를 활용하면 당장 부가 가치를 얻을 수 있는데, 개인적으로 번역해 주거나 요약해 주거나 정리해 주면 시간을 절약할 수 있을 것이다. 도움이 되지만 사실상 측정하기 어려운 경제적 효과가 나온다. 그러나 컴퓨터 공학을 만약에 받쳐준다면, 지금도 많은 분들이 버인 페이스부터 시작해서 코 파일럿을 비롯한 개발을 하실 때 프로그래밍을...

현재 AI 생성 기술이 많이 사용되고 있지만, 그 품질을 훨씬 더 높이고 프로그래밍의 복잡도를 증가시키면 이 분야가 완전히 변화할 수 있습니다. AI 서비스 애플리케이션 분야는 아직도 작습니다. 이 부분을 확대할 수 있는 기회가 있으며, GPT가 지금 51을 발표하는 것도 그 연장선상에 있습니다. 이렇게 레버리지를 주면서 앞으로 AI 서비스 시장과 AI 애플리케이션 시장이 폭발적으로 증가할 수 있으며, 여기서 산업 질서의 재편이 일어날 수 있습니다. 흔히 말하는 AI 에이전트 시장을 열어 주는 데 중요한 의미를 가지고 있다고 볼 수 있습니다. 말씀하신 대로, 사실 이 기술이 추론을 하지 않지만, 단계별로 생각하고 작업을 수행하여 우리에게 대답을 해 줄 수 있도록 프롬프트 엔지니어링을 하는 사람들이 필요합니다.

There were people who created that service, and the market could collapse as it emerges. As AI development tools advance and prompt engineering evolves, it's important to consider what markets will ultimately be disrupted. The era that allowed for the emergence of individual developers was created by Apple's iOS or the Android SDK, and another factor is the cloud environment. In a cloud environment, even if I lack backend knowledge or server expertise—though it would be better to have it—using tools available through Amazon AWS, Microsoft Azure, or Google Cloud, services can be developed by one person or two, whereas in the past, five or six people were needed, and teamwork was essential. This shift has been made possible by cloud technologies. Such tools are important, for example, cursor libraries.

When creating a system, there are various tools needed. At that time, this program has already learned what tools are available. It has been fine-tuned once. Then, when it comes to programming to achieve a specific purpose or function, it calculates the most effective order in which to utilize these tools and generates the program. For example, if we want to create a customized ERP system for our company, rather than using an existing ERP, we might consider tools like Cursor or Replit, or the new service called Agent Force that is being released by Salesforce. These tools are put together to generate the programming. Although we don't know when the ERP market will completely change, it is certainly challenging the ERP market, with Kakao Pay in Korea being a representative example.

Alternatively, there is a service called Clara in the European market where you can pay with Naver Pay. Clara has been using enterprise solutions from Salesforce and Workday. There are systems for enterprise management, such as HR management systems or the RPA market. In places like insurance companies, banks, and hospitals, automation is applied. For example, when an insurance claim is filed, it is automatically classified into a specific case, the document is scanned, and it is determined whether it was genuinely issued by that hospital. Once that judgment is made, the case is forwarded, and the insurance payment is automatically processed. This is all part of RPA. The market for this is enormous. Companies that create RPA solutions and consultants who provide consulting services make a lot of money. There are consulting firms like Center and many individuals as well. For example, they need to analyze the company's operations thoroughly to automate them according to that analysis.

They provide convenience, so both the workers and the customers need to have convenience. We can create this in advance with Salesforce, and we can customize it to fit our company using Azure services. However, what Rana has declared is that we don’t need all of this; our developers will use AI, specifically AI-related tools, to build everything. In running a company, we often use various SaaS solutions, like ERP for certain tasks, document analysis for others, and different messengers for communication. If we don’t do it this way, we would have to develop everything from scratch, which would increase development costs. That’s why we used SaaS in the first place. But now, the message seems to be that it’s hard to develop, so we just need to develop what fits us perfectly. This is the direction Clana is heading in. Of course, there are critical perspectives on this, especially since Clana is approaching an IPO and is focused on cost-cutting measures.

Is that why this is happening? The development market, which is receiving such help, has matured to that extent. After having a meeting, it’s possible that 5 won came out, and with that, this market is gradually becoming more robust. Now, this means that there is indeed a very large market in Korea, where the SI market exists. Although Korea is transitioning from the SI market to the SaaS market, fundamentally, the SI market is about companies needing certain functions, and those functions are being custom-developed by external companies. There are countless companies in Korea, like Gasan Digital City, that specialize in this. Large corporations, such as Samsung SDS, also have their own tailored solutions. This has led to the emergence of such companies. These SI companies exist globally in various forms and are significant in size, and there are also many consulting firms involved in this transition to SaaS.

There are many companies that provide services similar to sales posts, and one of the challengers in this space is an AI programming service company. Here, you can directly use tools like Copilot or GPT-4 for development. However, there are many hallucinations, and even if you create something, it may not be tailored to our company. This is because the AI does not accurately understand the tools we have. It doesn't know what our company's CRM or accounting tools are, so it provides very general development.

Using the library, we will update our tools and create a solution. Then, you can take this program for less than a SaaS offering. Of course, Replicator Cursor is also a SaaS program, but by doing this, you can take all the coding and implement it in your company, whether you use it on Amazon or in the cloud; it's up to you.

The video that became a hot topic features a person from a company called Cursor, who seems to have done some marketing. In August, they filmed a process where their 8-year-old daughter developed something using natural language, the kind of language we commonly speak. They encouraged her to create a game that she likes, and within just three minutes, the child made the game. The child asked AI how to implement a certain function, and the AI quickly provided the code. She simply copied and pasted it, completing the coding process. However, there is a bit of marketing behind this. The father of this child works at Cursor AI, which is currently known for its AI programming services, and he used Cursor to create this project.

In the back, there's GPT included. Claude, and so on, this is the kind of content that is included. However, in this context, it is being supported much more by the recent developments, indicating that the AI programming market can grow. We often refer to the AI agent market, and it is said to be becoming really important. So, if you look at it in this context, you can understand the value and significance of 51. Therefore, I believe that those in the corporate sector should definitely test 51 at least once. For example, if you want to buy a suit, how great would it be to have a custom-made suit? But it's expensive and takes a lot of effort, so people usually buy ready-made ones. They either buy a standard solution or pay a little more to have a tailored solution that fits their needs. But now, you can just click.

By clicking and simply entering my body measurements, it automatically creates a tailored fit and sends it to me. This means anyone can easily have something that perfectly fits their body. If we look at the current state of the AI market, it suggests that the previous SI market or SaaS market could collapse. Rather than a collapse, we might see a reorganization of the order. For example, what used to cost 10 billion won could now be entrusted to an SI for 2 billion won, which is a possibility. The SaaS market could also be reorganized. As you mentioned earlier about tailor shops, it's very accurate. Wealthy individuals order custom suits from high-end tailors, while the general public tends to wear ready-made clothing. There are also low-cost tailor shops available. However, now we can access high-end tailor services at affordable prices. This means that instead of wearing ready-made clothing, everything is becoming custom-fit in the ready-made clothing market.

The cloud market, the SaaS market, and the computer programming market are all very large. The SaaS market exists within this context. We're saying that we use SaaS in the cloud, but it's not like we have off-the-rack solutions. Instead, in the cloud, we will wear tailored suits from a tailor shop. However, the cost is cheaper than off-the-rack solutions, which is the most important point. You could have bought off-the-rack solutions because they are made by very skilled professionals. The off-the-rack solutions are produced by people who are often just as good, if not better, than those in tailor shops. This is because the best security experts and developers come together to create those tools, so the quality is never compromised. However, for those who wanted to use customized solutions but couldn't because of the high costs, they can now use these off-the-rack solutions, allowing companies to significantly reduce their expenses.

Companies want to customize their services as needed, adapting to the pace of their growth, changes in the company environment, and even the possibility of pivoting or adding new businesses. Each time, upgrading services can be quite expensive, especially when it comes to programming and maintenance. For instance, if a young individual is particularly skilled, they can modify ready-made suits; they can change the sleeves and adjust the length of the pants. This adaptability can be utilized as a sort of edge service. Thus, the custom market has entered the ready-made clothing market, which could potentially be cheaper than off-the-rack options. This is the expectation, and it is often referred to as the AI agent market. When the topic of AI agents comes up, people generally think that if AI agents are applied to our services, consumers will find them very convenient. However, AI...

Creating an agent is not an easy task, and you might think that. However, there are ways to make it a bit easier. AI agents can exist in both the B2C and B2B sectors. Depending on the classification, there are software agents, agents embedded in robots or robotic vacuums, and various types classified by model size, shape, and purpose. Therefore, academically, it is quite challenging to neatly categorize AI agents. But if I want to explain agents to you in a simple way, think of them as AI that can be used as tools. In software development, the ability to use tools is very important for achieving certain objectives. For example, a New York Times journalist traveled to Norway for the first time, and at that time, AI...

I used an agent service to go on a trip once, and I wrote an article about it. There are various services like Mindy and 100kc, and I used two of them. For example, if I want to go to Norway for four nights and five days, the agent will work accordingly. What they do is check the person's calendar app to find out when their vacation is. The person also makes a request, saying something like, My budget is this much, and I would like to stay in at least a three-star hotel. Once that information is provided, they will search for flights on sites like Hotels.com or Booking.com. They will then generate an itinerary that matches the person's preferences for Norway. Additionally, they will check if the museums are closed on Mondays.

After confirming whether it will arrive on Tuesday, various tools are used to make the schedule truly feasible. Moreover, it even sends text messages. This is because the app is still being developed, and the journalist went in June. So, what happens is that it sends a text message reminding you not to forget about the museum visit scheduled for 30 minutes later. To make it on time, you need to leave now. Also, you have a sauna appointment in two hours. If you want to make it, it's better to leave your current location and go to your accommodation first. If you want to go directly, you should leave by a certain time. It sends text messages like this. Using text messages is one of the tools. AI uses various tools, not just answering questions like before; this is what makes it an AI agent.

You can define it, and using this tool, you can realize your goals. I see this as extremely important. Based on the learned data, it has been compiled into books, and if we consider that until now, generative AI has been created by various developers who have developed tools that humanity has previously created, we can utilize tools like search engines, messaging apps, calendar apps, and map apps to share results with us. What this means is that we are evolving. Just as humanity has evolved, AI is evolving, and this is how we can utilize tools. I believe this will lead to the emergence of the AI economy. For example, mind trips are free. Why is it free? Because I can book hotels and purchase flights through it. Then, I can receive purchasing services there, like when planning a wedding.

Wedding planners don't actually pay for their services. They reserve our events and take their commission from that. So, the mindset is that a wedding planner becomes a travel planner, right? However, I believe that wedding planners can also evolve into AI agents in the future. This is something that is continuously emerging in the BQC realm. They don't just handle reservations; they also send alert messages and notifications tailored to the client's needs. These systems are being developed into apps, and this is currently emerging in the B2C market. Siri, for example, utilizes Fed UI and AI technology to create apps that can access data anytime. This also involves using tools, which is very important. It’s not just about providing a general methodology to achieve a goal; it’s about connecting APIs.

I simulate using all the tools to see which one yields the best results. By repeatedly simulating, I can determine that this approach is preferable. In professional English, this is referred to as system programming logic. However, the logic that can operate this system was originally created by humans. Now, AI can accelerate this process significantly. This does not mean that humans will disappear; they still need to oversee the work, as there can certainly be errors. Therefore, senior developers can work at an incredibly fast pace. I do not believe that AI will replace human developers, but I do think that we cannot ignore AI. By mastering AI, developers will inevitably be able to handle a much larger workload than before. This will lead to various competitive situations in the market.

The price is not determined by a single factor. For example, in the custom clothing market, one cannot assert that prices will drop quickly. Initially, when apps were being developed, external parties thought it would be difficult and thus paid a high price for the service. This was because there weren't many developers capable of creating apps. When smartphones first emerged, the cost of app development was extremely high. However, that price has been continuously decreasing, despite improvements in performance. Currently, there aren't many developers who can easily create software systems using AI tools like ChatGPT or Replit AI. Therefore, initially, there may not be any competitive dynamics in terms of pricing. However, once these elements start to become more commonplace, the overall costs of computer programming, including ERP systems, will likely change.

The costs of production and the prices of software that can utilize RPA may show a tendency to decline. This is why Salesforce has recognized a crisis. As a result, Salesforce is developing a new product called Agent Force, which will allow them to sell their enterprise SaaS offerings not just as off-the-shelf solutions, but in a customized manner. Salesforce is also advancing in this area, integrating internal capabilities like Curator and Ripple AI. They have expressed their intention to sell not only standard solutions but also tailored offerings. This indicates that Salesforce is reading the changes in the market. However, if version 51 has been released, and we assume versions 52 and 53 will follow, then the speed at which programming development in the computer sector is replaced will only accelerate.

So, what 51 symbolizes right now is that there could be 52. It means that there could be 53, and if we see O3 reaching 44, then developers will really have to work with AI; otherwise, they won't be considered developers. Even when using tools like Photoshop, it's clear that things have significantly improved. Companies like Salesforce need to be proactive, as if they stay idle, other companies could replace them. They will be the first to jump in and respond actively. I truly believe this will continue until 2025, 2026, and 2027. The SaaS market in Korea, the SI market in Korea, and the enterprise software market in Korea, as well as the consumer service market, will experience tremendous shifts. For instance, apps will be created that can greatly transform the landscape, leading to a plethora of new services.

It will come out, and perhaps I think we are entering an era where YouTube users can generate content. Now anyone can create apps and services through AI, and this could lead to businesses purchasing these at affordable prices, opening up a new market. In fact, we need to observe how the market will form a bit longer, but this time has laid the groundwork for the market to move. Traditional coding levels, such as those provided by models like Claude 3.5, have been surpassed, enhancing quality through thinking in terms of steps or step-by-step processes. Therefore, when O2 53 comes out, the market will inevitably undergo even more disruptive changes. How we prepare for this is crucial. For example, in Korea, services like Kers AI may emerge, and services similar to Replic AI could also appear, akin to SalesForce's Agent Force.

The same service could be offered by SaaS companies, and therefore, I believe there is a high possibility of a significant tectonic shift occurring. There was an expectation of a Cambrian explosion, right? When ChatGPT first came out last year, it was mentioned that it could serve as a starting point for this. To summarize, it has advanced significantly in the field of computer science, and because it can make step-by-step inferences, when an AI agent takes action, it can now think, as a servant, about how to effectively complete tasks on behalf of the owner. This means it has finally reached a point where it can consider that it can accomplish the task at hand. For something to change drastically, a lot of software needs to be released, and an AI agent capable of performing tasks accurately must emerge. The foundation for this is now being laid out. This is how it can be understood. And it is clear that Claude is also involved.

Anthropic and Google will also release products that are intertwined with the circle. We cannot let this slip away. Therefore, the step-by-step learning process, or the chain of thought, has been effectively implemented. I believe that even now, many are enjoying this segmented step-by-step learning process, including Llama and Anthropic. I think a new journey in history is beginning here, and this will lead to a competition based on quality. In the past, there was only a competition based on the size of the models from large companies; now, a competition to enhance the quality of reasoning is starting to emerge. I believe this will completely change the software market, the production market, and the development market used by all companies. This will lead to further advancements.

The market for apps that we can use is also likely to change significantly. Therefore, I believe that starting up for startups will become much easier. Whether or not there are developers available can affect the evaluation of a company or a startup, but now that may not be as important. Of course, if we take an example from a top university, I believe that much smarter people will do better. Those individuals will have their own areas of expertise and will be able to create much more advanced applications. However, if I want to create a consumer application but don't have a developer, I can outsource this service. Even with just a little development knowledge, one can create and advance a service. Of course, there may be differences when it comes to core competitiveness.

It is not an era without developers. However, we also have a considerable level of editing skills and capabilities, allowing us to create YouTube channels. The number of YouTube channels is enormous right now. It has been proven that only those from broadcasting backgrounds do not have to be the only ones creating content. Just as the number of YouTube channels is increasing, there will be a vast amount of software available as well, which will allow for personalized and tailored experiences for consumers. If we think back to the time when Pandora TV and Gom Player were popular, South Korea had already started the internet streaming market. However, it did not experience explosive growth like YouTube because, as you mentioned, only those from broadcasting backgrounds had the editing skills necessary at that time.

At that time, when I tried to edit something using the program available at school, it took several hours. For instance, if I inserted a one-minute clip and added some effects, I had to wait several minutes for it to process, so editing was quite difficult. However, with tools like Premiere becoming much easier to use, anyone could edit, which contributed to the growth of YouTube. Similarly, the tools being developed now are becoming easier to use, and thanks to this progress, we can streamline the process even further. In recent months, there has been a competition on how to use AI more affordably, with companies like Claude releasing new versions and Gemini launching their own models. Now, we are starting to see a shift back to models that allow for reasoning and step-by-step thinking, which other companies in the enterprise market can also adopt.

I see, you're going there now. I'm also trying to use the 5-won service, but I'm realizing that I can only use it for questions 30 or 50 times, and I'm not sure when I should use it. Since there’s a limit to how often I can use it, it feels wasteful. But if it can provide specific answers for certain functions, I would use it then. However, as a B2C consumer, I'm still unsure. On the other hand, it seems that there was significant explosive potential in the enterprise sector. That's right. Just as the mobile revolution was led by very easy development tools, we anticipated that the Cambrian explosion of AI services could start with the 5-won service. This could really change the landscape of software. Thank you for sharing these insights today.

Thank you for sharing insights that I never even considered. Thank you.

GPT-4o보다 느리고, 정리는 못해도 o1의 파급력이 더 큰 이유 (강정수 박사)

Technology(기술), Trend(트렌드), Transformation(혁신)에 관해 
꼭 알아야 할 지식.  

일하는 사람들의 지적 동반자 티타임즈TV. 

www.ttimes.co.kr

Contact
주소  서울시 종로구 청계천로 11(서린동,청계한국빌딩)
e-mail  ttimes6000@gmail.com
Tel  02-724-7718

GPT-4o보다 느리고, 정리는 못해도 o1의 파급력이 더 큰 이유 (강정수 박사)

GPT-4o보다 느리고, 정리는 못해도 o1의 파급력이 더 큰 이유 (강정수 박사)

스크립트