Cheetah Mobile Inc.

Good day and welcome to the Cheetah Mobile third quarter 2024 earnings conference call. All participants will be in listen-only mode. Should you need assistance, please signal a conference specialist by pressing the star key followed by zero. After today's presentation, there will be an opportunity to ask questions. To ask a question, you may press star then one on your telephone keypad, and to withdraw your question, please press star then two. Please note today's event is being recorded. I would now like to turn the conference over to Helen Jingzhu, IR for Cheetah Mobile. Please go ahead.

Thank you, operator. Welcome to Cheetah Mobile's third quarter 2024 earnings conference call. With us today are our company's chairman and CEO, Mr. Fu Shun, and our director and CFO, Mr. Thomas Jin. Following management's prepared remarks, we will conduct the Q&A section. please note that the CEO script will be presented by an AI agent. Before we begin, I refer you to the Safe Harbor statement in our earnings release, which also applies to our earnings conference call today, as well as make forward-looking statements. At this time, I will now turn the call over to our CEO, Mr. Fusheng. Please go ahead, Fusheng.

Hello, everyone. Thank you for joining us today. Cheetah Mobile once again achieved accelerated revenue growth in Q3, driven by our service robotics and internet businesses. This consistent growth results from our strategies to expand the use cases of our wheeled service robotics and expand into overseas markets, as well as the resilience of our legacy internet business. Industry demand for service robots continues to rise, especially in the overseas markets and in restaurants, hotels, factories, and offices. Business owners are using robots more often to help their staff and improve efficiency. In the past weeks, I have visited many customers and partners in Europe and associations Asia. In fact, I am still in Europe today, meeting with our local partners to further strengthen our presence Building a strong local distribution network is very important for our global expansion, as it will set us apart from our peers. That's why I have spent significant effort on this initiative. During my conversations with local partners, I learned that our robots are helping them solve labor shortages. Some customers in Europe share that using Cheetos robots have reduced employee absences and turnover. Meanwhile, some Japanese customers told us our robots are much more reliable than other players' offerings and are switching to our products. In September, we launched a new robot for factory and fulfillment century use. This robot can autonomously deliver goods to move low-payload items from transit warehouses to assembly lines. We are currently optimizing the product to better meet the needs of customers in overseas markets. This highlights the importance of receiving feedback and inputs from local partners. We believe this product will become an important part of our service robotics business in 2025. While the robotics industry is still in its early stages, it will be a trillion-dollar market. Robots are becoming essential helpers for humans, and it will happen in the bell markets first. And LLMs will speed up this growth by enabling robots to do more tasks. and making them easier to deploy than ever before. Through our conversations with investors, we've noticed a lot of interest in how LMs are making our robots smarter and driving steady revenue growth. Today, I will share what we've achieved so far with LMs in our products and what's coming next. First, we are using LMs to improve the way our service robots interact through voice. Thanks to our strong far-field voice recognition, our robots already hear users well. Now with LMs, they understand users' questions more clearly and respond better, making the overall experience much smoother. For instance, in restaurants, our robots don't just deliver food during busy hours. They can also help attract customers, boosting the return on investment for restaurant owners by taking on more roles. And because LMS break down language barriers. We are expanding these voice-enabled robots to international markets. We are also working on agentless, a system that lets customers set up tasks for robots using voice prompts. For example, you can tell a robot to check each table at 2 p.m. to see if anyone wants to order more food before the kitchen closes at 2.30 p.m. The robot will go to each table. skip the ones without customers, and even allow people to place orders if the ordering system is linked without LMs. This kind of functionality would be nearly impossible or would require writing a lot of complicated code. Secondly, we are using multimodality models to improve our robots' indoor autonomous driving. One area that we are working is to enable our robots to map out the large factory as they move and look around once the map is ready. Our MV team can mark key locations, and the robots can then navigate the factory on their own to deliver goods. If factory owners want to change these key locations, they can easily update the map whenever needed. Based on our initial testing, thanks to LLMS, the time it takes to set up our robots can dramatically reduce from about two days to just two hours. We are already using vision-based autonomous driving technology in some cases and plan to expand it further. For instance, our robots use vision-based autonomous driving technology to avoid people and obstacles and understand their surroundings. Over time, we aim to achieve an end-to-end navigation system. This will allow our robots to handle more complex environments entirely on their own. Third, we're adding robotic arms to our robots to help them do specific jobs. Some of these arms can press buttons, which is useful for delivering things between floors, in particular in the overseas markets where business owners are reluctant to adjust the elevator access control systems due to security concerns. Others can pick up and sort items, which is great for use in factories. These arms are powered by on-device multimodality models, making it easier to automate routine tasks. When it comes to LMs, we use advanced models through API calls to support some of the features we have discussed. At the same time, we are also developing our own models. In November, we launched an 8x7 billion mixture of experts model, covering many languages, including Chinese, English, Korean, and Japanese. We made it open source and use it to power our robots, especially the agent OS features. Additionally, we have trained smaller on-device models to support indoor autonomous driving and control robotic arms, turning to LM-based applications. We recently introduced AirDS, an high-based data service platform to assisting enterprises in data and building prompts for their LM-based application. AirTips was built on top of our insights into developing LMs and LM-based apps. So far, we have received positive customer feedback on our LM-based application, and we will continue to enrich our portfolio. Our goal is to offer relatively standardized SaaS products, allowing businesses to use LMs gain efficiency. Before handing the call over to Thomas for the financial highlights, I want to stress this. Cheetah Mobile is in a good position to tap into the growing markets for service robots and LM-based apps. We've been years of experience on the PC and mobile phone areas, as well as expanding into international markets. And we have strong LM expertise We've already made solid progress in growing our revenue and cutting losses. This is just the beginning of Cheeto Mobile's turnaround.

Thank you, Fusheng. Hello, everyone on the call. Please note that unless stated otherwise, all money amounts are in RMB terms. In Q3 2024, our financial results demonstrated solid execution on two strategic objectives. Number one, accelerating revenue growth driven by both the sales of robots and the legacy internet businesses. And number two, enhancing our operational efficiency to reduce our operating losses sequentially. In the third quarter of 2024, our total revenues increased 16.6% year over year, marking the second consecutive quarter of accelerating revenue growth, compared to 11.6% in Q1 and 12.3% in Q2. Our real-based service robots continue to be a key driver of growth. Additionally, our legacy internet business remains resilient, achieving solid revenue growth and margin expansion. On profitability, we made further progress. Non-GAAP gross profit rose 14% year-over-year and 7% quarter-over-quarter to $131 million, with non-GAAP gross margin expanding to 68% in the third quarter, from 65% in Q2 and 63% in Q1. Non-PAC operating loss was $61 million in the quarter, reduced from $63 million in Q2 and $66 million in Q1. We continue to focus on managing costs and expenses. Notably, we consolidated the teams of Cheetah and Beijing Orange Star, streamlining staff and services with overlapping functions. For example, in Q3, we reduced cost and cloud costs, professional and legal service fees, and certain labor costs related to G&A and operations. We are also decisively investing in AI, using AI to enhance our service robotic business. For example, our non-GAAP R&D expenses increased 25% quarter-over-quarter in Q3, and now about 60% of our revenues are invested in R&D. As Fusheng has said in the past, while we focus on developing products that generate immediate revenue and profits, we remain attentive to the latest technological advancements. Another highlight is the continued strength of our legacy internet business, which grew 26% year over year and 18% quarter over quarter. Operating margin excluding share based compensation expense for this segment improved to 10% from 6% last year. As of September 30th, 2024, we maintain a strong balance sheet with cash and cash equivalents of 1,831,000,000 or US dollar 218,000,000. Long-term investments of about 886,000,000 or US dollar 126,000,000. In closing, we are making solid progress in expanding revenues and narrowing losses. We are confident in our investments in AI because we see the significant market potential of integrating RLMs into our service robotics business. At the same time, we remain disciplined in reducing losses and driving efficiency in our AI operations. Thank you. This concludes our prepared remarks. Operator, we are now ready for the Q&A session.

Thank you. We will now begin the question and answer session. To ask a question, you may press star then 1 on your telephone keypad. If you're using a speakerphone, please pick up your handset before pressing the keys. If at any time your question has been addressed and you would like to withdraw your question, please press star then 2. At this time, we will pause momentarily to assemble our roster.

Everyone, for today's call, management will answer questions in Chinese, and an AI agent will translate management's comments into English in another line. Please note that the translation is for convenient purpose only. In the case of any discrepancy, or management's statement in Chinese will prevail. If you are unable to hear the Chinese translation, I think a transcript in English will be available on our iRap website as soon as we can. Operator, we are now ready to take questions. Thank you so much.

Thank you. Moving the first question up. And it's from Thomas Chong at Jefferies. Please go ahead.

Thanks management for taking my question and good evening. My question is about Twitter robot business. In 2025, what are our goals in terms of volume, revenue growth, and revenue contribution from the robot business? Thank you.

Okay, thank you. I'll answer this question. In 2025, some of our specific growth goals are still in development. I've been looking at many channels overseas recently. I think we can define the overall direction in this way. First, Our revenue will grow by 2025. And in terms of net profit, the ratio will also increase. We will combine this wave with Q4. We have summarized a lot of overseas market experiences and made some specific goals. And then what we think is that because the robot industry is very hot in the capital market today, but in fact, its growth is in the short term. I don't think it will happen like that of the Internet back then. I also want to tell you that it should be a continuous document and then gradually accelerate the process. Because today the robot is in the combination of software hardware. I'm sorry, I'm a little busy. The combination of software and hardware, and the landfall of the physical body, and the construction of channels, are much bigger than the investment on the Internet. That is, the digital growth of that year was much bigger. So I think we will look at the machine more as a long-term goal. We may, for example, in three years, be able to make the total revenue of the entire company reach more than half. And then, including us being able to be the first few in the world, at least the first three, such a service machine provider. This is our big goal. And then the detailed goal, we still need some discussion and serious promotion. Thank you.

Thank you, operator. Can we move to the next question, please?

Yes, our next question comes from Vicky Wei at Citi. Please go ahead.

From the whole picture, the Internet is no longer a high-growth industry. It has become a basic industry. So, as I said, it is very difficult to achieve that kind of index growth. But the second thing is that the bubble has moved. As we all know, in the past few years, we have encountered the impact of the price drop. It is also gradually coming out. Some of the new products of the Internet in China, including products on PC, including the appearance of AIPC due to the large model. The whole market of PC is also warming up, so it has brought some growth to us. If we talk about the expectations for next year, I think it is like this. The Internet has become a basic industry. It is the same as many traditional industries. We are already in this new We found our own growth path in the environment. So we think it should maintain gradual and stable growth. It shouldn't go down anymore. But this is just my guess. Because we don't know what will happen in this world today. Without this big black swan, as of now, we are confident in the growth of the Internet business next year. But it should be a more stable and not so fast growth. Our company's internal Internet business Thank you.

Thank you. Please move to the next question, please.

Thank you. Our next question comes from Yanlong Chen with CITIC Securities. Please go ahead.

I would like to ask another question. In terms of training methods, do we have the opportunity to train our robots in the future to improve their intelligence level? In addition, I would like to ask another question. In terms of training methods, do we have the opportunity to train our robots in the future to improve their intelligence level? How big is the gap between human-made video learning goals and our domestic industry compared to overseas companies? Okay, let me answer that. First,

So you're asking a very complex technical question. I'm trying to use my understanding and our company's understanding to simply help you with some of the analysis. The so-called training of an engineer, your data is missing. It's divided into several aspects. One aspect is that we turn engineers into several parts. One part is called navigation. It's a small-scale literature history. This thing. Because its environment is relatively limited in a room, so before this, there was no large-scale model or transformer as the bottom layer. Through some SLAM engineering metals, it has been basically solved this problem. But because of the transformer, such a bottom layer data makes the indoor navigation more, how to say, that is to say, its implementation cost will become lower, its dependence on sensors will become smaller. This thing we are already doing. Yes, what we have been doing recently is to make the indoor navigation of our robots more visualized. The next generation of robots will also change to a higher-calculated chip, which is to make the indoor navigation of robots more visualized. You can also compare it to today's new electric vehicles. In the end, Tesla's visualized FSD has made a major progress. and add laser radars and other types of radars, multi-modal radars. In fact, it's not as good as FSD. The main reason is because of the appearance of Transformers. The mechanism of the large-scale model can handle many things from the bottom to the end. So this is one aspect. As you said, we have deployed a lot of machines. We have run in all kinds of places. This can actually achieve a lot of data. And to be honest, its route is not as good This public road is so complicated and the speed requirement is so high. So in this part, we think it's okay. The data is not a problem. Especially for a company that has a lot of robots landing and serving every day. We are quite confident in the part of indoor navigation. The second is that you may mention a concept that is more popular now, which is the concept of giant intelligence. I think this concept is more of a concept. What is the definition of GSM? There is no clear definition. Some say it's a humanoid, others say it's a wheelchair, or a double arm. The data on this aspect is indeed relatively scarce. Because in the past, all of them, including the mechanical arm of the factory, were not built on the core of this data drive. It is the core of the automation of this code drive. On this point, we think our ideas are gradually coming one step at a time. I constantly express a point of view in many occasions. I think that humanoid robots have a long way to go to become commercialized products. I don't think it's possible to achieve commercialization in less than five or ten years. Although you can see that it has a lot of demonstration effects, it has to become a commercialized product. We should have a long way to go. So we are more in a realistic way. We will combine our scenes. For example, we are in some mechanical land and the interaction of the real world. Some simple tasks are completed and started. I won't talk about the details because this is related to our technical route. Our idea is not to come up with a perfect product. Everything can be done and all the problems can be solved. Our idea is because we have a scene today. We need to combine more of the scene itself to complete the continuous collection and training of this data. This, we think, needs some time. It may not be as optimistic as what we are investing in today, but I think it will be done step by step, one scene after another. You can also see some outstanding companies abroad. Their mission on the so-called infrastructure is also very simple. But I think this way is easy to fall. Yeah. It's not more optimistic than this, because it's more of a three-dimensional machine and more of a mechanical structure. This is our judgment of a big technology trend. There is a big difference between overseas and modern mechanical companies in China. To be honest, it's not that big. Because today it's a large-scale model, including you see the large-scale model, and it's very fast in China. Because the underlying algorithm is just You can put it this way. Once the algorithm itself has made a breakthrough at the AI level, it will not be difficult for everyone to learn. The real difficulty lies in how the algorithm is engineered, how to train more data, and how to train it more efficiently. In fact, the domestic Chinese team has an advantage. At least there will be no gap in this kind of large-scale data engineering. So I don't think there is much of a gap today. In terms of training that already exists, including the domestic large-scale model, productization is actually pretty good in all aspects. If there is a real difference in some new paths, for example, the emergence of a new method, I think there will be some differences in this country. Operator, please move to the next question, please. Thank you. Our next question comes from Yanni Su with Hightower Securities. Please go ahead.

Thank you, Mr. Guan. We have seen that since the beginning of this year, the company has been continuously reducing losses in every quarter. I would like to ask if we have any knowledge of the rhythm and specific plans for the reduction of losses in the future, and whether there is a clear time to achieve profit. Thank you. Okay.

China China China China Due to our participation in the development and training of large-scale models, we slowed down our pace a little bit. But after this round of MOE, we will put more of our power on the development of agents and on the landing related to robot agents. This kind of development cost will be much lower than the past training of large-scale models. There must be a reduction of losses I think the market is constantly changing, and so is the technology. We may not be able to make a very clear timeline, because I think... Just now, a friend asked about the progress of the robot training. Because we have seen some commercial opportunities. We also see that the service robot, due to the value of the large-scale model, whether it is its underlying planning ability, task decision ability, including its interaction ability, we think there will be a good improvement. Then it will expand its satisfaction in each market. It will expand the market. Operator, please move to the next question please.

Thank you. Our next question comes from Suri Yin with Sealand Securities. Please go ahead.

感谢管理层给我提问的机会。 我这边有一个问题,主要是想问一下您如何看待AI Agent 以及这块的一个技术的难易程度如何。 比如我们最近看到智普的AI Agent已经可以帮助用户去在百度地图 还有像美团上面查询和下单之类的功能。 Then I would like to ask if AI Agents can speed up the big model's landing at the C-end. How should we think about the value distribution? Thank you. Okay, thank you.

This question is also very professional. I think the word agent is very popular recently. Actually, agent itself is more like some traditional... uh, uh, uh, uh, uh um um It has to achieve the traditional system's high level of stability and satisfaction. What does that mean? For example, I give it a command. Will it be able to do what I want to do? This is a key point of the large-scale model today at the C-end and B-end. This point is not easy to be mentioned in the industry. But we found this problem when we were doing it ourselves. For example, you use traditional code to achieve it. You choose a point to make a order. Because people are very precise in operation. Basically, your operation is 100% effective, right? Back to your last question, can we accelerate large-scale applications on the C-end? Of course. Especially on many C-end apps, some apps have already started to use agent. For example, for example, translation and education. This has obviously begun. Yes, I think value distribution will still bring a wave of the rise of app manufacturers. This is my opinion. But personally, I am not particularly optimistic about the C-end, the C-end agent landing in China. Because the domestic big manufacturers are too capable in this regard. And for them, it will be done very, very quickly. So if you do a small innovation in a familiar field, it should be followed up very quickly. Thank you.

Operator, please move to the next question. Thank you.

Thank you. Our next question comes from Yi Zhu with Founder Securities. Please go ahead.

Hello, Director. I would like to ask a few questions about large-scale models in the field of closed-end applications. Because we have seen that the company has made a lot of explorations in large-scale applications this year, and also involved the field of BI training and sales management. So I would like to ask, how much does the company currently see the client's fee for large model applications? This is the first question. Then the second question, I would like to ask, are these applications still limited to some work scenarios with a relatively high melting point? In the future, as the technology improves, especially the appearance of agents, can it improve the illusion of large model to a certain extent, so that these applications can achieve more replacement for artificial intelligence? That's all. Thank you.

Yes, that's right. This question itself is very explanatory. I think basically all the points have been mentioned. First of all, I think the cost-benefit of the enterprise client for the large-model application is completely dependent on how much input and output it can bring to the application. And there is one thing that needs to be noted, that is, the input and output ratio must be relatively high. Because in essence, it means that this enterprise has to pay a lot of attention There are also re-definitions of some positions. This is for a company. If there is no such value, they will not be willing to promote it. At present, I think this is the popularity of corporate customers' use of large models. How do I put it? It's becoming more and more rational. I know that last year, many companies spent a lot of money for the authorization of a large model or the authorization of a four-dimensional model. This year, it's obvious that people will no longer pay for these or very little. And more so, what can you provide me with? Something that can be used directly. And then you asked about this. The current large model is indeed in a specific scene with a relatively high penetration rate. For example, just training. I think it's still in the entire industry. agent has been able to effectively improve and improve the problem of the large model. Because agent itself is moving some codes or task planning to frame the ability of the large model in a particularly vertical scene. At this time, when the scene of the large model is particularly vertical, its error rate will be reduced. Especially this wave, the current ability of the large model has reached this level. This is when everyone is invested in aging. So for this big model, this kind of application on the B-end, it will gradually realize more and more replacement. Here, I think the last question, I will also give an example. That is, today, if we do a big model application, if it is a fan application, it is theoretically difficult to make the customer satisfied enough. This is what we have been through for so long. I will give an example. For example, you bought Apple's latest 我也专门买了一个能用海外版的那个去尝试 你就发现它能够真正体现出来就很少 这个体验就没有那么强吧 然后包括微软推的Copilot 我认为这两类应用都是把它的 因为他们大公司把它定义的都过于宽泛 这一点上我觉得 但他们反正有时间 有足够的商业闭环 他就慢慢的在这个方向走 Operator, please move to the next question, please. Thank you.

No problem. And our next question comes from Ben Cheng Wu with TF Global. Please go ahead.

Thank you for giving me the opportunity to ask a question. I would also like to ask Vice President. Recently, everyone has been discussing a lot about the problem of scaling and slowing down. I would like to ask some of your opinions. And in terms of scaling and slowing down, what impact does it have on the entire industry development of large-scale applications?

Scaling and slowing down. We still have some discussions, but I don't know if ScandiLaw will be put on hold. But at least recently, especially in the past two months, everyone is talking about the lack of data, right? Because basically, the good data that can be used for big model training on the Internet is basic data. It's not necessarily the same in the industry, but I heard about 20T to 30T tokens. It's basically about this much data. Because some data, although there is, but its data is not high enough, it will make the model more difficult to use, right? Including now, our whole industry is focused on OpenAI. This GPT-5 has not been released yet, right? You see, these 12 days are basically the evidence of the productization of the past model capabilities. What does that mean? It means that in some sense, today's top model, the top model in the world's scope, is not easy to grow in a period of time. At least 4E has been around for many months. Before that, we looked at 3.5 to 4.0 to 4E, and each step was quite fast. But now it's been a long time. So we think that At least today, it seems that the growth of the top-end model capabilities in this industry is slowing down. I don't know if Scalino is slowing down, but this is definitely slowing down. But this is a good thing for the startup company itself, especially for companies like us who do applications. Because when the model capabilities were developing at a high speed, You did a lot of things. It's really what Sam Altman said at the time. Don't do it. I made my new model and it was covered for you. There were indeed some such projects at the time. After he finished, the model came out and added this ability to cover it for you, right? But today, due to the increase and decrease of the top model ability, everyone is thinking about how to use agent to use these abilities better. This is a big idea for application. So, So we think this is good for us. We can also be at ease, right? We don't participate in this model roll competition. We just do this agent itself. And then we today, due to this entry earlier, we still spent a lot of energy to do some scenes. So these scenes for us is the best way to polish the agent. Because the essence of the agent is to be combined with the customer and the market. You only know how satisfied this thing is. Thank you. Operator, please move to the next question. Thank you. Our next question comes from Alex Wong with SDICS International. Please go ahead.

Thank you, Mr. Wang, for accepting my question. I would like to ask a question about robots. I would like to ask you, Mr. Fu, what do you think of some changes in the trend of the robot industry in the past few months? And then, in addition, what is the role of the big model in the landing of robots? And how do you see the competition pattern of the robot industry in the future? Thank you.

For the past few months, I feel that the concept of a big model in the robot industry is getting more and more popular. And then because of this, Tesla's, this, today's addition of this humanoid robot is very popular. to build more sensors, very expensive laser radars, and very perfect algorithms to achieve this very strong individual, and then to achieve automatic acceleration landing. And the other route is the route of Tesla and at that time, I didn't think I would be able to do it. So I used visual sensors first, continuous sensors, and then I found that visual can be used, and then use more machines to run data from all over the place for end-to-end training. Today, we can basically say that the second one is that with the deepening of the scene and the continuous searching of data, the implementation of this technology It's much better than the original high-quality engineers, the best sensors, and this route. At present, it must be much better to go here. So I think the same goes for the robot. It won't be the same as the human being. It won't be able to do anything. In fact, the human being can't do anything either. In the end, he still has to drive a car, or push a small car, or use some tools to achieve more landing. And once this error occurs, it will cause the machine to fall. It's very difficult, right? Let me give you an example. For example, we are in the restaurant. In some of the scenes, you will find that you have to be under the supervision of no one. Run for more than ten hours a day, run hundreds of times, and then in a year like this, you can't make mistakes in thousands of scenes. So a little bit wrong for your customer's confidence and whether the customer can make it work is the biggest problem. So I think the big model will definitely help the robot in a big way, but when it really lands, it should be combined with the scene and a little bit of advance. So that's why I don't really look forward to the core of the humanoid robot. I think the future pattern of the robot is a very big problem. I believe that in the end, it will be like the example I gave earlier, where there will be more and more real-time early-stage manufacturers appearing. In one scene, the final competitors will emerge. I don't think it's a good idea to propose a big concept today, and then take action in the direction of the landing, so that it can emerge. Because there is another basic logic, which is This robot is an industry that combines hardware and software. And the hardware system inside is very complex. It's not like a car. It's more like a wheel structure or something. It has a lot of mechanical structure in it. And the improvement of the mechanical structure and hardware system is not based on Maltese law. You can have 18 months of software. What is the cost of education? It's a good deal. The hardware has to be a little bit more. It's been a hundred years since the car came out today. It's been more than a hundred years, right? There is a change in this smart car. So I think the future should still be... Anyway, we are firm on our own path, which is to continue to combine with the scene, and then add some mechanical force, and add some mechanical force to some scenes that just need pain, to complete some high-quality, high-reliability actions to achieve greater expansion. Thank you. Okay, thank you. No more questions. Okay.

Thank you. Please move to the next question, please.

Thank you. Our next question today comes from Yongping Biao with Gotai Junon Securities. Please go ahead.

Hi, management. Thanks for taking my question. And we have noticed that you have a large amount of net cash on your balance sheet. So do you have any plan of dividend distribution or share repurchase in the future? Thank you. Thank you for your question. My question is that the company has a lot of cash on its account. Will there be any repurchase or promotion plans in the future?

Thank you. Let me answer this question. I'm Thomas, CFO of the company. Thank you for your question. First of all, Leopold has always had an open attitude towards shareholder returns. The management team also pays great attention to shareholder returns. In history, we can see that we have sent shares twice, and we have also made many repurchasing plans. But sending shares twice is also based on the fact that we have an important investment project, we got a refund, and then returned it to our shareholders. But whether it will become popular in the future, or other means, we actually have a lot of factors to consider. For example, Vice President also mentioned that we actually In terms of technology, there is still some investment to be made, including our business, which is also transforming from 2C to 2B. Developing AI models and mechanical business also requires some investment. At the same time, we also feel that the overall economic environment is not very certain at the moment. So for us, it is more important to maintain a sufficient cash reserve and develop the business. So in the future, we will continue to maintain a more cautious financial strategy to ensure that the company has enough flexibility and risk-resistant ability in the face of market fluctuations. If in the future, our board of directors will approve the plan of repurchase and distribution, we will make an announcement to the market at the first time. Thank you. Okay, thank you. Very clear.

I have no other questions.

Thank you. Operator, please move to the next question.

Thank you. Our next question comes from Gigi Zhao with GF Securities. Please go ahead.

Thank you for giving me the opportunity to ask a question. I would like to ask how you see the development and competition of the domestic model. For example, in terms of actual operations, how do you see the development and competition From the perspective of performance and efficiency, is there a significant difference? Or is the competition of large models already surpassing the ability of the model itself? In the future, it may be more like productization and ecological construction. Yes, that's my question. Thank you. Okay, thank you for your question.

You also know that the domestic ecology is more complicated. I can't evaluate which model product is better. Everyone uses it by themselves. There will be a sense of body. I often use it, but I have to say that there is still a difference in this experience. I think you asked this question very well today. I think the competition of big models has actually surpassed the ability of models from the very beginning. Because the ability of models is the most important thing. We recently pushed a function called AI Data Tower. You will find that today the ability of the model itself depends on the data. And today's data, because everyone is more public on the Internet, you can do more engineering in the high-quality image management of this data in the markup and these aspects. Your model ability will be good. You think we really trained two models. One is 1.4 billion parameters, one is 7.8 billion, 7.1 billion MOE. In fact, the result of the list is also good. Because we are here, so today's big model competition is more about the importance of this matter and the data input. And I think at this stage, it is true that today's competition must go towards productization and ecological construction. Who can really go through the improvement of product experience to complete the satisfaction of users, who will have the opportunity to win. I don't think it will affect the gap in the underlying model. If you look at the list, today's this, tomorrow's that, and the next day's that. I think this thing might have already passed. Productivization and eco-development. I saw a post by Sam Altman not long ago. He asked what OpenAI lacks the most. He said it lacks products. So the progress of this AI I think its speed is very fast. So now it has gone from the original technical frenzy or technical criticism to product-based criticism and ecological criticism. This is a very obvious stage. I see some domestic experts or entrepreneurs recently mentioning that next year will be the year of the application outbreak. I agree with this principle because the ability of models has reached a relatively high level. Everyone is not much different. And it's not easy to get higher. I just talked about the data issue. Unless there's a new special way. Now everyone's efforts will go towards products and ecology. And since we've reached a certain level, everyone will invest in it. That will make the user experience more important. So next year, I think in terms of products and ecology, there will be a lot of progress. Thank you.

Thank you. Operator, please move to the next question.

Thank you. Our next question comes from Joanna Ma with CNBI. Please go ahead. Thank you for your question.

I would like to ask you about the basic problems of large model experiments. How is the technical ability of Chinese enterprises in the field of large model experiments? Because we also know that the new style characteristics of fortified learning lack the open-source model and academic paper as a direct reference. Thank you. This question is really academic. First of all, I think in the style of fortified learning, the technical ability of Chinese companies is not bad.

Fortified learning has not been around for a long time, right? I think this wave is mainly because of the launch of OpenAI's O1. Everyone found that in the language model, we need to do enhanced learning. You can look at the past. In the past, after AlphaGo was launched, some big companies and teams in China launched it. In fact, the level is not bad. Basically, as I just summed up, it may be a few points or two points worse in some specific assessment scales. I don't think this will affect the final productization. It includes our voice recognition and previous vision recognition. It can catch up with foreign companies at many points. So in terms of technical capabilities, China has accumulated a lot of talent in the field of Internet, big data, and software engineering over the past long period of time. I think this is enough to guarantee it. There is a lack of existing open source models. I don't think so. Thank you. Operator, please move to the next question, please. Thank you. Our next question comes from Chengru Li with WoYan Securities. Please go ahead.

Thank you, Mr. Guan. This is my question. I would like to ask, we can see more and more businessmen using service-type robots in China, and the most places we see are restaurants and hotels. I would like to ask, how is the market situation in these two scenarios? And if we look at it from a big perspective, how is the penetration rate of robots in the restaurant and hotel markets? In the future, Thank you. Okay, the ranking of market share, because this industry is relatively early, some reports are relatively rough.

Of course, we have to say that we did the restaurant and hotel relatively late, because we did the reception service first. I think we should be in the first few places. China China China China China I think the market space in this area is huge. You can imagine this kind of foreign market as the market three to five years ago. Today, some domestic markets are not entirely dependent on the machine market. For example, we all know that before the restaurant, there was a decline in sales. So their purchasing power is declining. This is also a fact. But in the long term, I'm very optimistic about these two scenarios, including some scenarios that will be extended. Our robot is not just a base model. Now we are using a large-scale model to enhance the ability of the voice. At a certain level, we believe it will greatly expand the scope of work in restaurants and hotels. In the next three years, in terms of the scope of these two sub-markets, Our goal, of course, is to achieve the maximum, right? If we can't achieve the maximum in three years, then this market means we haven't done well. I don't have this confidence, but can we achieve it? It depends on our team's execution. Of course, this time, this year or so, my biggest experience is like this. Because it's a 2B market. So, in the construction of sales channels, sales management, and sales team building, of course, it takes a lot of energy. This is also why I often visit many customers both overseas and domestically. Because I think our technical ability is definitely ahead in this industry. We can do a good job in terms of product quality, customer satisfaction, and customer satisfaction. So the real difficulty here is not technology and products for us, but more of the construction of sales channels and your entire sales network. I talked to a friend yesterday about this wave of Chinese companies going out of the sea. I think it's the biggest difference from the last wave of leopard movement going out of the sea. You have to go deep into the local area, do some localization of the channel construction, as well as the understanding of the local market, and be able to develop the corresponding method. If this can be done, I believe our Huchenghe will be much deeper than the previous wave, where we used the Internet to put ETPs and bind them. However, the experience and difficulty required will be relatively high. This is our view. But I am confident that I will win the first place in the market in three years. Thank you. We have other questions.

Thank you. As operator, if we have no further questions, we can just conclude the Q&A section.

We have no further questions, ma'am.

Okay. So thank you so much for joining Trita Mobile's third quarter 2024 earnings conference call. If you have any further questions, please do not hesitate to reach out to our IR team. Thank you so much. Bye-bye.

Thank you. This concludes today's conference call. We thank you all for attending today's presentation. You may now disconnect your lines and have a wonderful day.