Mark Zuckerberg and Susan Li Interview: Meta's AI Vision and the Growth of Social Apps

Mark Zuckerberg: It was a good quarter, with strong product and business momentum, and some of our long-term visions for the future of artificial intelligence and computing becoming clearer. We estimate that more than 3.20 billion people now use at least one of our applications every day, and we see rapid adoption of Meta AI and Llama, which are rapidly becoming industry standards.

So, let's start with the highlights of these apps. For WhatsApp, the US is still one of our fastest growing countries, and we have just surpassed the milestone of making 2 billion calls globally per day. On Facebook, we continue to see positive trends among young people, especially in the US. Instagram's global growth momentum remains strong, and this quarter we also launched teen accounts, which have built-in protection features that can restrict who teenagers send messages to and what content they can see. The Threads community currently has nearly 275 million monthly active users, with daily registration growth exceeding 1 million, and active levels are also growing. We will continue to move towards our goal of making it the next major social app.

We have also made great progress in our efforts in artificial intelligence. We see that artificial intelligence has had a positive impact on almost all aspects of our work, from core business engagement and monetization to long-term roadmaps for new services and computing platforms. I think part of this comes from having a vision and roadmap that aligns with the direction of technological development, but more importantly, our team has done an excellent job in many aspects. Meta AI now has over 500 million active users per month. This year alone, our improvements to AI-driven information flow and video recommendations have increased user usage time on Facebook by 8% and Instagram by 6%. Last month, more than 1 million advertisers used our gen AI tool to create more than 15 million ads. We estimate that businesses using image generation will see a 7% increase in conversion rates, and we believe there is still a lot of room for growth in this area.

We are also seeing strong momentum in Llama, and this year, the usage of Llama tokens has grown exponentially. The more widely Llama is adopted and becomes an industry standard, the more its quality and efficiency improvements can be fed back into all of our products. This quarter, we released Llama 3.2, which includes leading small models running on devices and open source MultiModal Machine Learning models. We are working with enterprises to make it easier to use. Now, we are also working with the public sector to make Llama widely adopted in the US government. Llama 3 can be said to be a turning point in the industry. But I am more concerned about Llama 4, which is currently in development. We are training Llama 4 models on a cluster larger than 100,000 H100s, larger than anyone else I have seen doing. Smaller Llama 4 models will be ready first, and we expect them to be released sometime early next year. I think they will play an important role in multiple ways, including new modes, new capabilities, stronger inference, and faster speed. In my opinion, open source will be the most cost-effective, customizable, reliable, highest-performing, and easiest-to-use option available to developers. I am proud that Llama is leading in this regard.

Meta formulates a budget for next year every year, which is still being developed. But I want to share some things that have impressed me so far as we go through this process. Firstly, it is obvious that there are many new opportunities to accelerate our core business using new artificial intelligence advances, which should have a high return on investment in the next few years. So, I think we should invest more in this area. Secondly, our artificial intelligence investment still requires a lot of infrastructure, and I expect to continue to invest heavily in this area. We have not yet determined the final budget, but these are some directional trends that I have seen.

This quarter, we have made several milestones in Reality Labs and integrating artificial intelligence into wearable devices. Ray-Ban Meta glasses are a typical example. They have a beautiful appearance and can allow you to take photos and videos, listen to music, and answer phone calls. But what really sets them apart is Meta AI integration. With our new update, it can not only answer your questions all day, but also help you remember things, provide real-time multimodal AI suggestions when you do things, and even translate other languages directly in your ear. I still believe that glasses are the ideal form of artificial intelligence because you can let your artificial intelligence see what you see, hear what you hear, and talk to you. The demand for glasses is still very strong. The brand new transparent version we released on Connect sold out almost immediately, with an online transaction price of over $1,000. We have deepened our cooperation with EssilorLuxottica to create future smart glasses that combine cutting-edge technology and fashion style.

At the Connect conference, we also showcased our first holographic AR glasses, Orion. We have been developing this glasses for about ten years. We are not far from being able to provide beautiful-looking glasses that seamlessly integrate the physical and digital worlds, so that no matter where they are, you can feel the feeling of being with others. We are beginning to see the emergence of the next generation of computing platforms, which is very exciting.

We also released the latest Mixed Reality earbuds, Quest 3S. It brings the best features of Quest 3, high-quality color pass-through, new chipset, and more, at a cheaper price of only $300. So far, the reviews have been good, and I'm looking forward to seeing how it performs this holiday season as more and more people buy it.

Overall, this is a good quarter. I am very excited about all the work we are doing so far. This may be the most dynamic moment I have ever seen in our industry, and I am focused on ensuring that we build something great and fully utilize future opportunities. If we can do this well, Meta and everyone who builds with us will have great potential.

Susan Li: Good afternoon. Let's start with consolidated results. All comparisons are year-over-year unless otherwise noted.

Total revenue for the third quarter was $40.60 billion, an increase of 19% or 20% in constant currency. Total expenses for the third quarter were $23.20 billion, an increase of 14% from last year. In terms of specific projects, revenue costs increased by 19%, mainly due to increased infrastructure costs.

R & D expenses increased by 21%, mainly due to the increase in employee headcount-related expenses and infrastructure costs. Sales expenses decreased by 2%, mainly due to the decrease in restructuring costs. Administrative expenses decreased by 10%, mainly due to the decrease in legal-related expenses. As of the third quarter, we had more than 72,400 employees, a year-on-year increase of 9%, mainly due to our recruitment in priority areas such as monetization, infrastructure, reality labs, generative AI, and regulatory and compliance.

Third quarter operating profit was $17.40 billion with an operating margin of 43%. Our tax rate for the quarter was 12%. Net Profit was $15.70 billion, or $6.03 per share. Capital expenditures, including principal payments for financing leases, were $9.20 billion, primarily from investments in Servers, Data centers and network infrastructure. Our capital expenditures were impacted in part by the timing of third quarter server deliveries, which will be paid in the fourth quarter. Free cash flow was $15.50 billion.

During the third quarter, we completed $10.50 billion of debt to publish, repurchased $8.90 billion of Class A Common Stock, and paid a $1.30 billion dividend to our shareholders. At the end of the quarter, we held $70.90 billion of cash and marketable securities, and $28.80 billion of debt.

Now let's talk about our segment performance. I'll start with our App Series segment. Our App Series community continues to grow, with more than 3.20 billion people using at least one of our App Series every day in September. Total App Series revenue for the third quarter was $40.30 billion, up 19% year-over-year. App Series advertising revenue for the third quarter was $39.90 billion, up 19%, or 20% in constant currency. In advertising revenue, e-commerce was the largest contributor to year-over-year growth, followed by healthcare and entertainment and media. From the perspective of user regions, the strongest growth in advertising revenue was in the rest of the world and Europe, with growth rates of 23% and 21%, respectively. The Asia-Pacific region grew by 18%, and North America grew by 16%. From the perspective of advertiser regions, total revenue growth in North America and Europe was the strongest, reaching 21%, while the rest of the world grew by 17%. The Asia-Pacific region was the slowest growing region, with only 15%, lower than the 28% growth rate in the second quarter, mainly due to the strong demand from Chinese advertisers in the same period last year.

In the third quarter, the total number of ad impressions placed in our services increased by 7%, and the average price per ad increased by 11%. Impressions growth was primarily driven by the Asia Pacific region and the rest of the world. Pricing growth was driven by increased demand from advertisers, in part due to improved ad effectiveness. This was partially offset by impressions growth, particularly from regions and services with lower monetization.

App Series Other revenue was $434 million, up 48%, primarily driven by growth in WhatsApp Business platform business messaging revenue. We continue to invest the majority of our investment in the development and operation of the App Series.

In the third quarter, application series expenses were $18.50 billion, accounting for approximately 80% of our overall expenses. Application series expenses increased by 13%, mainly due to the increase in infrastructure and employee-related expenses, but the decrease in legal-related expenses partially offset this growth. Application series operating profit was $21.80 billion, and the operating profit margin was 54%.

In our Reality Labs division, third quarter revenue was $270 million, up 29%, primarily due to hardware sales. Reality Labs expenses were $4.70 billion, up 19% year over year, primarily due to increased employee-related expenses and infrastructure costs. Reality Labs had an operating loss of $4.40 billion. Now let's talk about the business outlook.

The main factors driving our revenue performance are two: our ability to provide engaging experiences to the community, and our effectiveness in monetizing this engagement over time. Firstly, we focus on improving the experience of current users in our app and invest in long-term plans that may promote engagement in the coming years. We expect our content recommendation roadmap to cover both time frames, as we have a multi-year plan to focus on improving recommendations and developing innovative new methods.

In the third quarter, we continued to see year-over-year growth in daily usage on Facebook and Instagram, both globally and in the US. On Facebook, we saw strong results from the global rollout of the Unified Mobile Pentium 4 in June. Since the launch of the new experience and prediction system that supports the player, we have seen a 10% increase in time spent on Facebook Mobile Pentium 4. This month, we move into the next phase of our Facebook video product development. Starting in the US and Canada, we will be updating the standalone video tab to a full-screen viewing experience, which will enable people to seamlessly watch videos for a more immersive experience. We expect to complete the global update in early 2025. On Instagram, Reels continues to receive widespread attention, and we have made continuous progress in promoting original content. Currently, over 60% of recommendations come from original posts in the US. This not only helps people find unique and differentiated content on Instagram, but also helps early creators to be discovered.

Next, let me further introduce our years of recommendation paths. Previously, we operated separate ranking and Recommender systems for each Product because we found that if we expanded the model size and computing power beyond a certain point, the performance would not expand. However, inspired by the expansion patterns we observed in large models, last year, we developed a new ranking model architecture that can learn more effectively from much larger datasets. First, we have deployed these new architectures to our Facebook ranking-video ranking model, which allows us to provide more relevant recommendations and significantly improve viewing time. Now, we are exploring whether these new models can bring similar improvements to recommendations for other applications. Afterwards, we will consider introducing cross-application data into these models, so that our system can understand what users are interested in from one application and use it to improve recommendations on another application. This will take time to achieve, and we will also explore other things at the same time. However, over time, we are optimistic that this will unlock more relevant recommendations, and because we have fewer recommendations running, engineering efficiency will also increase.

In addition to recommendations, we have also made progress in other long-term engagement priorities, including generative artificial intelligence and threads. As we offer Meta AI in more countries and languages, its usage continues to expand. As we improve the model and introduce many enhancements in recent months to make Meta AI more helpful and attractive, we have seen an increase in usage.

Last month, we started introducing voice functionality so you can talk to Meta AI more naturally. Now users in the US, Australia, Canada, and New Zealand can use English. In the US, people can also upload photos to Meta AI to learn more, write titles for posts, and add, delete, or change image content with simple text prompts. All of these are built using our first MultiModal Machine Learning base model, Llama 3.2.

Threads is still another area where we see exciting potential. We are attracting more and more new users every quarter, while customer engagement is constantly improving. Looking ahead, we plan to launch more features to make it easier for people to understand the topics they care about.

Now, let's talk about the second driving factor of revenue performance, which is to improve monetization efficiency. This work is divided into two parts. The first is to optimize the advertising level in natural interactions. We continue to see opportunities to increase advertising supply in areas with low profitability such as video. Within Facebook, with the unification of Mobile Pentium 4, video interactions continue to shift to short form, and we expect this situation to continue as the video tab transitions to full-screen format. This has led to a faster growth rate of natural video display than overall video time on Facebook, providing more opportunities for advertising. On Facebook and Instagram, we are also continuing to carry out broader work to optimize the time and location of displaying ads in user sessions. This allows us to drive revenue and conversion rate growth without increasing the number of ads.

The second part of improving monetization efficiency is improving marketing effectiveness. Similar to organic content ranking, we are looking for opportunities to achieve meaningful advertising effectiveness improvements by adopting new modeling methods. For example, we recently deployed new learning and modeling techniques that enable our advertising system to consider a series of actions taken by users before and after seeing ads. Previously, our advertising system could only aggregate these actions together, not map the order of these actions. This new approach enables our system to better predict audience reactions to specific ads. Since we adopted the new model in the first half of this year, we have seen conversion rates increase by 2% to 4% based on tests conducted in selected segmented markets.

We are also improving our ad platform to ensure that the results we drive can be customized based on each business's goals and how they measure value. In the third quarter, we made changes to our ad ranking and optimization models to consider the process of cross-platform delivery more, which we expect will increase the meta-attribution conversions that advertisers see in their third-party analytics tools. We are also testing new features and settings for advertisers to optimize their ad campaigns based on the aspects they value most, such as driving incremental conversions rather than absolute conversions. Finally, our Advantage + solution (including our ad creative tools) continues to maintain strong momentum. We found that advertisers who use our AI-based generative image extension, background generation, and text generation tools have high Retention Rates, and these tools have already brought performance improvements to advertisers in the early stages. Earlier this month, we began testing our first video generation feature, video extension, and image animation. We expect these features to be more widely launched by early next year.

Next, I would like to discuss our capital allocation situation. We continue to operate our business with a long-term perspective, which involves investing in a range of opportunities that we expect to generate returns at different times. We are very optimistic about the current range of opportunities and believe that investing in infrastructure and talent now will not only accelerate our progress, but also increase the possibility of maximizing returns in each area. This includes investing in short-term plans to achieve sustained healthy revenue growth in our core business, as well as investing in long-term opportunities that have scale and can bring considerable returns over time. Considering the preparation time for our long-term investments, we will also continue to maximize our flexibility so that we can respond to market developments. Within Reality Labs, this has benefited us greatly as we have improved our path to respond to the earlier-than-expected success of smart glasses. In the field of generative artificial intelligence, we expect to significantly expand infrastructure capacity while prioritizing interchangeability, which will also put us in a favorable position to cope with the development of technology and the market in the coming years.

Now let's talk about our financial outlook. We expect total revenue for the fourth quarter of 2024 to be in the range of $45 billion to $48 billion. Our guidance assumes a roughly neutral impact of foreign currencies on year-over-year revenue growth based on current exchange rates. Now let's talk about the outlook for expenses. We expect total expenses for the full year of 2024 to be in the range of $96 billion to $98 billion, an update from the previous range of $96 billion to $99 billion. For Reality Labs, we continue to expect operating losses to increase significantly year-over-year in 2024 as a result of our ongoing product development efforts and investments to further expand our ecosystem.

Speaking of taxation, if there are no changes in our tax environment, we expect the tax rate for the fourth quarter of 2024 to be below 10%. In addition, we will continue to monitor active regulatory environments, including increasing legal and regulatory resistance in the European Union and the US, which may have a significant impact on our business and financial performance.

Finally, this is another good quarter for our business. Our global community continues to grow. We see our core business continuing to grow, and we have exciting opportunities in front of us to further grow our core business in 2025 and take advantage of future long-term opportunities.

Susan Li: Your first question is what kind of repeated interactions we see between people and their use of Meta AI. We see - first of all, I think, as Mark mentioned, we are excited about the progress of Meta AI. Obviously, it is still in its infancy, but it is expected to become the most widely used artificial intelligence assistant in the world by the end of this year, with monthly active users exceeding 500 million. People use it for many things. Some common use cases we see include information collection, helping to complete operational tasks, which is the largest use case. But we also see people using it to deepen their understanding of interests, find content in our services, and generate images . So far, this is also another very popular use case. I would say that in the short term, our focus is on making Meta AI more and more valuable to people. If we succeed, we think people will use it for a wider range of searches, including more monetizable queries over time.

Regarding the second part of your question, Meta AI extracts content from the internet to promptly solve user problems and provides sources from our search engine partners for these results. We have integrated with Bing and Google, both of which provide excellent search experiences. Like other companies, we also use publicly available content on the internet to train our artificial intelligence models and crawl the web for various purposes.

I think your second question is actually about how we are thinking about headcount in 2025. We are still working on the budget process for 2025. This is one of the reasons why we changed our forward guidance approach to provide guidance on the next conference call. But as we approach this issue, we are looking for opportunities to invest in our strategic priorities. This includes monetization, infrastructure, reality labs, artificial intelligence, and our continued investment in regulation and compliance. We are indeed evaluating each opportunity, looking at measurable return on investment or strategic opportunities, depending on the field. We support this by continuing to truly focus on streamlining other businesses. Therefore, we do not have specific information on the increase in the number of employees in 2025, but this can give you a glimpse of our progress in the budget process.

Mark Zuckerberg: I think the main point here is that it seems to be broadly applicable to a wide variety of products. Therefore, some areas are more part of our core business, from making Feeds and Reels more relevant, to making ads more relevant, to helping advertisers make better ads, to helping people create the content they want, to helping our integrity operations and compliance and the work we do there. This is important. It is very valuable in all these aspects of the core business. But it will also enable new types of services, such as things like Meta AI that we did not have before. We haven't had anything like Ray-Ban Meta glasses before, and artificial intelligence will be a very important part of all these products. There are other similar new products around AI Studio. This year, we are really focused on launching Meta AI as our single assistant that people can ask any question to.

But I think that in terms of consumer and commercial use cases, we will see more opportunities next year. People can interact with various AI agents, and consumers can use artificial intelligence studios, whether it is different creators or different agents created for entertainment. Or in terms of business, we do hope to continue to make progress on this vision, allowing any small business or any enterprise to establish an AI agent with just a few clicks of the mouse over time, helping to provide Client Server and sell products to all customers around the world. I think this is a huge opportunity. So, it is very broad. I think the part we see is that there are many opportunities.

Some long-term investments involve Meta AI or AI Studio. These may not necessarily be huge profit opportunities in the next few years. But there are many things in the core business around engagement and monetization, which I think will happen in the next few years. So, I think we are working hard to ensure that we find the right people to do this and ensure that we have appropriate investments that will be used for what we believe are very, very big opportunities.

Mark Zuckerberg: I can answer the question about Meta AI, although I intentionally don't talk too much about the new features and modes of Meta AI that we will launch through Llama 4. What I mean is that I pointed out in the previous comments that with each major intergenerational update, I expect a lot of new capabilities to expand. But I think this is only part of the reason why I am excited, and we will talk more about this issue next year when we are ready.

However, I think one trend we will see is that these models not only provide support for our single assistant Meta AI, but also for AI Studio and Business Agent, making it constantly evolving. What I mean is that this year, if you look back at the situation about a year ago, we started launching Meta AI. This year, we did successfully develop it and let many people use it. Obviously, we hope to add more engagement depth and new use cases over time.

But what I want to say is that the achievements we have made in AI Studio and commercial AI today are roughly the same as what we achieved in Meta AI a year ago. Therefore, I think our goal for next year is to work hard to make these use cases widely applicable, although it will take many years to achieve the deep usage and business results we want. Therefore, there are still many things to be done here, and I am happy to start talking about these things from the beginning of next year.

Susan Li: Your second question is about capital expenditures in the fourth quarter. Capital expenditures in the fourth quarter are expected to increase compared to the third quarter, partly due to increased server expenditures and (to a lesser extent) increased data center capital expenditures. However, there is a timing issue for servers, as our server deliveries were completed at the end of the third quarter. Therefore, cash will not flow out until the fourth quarter, when you will see capital expenditures. Given the nature of capital expenditures, there will generally be some fluctuations between quarters. So, it is difficult to extrapolate from any specific quarter. Overall, I think we will significantly increase infrastructure investment this year, and we expect to achieve significant growth again in 2025.

Susan Li: Regarding the use of artificial intelligence and employee productivity, this is certainly something we are very excited about. I don't know if we have anything particularly quantifiable to share now. I think there are different efficiency opportunities for artificial intelligence, and we have been focusing on these opportunities, that is, we can reduce costs over time and save expenses by improving internal productivity in areas such as coding.

For example, although it is still in the early stages, we have seen a lot of internal adoption of our internal assistants and coding agents, and we continue to make Llama more effective in coding, which should make this use case more valuable to developers over time. We also hope to deploy these tools into many of our content moderation work over time to help us carry out a lot of content moderation work and improve work efficiency and effectiveness. In many other parts of the company, I would say that we are still in a relatively early stage of exploring how to use LLM-based tools to improve the efficiency of different types of workflows. So, all of this makes us very excited.

Regarding your second question about the number of employees, our budget is still at a moderate level. Therefore, we don't have much clear information to share at the moment. However, when we evaluate where there are good investment opportunities, we do believe that there are many opportunities for personnel growth driven by investment returns. We strictly consider the way of return, what the return opportunities are, how we view the possibility of these returns, and how much the overall increment of these investments is. These are all things we evaluate when considering where to invest in our core business and where we believe we can achieve investment returns in the short term. At the same time, we are also evaluating opportunities in some medium and long-term strategic investment areas.

This includes our efforts in artificial intelligence and the infrastructure needed to support it. This includes our investment in Reality Labs. Therefore, these are all things we are evaluating to assess what we can do in 2025 - what we think we will do, and some ideas. First, where can we establish the greatest flexibility to consider infrastructure or employee plans? Second, we are truly focused on efficiency work throughout the company and ensuring that we feel we are continuing to drive the entire company, including areas where we expect to invest additional employees to think about how to be more efficient in 2025 than in 2024.

Mark Zuckerberg: Yes. I can try to explain this in more detail. I mean, I think there are several forms of improvement for Llama. One is quality improvement, and the other is efficiency improvement.

There are many researchers and independent developers working on Llama because it is available. They work on Llama, make improvements, and then release it, making it easy for us to re-integrate it into Llama and our Meta products, such as Meta AI or AI Studio or commercial AI, because the examples shown in the work are what people do on our stack. Perhaps more importantly, efficiency and cost. What I mean is that these things are obviously very expensive. If someone comes up with a better way to run it, if they can run it 20% more efficiently, it will save us a lot of money.

This is our experience with Open Compute, and one of the reasons why we rely so much on open source, because we found that the release and sharing of our computing architecture and design with Open Compute has made the industry more standardized, which is contrary to our intuition. We also received some suggestions that helped us save costs, which is very valuable to us. One of the biggest costs here is the chip. In a lot of infrastructure there, what we see is that as Llama is increasingly adopted, companies like NVIDIA and AMD have optimized their chips more to run Llama better, which is obviously beneficial to us.

Therefore, it benefits everyone who uses Llama, and it makes our product better, right? Instead of just building a model on an island that no one has standardized in the industry. So, this is some of the situations we see around Llama, and I think doing this in an open way is beneficial to us.

In terms of expanding infrastructure, what I mean is that when I talk about our team executing well, some of it is used to provide more attractive products, and some is used to bring in more revenue. In terms of infrastructure, it helps to increase expenses faster, so I think part of what we have seen this year is that the execution of the infrastructure team is quite good. I think this is why we have been able to expand production capacity in this year. I mean, going into this year, we have a range of things that may be done. I think we have been able to do more than we hoped and expected at the beginning of the year. Although this reflects higher expenses, I am actually pleased that the team has been able to execute well.

I think this execution makes me more optimistic that we will be able to continue building this project at a good pace, but this is part of the whole thing, that is, the formula around infrastructure building may not be what investors want to hear in the short term, and we are developing this project. But I think the opportunity here is really great. We will continue to invest a lot of money in this area. I am proud of the teams that work hard and build a lot of capacity so that we can deliver world-class products.

Susan Li: Your first question is about Threads. We have made great progress in this area. We will continue to launch more features and improve our ranking stack.

We are very pleased with the continued growth of Threads' users. Our user numbers are increasing every quarter, and engagement is also constantly improving. In the third quarter, we saw particularly strong user growth in major markets such as the US, Taiwan, and Japan. We added many new features in the third quarter, including providing account insights for businesses and creators to understand their post performance, as well as the ability to save multiple drafts, continuing our commitment to integrating Threads with the metaverse.

Basically, we are very focused on continuing to build the functionality of Threads over time and responding to what users tell us they are interested in. Specifically, in terms of monetization, we do not currently believe that Threads will be an important driver of revenue in 2025. We are satisfied with the growth trajectory and are once again truly focused on introducing valuable features that the community deems valuable and striving to deepen growth and engagement.

Your second question is about the increase in the average price per ad. This number increased by 11% year-on-year, thanks in part to strong demand from advertisers and the continuous improvement of ad effectiveness over time. We found that the CPM growth rate was slightly faster than the 10% in the second quarter, partly due to the lower growth rate of impressions in the third quarter. However, from a broader perspective, when we consider pricing growth and this indicator, that is, the year-on-year growth of the reported price per ad, there are many factors, including auction dynamics caused by fluctuations in the growth of impressions.

We feel that one of the things we are very concerned about is the input metrics. What are the conversions we bring to advertisers? Have they gained more value over time? The price per ad in this comprehensive report is very complex because all these factors will be combined. There are many goals that advertisers optimize for. The value of these goals varies greatly, making it difficult to compare them to the same category. But we are very concerned about conversion growth, which is growing and continues to grow at a faster rate than display growth. Are we seeing a healthy trend in cost per action or cost per conversion? The answer is yes. As long as we continue to drive better conversions for advertisers, over time, this should have the effect of increasing the cost per thousand impressions, as we provide more conversions per impression, which will lead to higher value impressions.

Mark Zuckerberg: Yes. We are developing a lot of consumer products. With Llama, I hope that application developers can also develop a lot of really good things. I mentioned Meta AI, AI Studio, and business AI many times, and I hope they will become an important part of the consumer experience.

Another part that I haven't talked much about yet is the opportunity for artificial intelligence to help people create content and improve their dynamic experience. But if you look at the big trend of dynamic in the company's history, you will find that it started with friends, right? So, all the updates there basically come from the content posted by your friends. Then we entered a new era, and we also added creator content. Now, a large part of the content on Instagram and Facebook is not from your friends. It may not even come from people you directly follow.

It may just be recommended content from creators, and we can determine through algorithms that this content is interesting, attractive, and valuable to you. I think we will add a new category of content, namely content generated or summarized by artificial intelligence, or existing content integrated by artificial intelligence in some way. I think this will be very exciting for Facebook, Instagram, Threads, or other types of information flow experiences. We are starting to test different things around this.

I don't know if we know what will really work. Some things are very hopeful. I don't know - I guess this won't have a significant impact on the business for 25 years. But I think - I am very confident that this will be an important trend and one of the important applications in the next few years.

But you will get these, you will get Meta AI, AI Studio, commercial AI, and many things that developers can do with Llama 2.

Susan Li: I'm happy to answer these two questions. So, your first question is about the plan to provide advertising in commercial search. I think I may have mentioned this in an earlier question. Currently, we are really focused on making Meta AI as attractive and valuable a consumer experience as possible. We believe that as time goes by, the search scope of people using it will continue to expand. I think that as time goes by, profit opportunities will exist - as time goes by, we will achieve this. But now, I want to say that what we really focus on is consumer experience. This is a strategy for the products we launch to the world. Before focusing on profit prospects, the first thing we need to consider is consumer experience.

The second part of your question is about Reality Labs. Currently, we have not shared our expectations beyond 2024. Of course, when we consider the budget process for Reality Labs in 2025, we are considering where we want to focus and energize. We are very excited about the progress of smart glasses and the strong interest of consumers in them. Therefore, we are considering where we can make appropriate investments after the consumer growth momentum we see. Overall, I think Reality Labs is clearly one of our long-term strategic priorities, and we expect it to be an area where we make significant investments - as we are moving towards the very ambitious product roadmap we have developed there.

Susan Li: Your first question is about the revenue background in 2024. You mentioned events held every four or five years, so I think we are talking about the Olympics. Historically, we have never seen an event like the Olympics bring meaningful incremental contributions.

We believe that this year's situation is basically like this. Therefore, when we consider the prospects for the fourth quarter and next year, we usually expect growth to continue to benefit from the healthy global advertising demand we see. We believe that our investment in improving advertising effectiveness will continue to bring benefits to advertisers. But obviously, there are various macro backgrounds, which we try to reflect in the series of revenue guidance we provide. But I don't know how many specific events will have a significant impact on revenue in 2024.

Your second question is about the monetization of WhatsApp and our current situation. Now, I would like to emphasize again that click-to-message is indeed a key area of focus for us. We see that this area continues to receive attention. Especially the growth momentum of click-to-WhatsApp ads is still strong. Therefore, we will continue to focus on expanding the scale of click-to-WhatsApp ads in more markets with high WhatsApp user usage (such as Brazil). In the US, this trend is obviously still relatively early, but we see good growth momentum in click-to-WhatsApp ads, and we will continue to invest in the expansion and consumer usage of WhatsApp in the US, which will create greater opportunities in the future. Of course, we are doing a lot of work to make click message ads more effective and help advertisers focus on specific conversion events they care about. I would say that another source of revenue for WhatsApp is paid messaging. This quarter, paid messaging continued to grow strongly. In fact, it is still the main driver of growth for other revenue lines in our application series, with a 48% increase in the third quarter. We found that the overall volume of paid conversations has increased significantly, thanks to the growth in the number of businesses adopting paid messaging services and the growth in conversation volume per business.