A：The current structure of the AI industry is similar to a regular pyramid, where the bottom layer of chips captures the majority of the value, while the value of the models and applications at the top layer is minimal. This structure is clearly unhealthy because chips themselves do not directly generate value, it is the application layer that truly creates value. There are doubts in the market now, suggesting that there is a bubble in AI, and the ideal inverted pyramid structure should be one where the chip manufacturers make enough money to allow models to capture 10 times the value, and applications developed based on models can generate 100 times the value.

A：In the past year, the capabilities of AI models have gone beyond the scope of chatbots. Technologies such as digital human technology, code intelligent agents technology, and intelligent agents that can autonomously evolve to find global optimal solutions have all made significant progress. Now, the focus is on how to integrate AI with tasks in various industries, becoming a driving force for enterprise development and personal growth. This enables businesses and individuals to internalize and effectively use AI capabilities to improve decision quality, discover growth opportunities, reduce costs, increase profit margins, and innovate faster.

A：The emergence of AI effects is reflected in the renewal of business decisions, the burst of creative inspiration, the reshaping of user experience, and has become a new growth engine for enterprises and individual creativity engine. For example, in the context of intelligent life, digital human technology achieves more natural human-machine interaction through multimodal new product forms, is widely used in various fields, and has achieved significant effects in e-commerce live streaming, such as increasing conversion rates in live broadcast rooms and reducing order costs. At the same time, real-time interactive digital humans will play a greater role in industry consultation, after-sales service and other scenarios, providing efficient connections.

A：Digital humans have many applications in smart living, such as the performance of digital human anchors in e-commerce live broadcasts, which has achieved the effect of a super-real person, increasing the conversion rate of the live room and reducing order costs. The digital human live room created by Zero Run Cars in collaboration with Baidu has increased the conversion rate of leads by 44% and reduced order costs by 64% through the intelligent capabilities of digital humans. In addition, digital humans are also applied in knowledge popularization, historical and humanistic emotional interpretation, and other fields, becoming the intelligent partner of content creators.

A：Real-time interactive digital humans have characteristics such as real-time response, multimodal coordination, and expressive voices, and they can provide timely feedback based on real-world information and make judgments based on the on-site environment. In the future, they will play a greater role in service scenarios such as industry consultation and after-sales service that require high trust and timeliness, helping businesses solve the problem of scarce high-quality anchors and improve service efficiency and quality.

A：The highlights of this Winter Olympics are abundant, including the Italian-style opening ceremony that cannot be missed and the popular events where the Chinese team has a chance at winning gold. In addition, besides watching the games, you can also visit the Milan Cathedral, experience the local culture, and feel the exotic New Year festivities in the second half of the games. The Milan Cathedral is an important landmark in the center of Milan, with 135 spires and over 3000 statues, like a miracle. Stepping into the cathedral, you can see 52 towering stone columns, a soaring dome, giant stained glass windows, and lifelike sculptures, all in a solemn and holy architectural style.

A：After the release of Wenyi's statement by Baidu in 2023, they decided to become the first company to AI restructure all of their products, with the first product to be restructured being the search engine. Through AI restructuring, Baidu's search engine has transitioned from being primarily text-based with links to AI applications focusing on multimedia content. Now, the majority of search results are generated by AI, with self-media coverage of the top results reaching up to 70%. The AI-restructured search engine is capable of handling ambiguous questions, providing multi-modal understanding, and allowing users to ask questions in a more flexible manner. For example, for niche questions, AI search can use AIGC to generate video content; for personalized needs, it can call on digital human intelligence for one-on-one consultations; at the same time, AI search can also call on real-world service links to simplify complicated operations, such as providing one-click access to information on electric vehicle subsidies and offering purchase links.

A：AI search, based on user interests, personality traits, etc., can infer the user's next intention and generate personalized content. For example, based on a user's gastrointestinal disease, raising orange cats, and interest in AI tools, it actively customizes dietary recipes for them, providing recommended content that better fits user needs.

A：Baidu will open up its 25 years of accumulated search technology to global developers in the form of AIAPI, collaborating with manufacturers such as Samsung, Honor, Vivo, etc., and enabling its underlying capabilities to be widely used in various AI applications. Currently, 625 manufacturers have connected to Baidu Intelligent Cloud and are using the search API, making Baidu's AIAPI a foundational technology in the industry.

A：To configure WeChat Pay, you first need to fill in the APP ID and merchant number, and then upload the API certificate and public key. After a simple few steps of configuration, WeChat Pay setup can be completed.

A：The merchant backend has functions such as product management, order management, and user management. To add a new product, you need to enter the product management module, upload product images, fill in product details such as name, price, stock, and category information, choose whether to enable the new product recommendation function, and finally publish the product. Users need to log in with a mobile phone verification code, and after logging in, they can browse the products on the shelves of the merchant, such as down jackets. Through the try-on function provided by the AI11 smart body, users can see the effect of wearing the clothes, then choose to buy immediately, fill in the delivery address, and make payment via WeChat Pay. After successful payment, they will receive the product.

A：AI11 enhances the shopping experience by identifying the needs and integrating relevant plugins, such as the photo merging plugin, allowing users to see the actual effect of purchasing goods in real time.

A：Users can push their applications to the public network through the "publish" function and after being reviewed, they will be included in search engines. As a leading global autonomous driving service platform, Carrot Run has already achieved fully autonomous commercial operation, providing shuttle services in places like Abu Dhabi and planning to expand to more cities. It is estimated that by 2025, the number of global travel service instances will reach hundreds of millions, with costs significantly reduced, allowing more people to accept and use autonomous driving travel services.

A：Since its establishment in 2011, the Kunlun chip team has undergone ten years of research and development. In the first half of this year, they successfully activated the Kunlun chip P800 30,000 card cluster, which is an important milestone in the development of domestic computing power. The P800 not only undertakes the majority of inference tasks within Baidu, but also trains multimodal models based on a 5,000 card cluster. As it expands to clusters of more than 10,000 cards, it is training even larger-scale models. In addition, Kunlun chip has been widely used in various industries such as internet finance, energy manufacturing, transportation, and education, providing services to hundreds of customers including China Merchants Bank, Southern Power Grid, China Iron and Steel Co., National Pipe Network, Geely Automobile, and vivo. The delivery scale has expanded from dozens of cards to clusters of over 10,000 cards.

A：The new generation Kunlun Xin M100 is optimized for large-scale reasoning scenarios, which will significantly improve the inference performance of the MOE model and is expected to be launched in early next year. The next generation Kunlun Xin M300 will be launched in early 2027, mainly aiming at in-depth optimization of training and inference for super large-scale multimodal large models, with outstanding performance. These two chips are designed to provide ultimate performance and low cost to meet the autonomous and controllable AI computing power needs of Chinese enterprises.

A：As the size of AI models increases, the amount of communication between multiple chips during training and inference processes significantly increases, which can easily lead to communication bottlenecks in traditional single-machine multi-card nodes. To address this issue, Baidu has created super nodes by combining dozens or even hundreds of cards, enabling tight connections and greatly improving inter-card communication efficiency. For example, the newly released Tianchi 256 and Tianchi 512 super nodes, based on the P800 card, have powerful inter-card connectivity and outstanding performance, enabling efficient completion of training tasks for trillion-parameter models. In the future, Baidu will continue to release thousand-card and four-thousand-card super nodes based on the Kunlun chip M series.

A：Baidu provides the Qianfan platform in the agent info field, which includes the latest version of the Baidu Wenxin large model and more than 150 mainstream models for users to use. The Qianfan platform also supports model customization development, and has rich data services and tool services that can help companies convert accumulated data experience into understandable information for models. At the same time, through Qianfan's tool services and MCB access to internal tools of the enterprise, users can build enterprise-level agents on top of it, which can be applied to core business scenarios. For example, Baidu and customers co-created a one-click multi-person collaborative SOP analysis agent, helping fast food restaurants accurately monitor employee operating procedures, correct errors in a timely manner, and improve production efficiency and service quality. In addition, this agent can also be applied in the financial field to assist traders in quickly generating quote proposals and increasing business scale; as well as in the travel field, providing users with one-stop service through digital employees, optimizing user experience.

A：Now, with AI-based distribution grid monitoring agents and operation ticket audit agents, dispatchers can analyze alarm information within one minute and complete operation ticket audits within 100 seconds, achieving efficient and safe operation of the power grid system. Previously, highly manual processes have been replaced by automation, allowing dispatchers to focus on handling more complex scenarios and emergency response.

A：AI and agents can help businesses automate tasks at various stages, such as automatically downloading and categorizing data, organizing notes, and conducting data analysis, greatly enhancing work efficiency, and converting model capabilities into the productivity of the enterprise.

A：Baidu Smart Cloud, with intelligence as its core, has been committed to providing an intelligence-first cloud platform since its establishment ten years ago. In the future, Baidu Smart Cloud will continue to invest, enhance its intelligent capabilities, expand intelligent penetration, provide solid AI infrastructure for enterprises, and support application innovation for Chinese businesses.

A：G Flow 3.0 has the characteristics of a universal intelligent agent on all platforms. After upgrading to version 3.0, it can efficiently and automatically complete a series of tasks such as data analysis, information organization, document writing, etc. It can also provide personalized services based on the user's habits and preferences, and has a global memory and intervention mode, enabling it to simultaneously advance multiple tasks and ensure high-quality completion.

A：In the workplace, users can delegate tasks such as data analysis, prototype design, and summary reports to g flow 3.0 for processing. It can quickly execute and integrate data in various formats, and adjust content and format in real-time according to user needs, quickly generating high-quality results.

A：Users can unleash their creativity through g flow 3.0, such as transforming photos into artistic works or fun videos. It supports collaboration among multiple people, making the creation process easy and enjoyable. The final works generated vividly reflect users' creative ideas.

A：Baidu Wenku and Wangpan's products significantly enhance users' data management and work efficiency by providing one-stop intelligent services such as intelligent downloading, classified management, automatic alarm task execution, global memory, and human-machine collaborative creation. It has become the world's largest general intelligent body - Baidu, with over 20 million active users.

A：The main application scenario of the genetic algorithm is to find the global optimal solution, borrowing from the concept of evolution theory in the biological world. It can simulate billions of years of evolution process and discover the global optimal solution or the best solution to complex problems that have never been discovered by humans in just a few days or hours. Compared to traditional methods, genetic algorithms can dynamically iterate based on changing conditions, provide the optimal solution, and are highly versatile, applicable to various fields such as transportation, energy, finance, and logistics. By simulating genetic variations and natural selection in the evolutionary process in the biological world, genetic algorithms achieve the ability to rapidly abstract problems, model, and automatically iterate and refresh the optimal solution based on environmental conditions. This means that the solution is not fixed, but dynamic and up-to-date, able to simulate or even surpass top algorithm experts in finding the best solution.

A：In the scene of commuting across the river between the new and old city areas of Ejin Horo Banner in Ordos, through the use of Baidu Intelligent Cloud SaaS Credit Control Platform and AI Traffic Light Configuration Plan, the average delay of vehicles has been reduced by 13%. It can also adjust and optimize the signal light configuration efficiency in real time, effectively reducing the average delay of vehicles and improving traffic efficiency even in situations of increased traffic flow.

A：In the field of finance, artificial intelligence can automatically evolve a large number of risk feature combinations in a short period of time, uncovering high-value features that are overlooked during manual selection, and generating the optimal risk control solution. For example, in the application at Citic Bank, artificial intelligence not only quickly reached the level of professional data mining engineers, but also provided a more comprehensive and accurate assessment than manual monitoring, significantly improving risk control capability.

A：In the planning of the pipeline network on offshore energy platforms, Fa Mu aims to achieve breakthrough changes in cost and efficiency by optimizing the platform layout to the extreme. This helps engineers efficiently plan the complex layout of the pipeline network, saving costs and time, while allowing manpower to focus on more macro creative issues.

A：Baidu partners with Chinese Shulian Logistics and Liaoharbour Group to use cutting-edge technology to achieve intelligent scheduling and path optimization for the largest port in the Northeast region. This allows for efficient collaboration between ships, cranes, trucks, and factories, achieving the optimal balance between efficiency, cost, and risk, significantly improving port operational efficiency.

A：The application of decision-making algorithms has gradually infiltrated into more fields, including natural gas demand forecasting, electricity trading centers, logistics mainline scheduling, and precision irrigation in farmland, solving efficiency improvement problems in various industries' core processes. In addition, in fields such as new drug research and development, inventing new materials, and the birth of new theories in natural science, decision-making algorithms can accelerate the discovery of more possibilities, demonstrating their enormous potential in scientific discovery. Decision-making algorithms are now officially open and are provided through invitation codes. Enterprises and ecological partners are welcome to submit applications to jointly explore opportunities for extreme efficiency improvement.