Where is the biggest development opportunity in the era of big model? Basic service or industry service? Li Yanhong, founder, chairman and CEO of Baidu, believes that it is not, but in application. Among the 18 "catch-up rules for starting a business" published recently, Li Yanhong said that for entrepreneurs, the big model is meaningless and the application opportunities of the volume are greater. Baidu should be the first company to "reconstruct" all its products.
On October 12th, Baidu held the "Baidu World 2023 Media Pre-communication Meeting", which revealed the latest development of network disk, intelligent work platform such as Streaming and Apollo Intelligent Cabin based on large model reconstruction. In addition, it is understood that at the "Baidu World 2023" held on October 17, Baidu will also release the first generative business intelligence product in China-Baidu GBI.
The first generative business intelligence product in China
In the business world, shopping malls are like battlefields, and enterprises compete, not the big fish eat small fish, but the fast fish eat the slow fish. If an enterprise wants to win, the most inseparable thing is business analysis. However, traditional BI tools are difficult to find data, use tools and have limited index coverage. Both Excel and data analysis platforms are designed for high-frequency (preset) scenes, so they can’t flexibly deal with various problems at any time. The analysis and insight of a complex problem often takes hours or days to complete. In the fierce and rapidly changing market competition, such a long decision-making cycle may lose important business opportunities.
Sun Ke, Chairman of Baidu AI Cloud Technical Committee and Chief Architect of Baidu AI Cloud Application Products Center, said that Baidu GBI, which will be released on October 17th, can shorten the data analysis from days to minutes.
First of all, traditional BI can only be operated by professionals, while GBI can directly understand the president’s question, implement it in real time, and draw conclusions quickly. Secondly, GBI provides a convenient access mode, so that enterprises can access data, ask questions and analyze any data, and there is no need to manually analyze across databases and tables. Third, GBI also has the ability to learn, and enterprises can inject professional knowledge into the industry and make it an industry expert.
Baidu’s network disk will be upgraded again, and everything in the video will be refined.
"In the AI era, the network disk no longer focuses on file transfer or storage," said Wu Tianhao, general manager of Baidu AI Cloud Network Disk Products Department. "Instead, it further moves towards knowledge management of individuals and enterprises, and realizes the transformation of information from data to knowledge."
In the past 11 years, Baidu’s network disk has served 800 million users, and users will upload more than 1 billion pictures every day. Therefore, in the direction of AI reconstruction, Baidu’s network disk focuses on the intelligent service of personal files.
"One Cloud" is Baidu’s first domestic network disk intelligent assistant based on Wenxin model. It not only realizes the transformation from graphical interface interaction to natural language interaction, but also enhances multimodal information understanding. Users can operate files, pictures and videos in the network disk with only one sentence, which is convenient for users to "find things" in the network disk and video. It is worth mentioning that Baidu’s network disk "Cloud One" can also help users quickly understand the video content, refine the content focus from the video, add subtitles to the video, export all the subtitles to the manuscript, and even add the title of the manuscript, greatly improving the information understanding and communication efficiency.
Recently, Baidu’s network disk also launched the "Cloud One File Assistant" which can be used on WeChat. By forwarding any WeChat official account article to Cloud One, it can "read" directly, and summarize and refine the main points within 10 seconds, so that users can get the "essence dry goods" in a short time.
"A powerful video processing and understanding ability of Baidu Netdisk Cloud will also be shown to you at Baidu World Congress, so stay tuned," Wu Tianhao said.
The gospel of migrant workers is to have a "super assistant" who understands you and is professional.
At present, there are still many pain points in the online office of enterprises. How will the big model reshape intelligent work?
"The big model is a great opportunity in the field of corporate office." He Wei, director of Baidu’s intelligent office platform department, said, "With ERNIE Bot’s blessing, Baidu’s intelligent work platform has reconstructed intelligent work based on AI’s original thinking, stimulating enterprise innovation and improving efficiency."
He introduced the "super assistant" created by Ruliu, which has three characteristics: knowing you, professionalism and real-time companionship. It has been aroused at any time and anywhere, and has been applied in a wealth of office application scenarios.
During the execution of intelligent tasks, the super assistant evokes through natural language voice, and realizes one-step direct access to complex systems for scenes such as meeting reservation, vacation and poor travel. In the intelligent document processing scene, the super assistant can quickly find out the relevant documents according to the instructions, which doubles the efficiency of knowledge acquisition, and can also quickly consult, summarize and translate the documents on the Web side of the browser; In the high-frequency communication scenario, "IM intelligent summary" and "AI meeting insight" and "AI meeting minutes" can quickly refine the main points and generate structured minutes. The data shows that through the use of the AI conference insight function, the current reading volume of conference content has increased by 3.5 times.
He Wei believes that "unlike the stronger tool attribute under the concept of assistant, super assistant can be competent for more complicated tasks, and at the same time will help you finish your work more actively, so that intelligent work can replace hard work." At the scene of the 2023 Baidu World Congress, the reconstructed new features such as streaming will be unveiled.
Large-scale model reconstructed intelligent cockpit, which will be carried in the mass production of Extreme Yue 01.
The big model not only reshapes thousands of industries, but also profoundly affects the automobile industry. When the large model is combined with the car cockpit, the car will become an automobile robot with EQ and IQ.
Su Tan, general manager of the intelligent car business department of Baidu Intelligent Driving Group (IDG), said: In the era of big model, based on the idea of reconstruction, we have the opportunity to change the relationship between man and machine in the car cockpit into the relationship between man and virtual person. Based on the Wenxin model as the basic model, and the massive data accumulation of Baidu Apollo in millions of smart cars and different scenarios, the Apollo intelligent cabin model and intelligent cabin development tool chain are further enhanced.
The intelligent emergence of large-scale models brings core abilities such as understanding, generation, reasoning and memory, so that the format of intelligent cockpit will be reconstructed, including interaction, development mode, architecture and user operation mode. The first thing to be reconstructed is the interaction between people and vehicles. From "imperative" to "conversational", the naturalness of interaction will be greatly improved. The improvement of interactive experience will also greatly improve the user experience of just-needed scenes such as in-car navigation and car use. At the same time, the demand for in-car entertainment and services will be further released, and the user’s usage scenarios will also be reconstructed. Baidu Apollo has built an exclusive large-scale model technology base for the car cockpit. Based on the Wenxin large model, it has enhanced the effect of the large model in the car cockpit and provided more anthropomorphic intelligent interaction, including improving the understanding in the cabin, adding multi-modal understanding, active interaction ability, dynamic recovery ability, response time optimization and other capacity building to meet the intelligent needs of users for the cockpit.
In order to make the smart cockpit empowered by the big model land more quickly, Baidu Apollo has also reconstructed the technical route of the smart cockpit of the car, and the development mode of the native application of the car AI based on the big model is deeply integrated with the localization. Apollo streamlines and implements the intelligent cabin AI native application development paradigm, reduces costs and improves efficiency in the whole link, and helps the industry to explore. Automobile host manufacturers can independently develop their own brand models and application scenarios, and Apollo will also provide a reference for boutique in-vehicle native applications, and build a large number of commonly used plug-in scheduling capabilities for the cockpit ecological environment, greatly reducing the input costs of automobile host manufacturers.
At present, the car voice products blessed by Baidu Apollo Intelligent Cabin Model have been mass-produced in models such as Extreme Yue 01, Cadillac Ruige, Buick E5, Geely Yinhe L7 and Geely Yinhe L6, and brands such as Geely Yinhe and Haval will soon be launched. Among them, Extreme Yue 01 is at the leading level among the mass-produced vehicles in the industry in terms of its core capabilities such as millisecond response, full-time wakeup-free interaction, multi-channel simultaneous interaction, and full-page seeing and speaking.
The scale of users of Qian Fan’s large model platform has grown rapidly, covering more than 400 scenes.
Baidu is reshaping its products with big models, so the next question for entrepreneurs and enterprises is, how to build AI native applications?
Li Jingqiu, deputy general manager of Baidu AI Cloud AI platform, said at the communication meeting that Baidu has summarized these successful practical experiences into tools and methodologies, and deposited them on the world’s first one-stop large-scale model service platform "Qian Fan" to help entrepreneurs and enterprises build their own AI native applications with low cost and high efficiency.
Baidu AI Cloud continues to improve and upgrade the tool chain on the Qian Fan platform. At present, 41 data sets and 226 sets of high-quality Prompt templates are preset on the Qian Fan platform, so enterprises can quickly optimize the model effect according to their own business scenarios. The platform also manages 42 mainstream large models at home and abroad, providing Chinese enhancement, performance enhancement, context enhancement and other capabilities. Qian Fan platform provides an extremely stable training environment for the performance guarantee issues that corporate customers are concerned about. Under the conventional method, engineers spend 30%-40% of their time on fault tolerance and fault recovery. Now, the fault management mechanism of Baidu AI Cloud’s self-developed cluster networking makes the effective training time of the model reach more than 95%.
Li Jingqiu, deputy general manager of Baidu AI Cloud AI platform, revealed at the communication meeting that the user scale of Qian Fan platform has increased rapidly since March. By the beginning of September, there were nearly tens of thousands of monthly enterprises on the platform of Baidu AI Cloud Qian Fan, covering more than 400 scenes in manufacturing, energy, transportation and other industries.
"Around the accelerated landing of AI native applications, Qian Fan platform will release more new products at Baidu World 2023 on October 17." Li Jingqiu, deputy general manager of Baidu AI Cloud AI platform, said.