"The emergence of intelligence brought about by large models is the basis for us to develop AI native applications." October 17,Robin LiexistBaiduWorld 2023. On the same day, Robin Li delivered a speech entitled "Hand in Hand Teach You How to Make AI Native Applications", released the 4.0 version of the Wenxin Big Model, and brought more than ten AI native applications such as new search and new maps.
At the conference, Robin Li announced the official release of Wenxin Big Model 4.0 and the start of invitation testing. He said that this is the most powerful Wenxin Big Model to date, which has achieved a comprehensive upgrade of the basic model, with significant improvements in understanding, generation, logic and memory capabilities, and its comprehensive ability is "not inferior to GPT-4". Robin Li introduced that Wenxin 4.0 also started invitation testing at the same time. The audience can experience the professional version of Wenxin Yiyan by scanning the QR code of the guest certificate, logging in to the official website of Wenxin Yiyan or downloading the latest version of Wenxin Yiyan APP; in addition, corporate customers can also apply to test the Wenxin 4.0 API through the Baidu Smart Cloud Qianfan Big Model Platform.
He demonstrated on site more than ten AI native applications such as Baidu Search, Ruliu, Maps, Cloud Disk, and Wenku, which were reconstructed based on Wenxin Yiyan, hoping to expand everyone's imagination and "inspire everyone to work together to create more amazing AI native applications."
The most powerful Wenxin Big Model 4.0 is released, and its comprehensive capabilities are no less than GPT-4
In Li Yanhong's view, the birth of AI native applications benefited from the four core capabilities of understanding, generation, logic and memory of large models. Baidu's AI native applications were also developed based on Wenxin Yiyan. "These capabilities were not available in the past, and therefore can open up unlimited space for innovation."
Based on Wenxin Big Model 4.0, Li Yanhong demonstrated the characteristics and application scenarios of the four major capabilities in turn. In terms of comprehension, he asked about the policy of housing provident fund loans in different places, and demonstrated Wenxin Yiyan's ability to understand complex prompts such as disordered order, ambiguous intentions, and subtexts. For example, "working in Beijing" is equivalent to "paying housing provident fund in Beijing", etc. "Today, every word you say, it is likely to understand."
In terms of generation capabilities, Robin Li demonstrated how Wenxin Yiyan can quickly generate a set of advertising posters, five advertising copywritings and a marketing video based on a material image in just a few minutes. It is reported that based on this series of capabilities, Baidu has launched the AIGC marketing creative platform Qingduo, which allows "one person to become an AI marketing team."
At the same time, he also demonstrated the logical ability of the big model through scenarios such as solving math problems and summarizing knowledge points; demonstrated the memory ability of the big model through writing thousands of words of novels and setting up characters and plots; and demonstrated the comprehensive application of the four major capabilities by using digital human doctors to help patients interpret drug instructions.
"The previous demonstration shows the progress of the Wenxin model in the four major capabilities of understanding, generation, logic, and memory. These capabilities are the foundation for the survival of all AI native applications," said Robin Li.
More than ten AI native applications were released
Rich AI native applications are the value of the big model. At the conference, Robin Li announced that "our search, streaming, maps, cloud storage, document library, etc. will all meet you with a brand new look," and said that the purpose of sharing these applications is to expand imagination and inspire more people to create more amazing AI native applications.
Robin Li introduced that Baidu's new search has three characteristics: extreme satisfaction, recommendation stimulation, and multi-round interaction. When users search for questions, the new search will "no longer give you a bunch of links", but will generate multi-modal answers of text, pictures, and dynamic charts through understanding the content, allowing users to get answers in one step. When it comes to complex needs, the "multi-round interaction" feature can also meet users' more personalized search needs through prompts, adjustments, etc.
At the same time, Robin Li also demonstrated the first generative business intelligence product in China built with AI native thinking: Baidu GBI. It is reported that compared with the high threshold and difficult data analysis of traditional BI software, Baidu GBI can perform data query and analysis tasks through natural language interaction, and also supports professional knowledge injection to meet more complex and professional analysis needs.
By understanding and regenerating massive amounts of documents, images and videos, Baidu Netdisk and Wenku have acquired creative capabilities: Netdisk can not only accurately locate a certain frame of a video, but also summarize a 1-hour video content in a few seconds, and extract golden sentences and key points from it; Wenku is based on 1 billion high-quality materials, and can realize tasks such as writing articles and making PPTs, becoming a veritable "productivity tool."
Baidu Maps and smart office platforms such as Ruliu have also become more considerate travel guides and super assistants through their understanding and memory capabilities: on the map, users only need to state their needs, and the map will mobilize thousands of service interfaces to help users recommend restaurants, compare information from multiple locations, and give travel suggestions; Ruliu can target office pain points such as large amounts of information in group chats and "highlight key points in one second." The travel assistant can not only book air tickets and hotels, but can even summarize background information and conversation references for visiting customers by accessing company systems such as CRM.
As Robin Li said before, AI native applications are not a simple repetition of mobile Internet apps and PC software, but should be able to "solve problems that could not be solved or solved poorly in the past."
Plug-ins and APIs help the ecosystem prosper and promote economic growth
"Big Model will open up a prosperous AI native application ecosystem," Li Yanhong emphasized that plug-ins are a special type of AI native application with the lowest threshold and the easiest to use, allowing developers and entrepreneurs to quickly join the ecosystem. For example, he said that the "intelligent legal assistant" that Big Model accesses authoritative legal data can provide users with relevant suggestions for legal consultation, and the resume assistant plug-in can help users generate resume templates with one click.
According to reports, the data, capabilities or applications of individuals and enterprises can be quickly turned into AI plug-ins to enhance the capabilities of large models and make them more practical and easy to use. Li Yanhong said that a month ago, Baidu launched the Lingjing plug-in platform, and currently 27,000 developers have applied to join, covering multiple fields such as law, workplace, and learning.
When developing AI native applications, the basic capabilities of large models are crucial. Robin Li said that API is the main way for AI native applications to call basic large models. Enterprises and developers can call large model APIs including Wenxin Yiyan on Baidu's Qianfan large model platform. At present, Qianfan large model platform has become China's largest large model development platform, with 42 mainstream large models settled in, covering nearly 500 scenarios in various industries. From now on, enterprise customers can also apply to test Wenxin 4.0 API on Qianfan large model platform.
"China has rich application scenarios, and Chinese users are naturally willing to embrace new technologies. With advanced basic large models, we can build a prosperous AI ecosystem and jointly create a new round of economic growth," said Robin Li.
In addition, Robin Li said that the future AI native applications must be multimodal, and will reconstruct the physical world in addition to the information world. Autonomous driving is a typical application of visual big models to reconstruct the physical world. Big models will enable Baidu's autonomous driving capabilities to surpass empirical systems, handle complex scenarios more intelligently, and achieve wider space-time coverage. At present, Baidu's autonomous driving travel service platform Luobo Kuaipao has provided services for more than 4 million times in total, and has become the world's largest autonomous driving travel service provider.
"A large number of AI native applications will continue to emerge, and digital technology and the real economy will be deeply integrated...Big models are becoming an important driving force for new industrialization." Li Yanhong said. As the theme of Baidu World 2023 is "Generating the Future", at the end of his speech, Li Yanhong announced that we are about to enter an era of AI native, an era where humans and machines interact through prompts.