"Every product is worth redoing with big model technology," and now Baidu is turning that statement into reality.
On October 17, the 2023 Baidu World Congress was held in Beijing Shougang Park. As Baidu's most important product, technology and strategy conference, Baidu World Congress has been held for 17 consecutive sessions, and it will be held online from 2020 to 2022 due to the impact of the epidemic. After three years, Baidu World Congress returned offline this year, becoming Baidu's most significant offline event in recent years.
MORE IMPORTANTLY, THIS YEAR'S BAIDU WORLD CONFERENCE IS ALSO A MAJOR EVENT HELD BY BAIDU IN THE ERA OF AI BIG MODEL, THE THEME OF THIS BAIDU WORLD CONFERENCE IS "GENERATING THE FUTURE PROMPT THE WORLD", LARGE MODEL AI HAS BECOME THE ONLY THEME, BAIDU COMPANY AND ITS CORE BUSINESS HAVE NEW PRODUCTS RELATED TO LARGE MODELS, THE MOST CONCERNED LINK IS LI YANHONG PERSONALLY APPEARED "HAND-IN-HAND TO TEACH YOU TO DO AI NATIVE APPLICATIONS" THIS SCENE.
Baidu is a pioneer of AI technology in China, and has been paying attention to AI technologies such as deep learning since around 2010, and became a pioneer in deep learning AI technology in mainland China in 2013. Baidu has announced that it has invested more than 140 billion yuan in research and development in the past decade, most of which has been invested in AI technology. In 2023, the big model boom is rushing, and the "100-model war" is in full swing. When AI enters the big model stage from the deep learning stage, Baidu still plays a similar role in the AI industry: a leader in technology, a pioneer in application, and a promoter of landing. How to promote the faster implementation of AI models in the whole industry? AI native application is the answer given by Robin Li. AI native applications rethink the more natural interaction mode between humans and machines in the era of large models and redesign applications, and at the same time, the value boundary of AI native applications has also achieved a breakthrough. Unlike "chat" assistants such as ChatGPT or "graphing" tools such as Midjourney, which emphasize "fun and fun", AI native applications emphasize non-entertainment values such as work, study, life, and business.
What is the strength of AI native applications?
At the Baidu World Conference, Baidu released a number of AI native applications, all of which are well-known products based on large model technology and AI native thinking, thereby bringing users a completely different experience and value.
Search is Baidu's starting business, and it is also Baidu's "nuclear within core" business. Among the many AI native applications, the most popular is undoubtedly the new generation of Baidu search engine and the industry's first AI interactive search engine - Baidu New Search.
Over the years, search engines have been based on the logic of "query" to match answers to users in massive Internet content, although search giants such as Baidu are constantly improving technology, improving experience and even deeply applying deep learning technology, but the underlying technical architecture of search engines is "crawler + index + query". Baidu's new search reconstructed with a large model has three innovative capabilities: accurately understanding intent, prioritizing personalized content recommendations, and multiple rounds of interaction. The change of the underlying technical architecture has made the search engine no longer just a passive information query tool, but a "digital partner" that can actively respond to users and communicate with users, and the value of search engines has been completely redefined.
When ChatGPT became popular, some people thought it would disrupt search, but judging from the results, ChatGPT is difficult to become a daily search tool because of the uncertainty of results. On the contrary, Baidu New Search, a search engine reconstructed with AI native thinking, is what future search should look like.
Baidu's new library started as early as August this year, and has now been upgraded to a "one-stop intelligent document platform". Compared with the old version, the changes in the new Baidu library are mainly reflected in the upgrade of four major capabilities: PPT intelligent generation, document intelligent generation, intelligent editing and intelligent auxiliary reading, which Luo Chao Pro has previously evaluated, Baidu Wenku's AIGC ability is not just "fist embroidery legs", but can really help workers and other key users of office software, to achieve a significant reduction in repetitive work, a significant increase in work efficiency.
On the one hand, Baidu Library processes massive content, especially complex content such as documents, which is exactly the field of "big models are good at and deep learning is not good at", and can test the technical strength of large models; On the other hand, Baidu Wenku has long become a tool for people's work and learning, with high-frequency and heavy use and scenarios. Now it seems that after being armed with AI large model technology, Baidu Wenku has become a new generation of productivity tools, which will benefit more users from work and learning.
The AI native reconstruction of Baidu Maps is reflected in two levels: new interaction and new navigation. New interaction refers to the use of AI technology to improve the fluency of natural language dialogue functions, the accuracy of receiving and analyzing information. The new navigation refers to the newly launched "AI Wizard" function - this is a new map reconstructed through the Wenxin large model, which can provide a natural communication experience closer to a real person, and the user can not only communicate with the system at any time during use, but also actively inquire and accurately identify the user's intention.
For commercial scenarios, Baidu Xinruliu and the newly released first generative business intelligence product in China, Baidu GBI (full name Generative Business Intelligence, generative business intelligence), are the masterpieces of AI big model technology.
Baidu Ruliu is committed to providing enterprises with one-stop knowledge and information management services, and its main functions include enterprise communication, collaborative office, etc. After the reconstruction of the AI large model, the new Baidu Ruliu has reshaped the interaction mode, demand satisfaction mode and decision-making process, and incarnated as a "super assistant" for enterprise employees, and its interaction mode is mainly natural language, and users can book meetings and arrange travel through voice interaction, greatly improving efficiency; In terms of functional requirements, it provides an end-to-end complete service chain, such as conversational business intelligence for BI scenarios, which integrates functions such as sales data query, analysis, and business opportunity identification, and users can get a detailed report by simply entering instructions through natural language; The operation steps are also greatly simplified in the process, such as the "IM intelligent summary" function supports multiple conversation summaries and one-click forwarding, which increases the reading efficiency by nearly 3 times.
If Baidu's advantage is comprehensive, Baidu GBI is strong in differentiation, especially compared to traditional BI tools.
The biggest pain point of traditional BI tools is low efficiency: limited indicators can be called, and data analysis formulas are rigid, which will reduce analysis efficiency and lengthen the decision-making cycle of enterprises. AI big model technology endows Baidu GBI with conversational natural language interaction capabilities, proactive data query capabilities and data analysis capabilities. In addition, Baidu GBI also connects massive data sources and connects enterprise data warehouses, breaking the data barriers of traditional BI tools and clearing obstacles for rapid data mobilization and integrated analysis.
(Image from Baidu GBI official website)
The above applications are all AI native applications, and it is not difficult to find their common points: based on AI large model technology reconstruction, the interaction can be as natural as possible, and it can be active and active, reducing user learning costs and improving user efficiency; Do more in functions and reduce user operation steps; The value does not emphasize "fun", but helps users to solve the problems faced in work and study.
What is the unique value of AI-native applications?
Technology is only a means to achieve the product rather than the end goal, why should Baidu launch AI native applications? The answer is that AI native applications have a completely different value than traditional AI applications.
AI applications in the era of deep learning are all "patches" that is, "optimization" of Internet applications, while AI native applications are "starting with the end" to think about what the application of the big model era should look like, and then reconstruct the application, which is the connotation of the word "native".
Why did "mobile QQ", which came out earlier than WeChat, fail to become a national mobile social tool? Theoretically, it is not difficult for mobile phone QQ functions and services to be done the same as WeChat. The deep-seated reason for the success of WeChat and the failure of mobile QQ is that WeChat is a mobile native application, and mobile QQ is not. Not only in the mobile Internet, but also in the development of many technologies of cloud computing, "building native applications based on new technologies" has shown great vitality.
Speaking of AI native applications, Luo Chao Pro believes that its new value is mainly reflected in several aspects:
First, it broadens the application boundary and extends the application scope of AI large model technology to work and learning scenarios, making it more functional and practical.
Prior to this, most AI applications for C-end consumption scenarios stayed at the level of entertainment and information acquisition, such as chatbots such as ChatGPT and Google Bard and Midjourney painting applications.
The native application of AI will completely open up the application space of large model technology, and gradually penetrate into various scenarios and industries, and "each product is redone with large model technology" is accelerating.
Take the cases mentioned earlier: in the work scenario, Baidu Xinruliu and Baidu GBI help the daily management of enterprises, and play an important role in meeting record analysis and collation, data analysis, document processing and other levels; In the learning scenario, in addition to generating PPT, Baidu New Library can also disassemble long articles and assist reading through intelligent abbreviation, continuation and other functions, and Baidu network disk cloud can summarize and refine the main points of the article, add subtitles to course videos, and export documents within 10 seconds, greatly improving learning efficiency.
Second, realize the penetration from online virtual scenes to offline real scenes, and comprehensively transform the physical world.
There are many ways to define AI in academia, but one thing is certain: it can better understand the real world, better understand human intentions, and better meet human needs.
The big model will accelerate the arrival of the stage of "industrial AI and AI industrialization", and AI will have a greater presence in the physical world. Compared with traditional applications, AI native applications have a direct impact on human real life, truly entering human learning, work and life, and building a bridge between the digital and physical worlds.
Li Yanhong introduced at the Baidu World Conference that Baidu AI native applications armed with large model technology have been widely used in real industries such as manufacturing, energy, electric power, chemical industry and transportation, such as relying on intelligent scheduling and intelligent information control of AI native applications to alleviate traffic congestion, and intelligent inspection can reduce potential safety hazards. In the era of big models, the real impact of AI technology on the physical world is becoming more and more extensive and deeper.
Third, it reshapes the human-computer interaction process, which is smoother, more natural, and more in line with human nature and user habits. The evolution of Internet digital technology has always had a dark line: human-computer interaction. PC is "keyboard and mouse" interaction, mobile Internet is "touch" interaction, and AI era should be "sensorless, natural, conversational" interaction.
Large model technology is strong in "cognition", machines can better understand human intentions and real scenes of the world, at the same time, large model technology can achieve AIGC, which allows machines to better express themselves and talk to people. Because of this, many of Baidu's AI native applications can achieve natural multi-round conversational interaction, they understand users better, and human-machine communication is smoother.
Marshall McLuhan, a 20th-century original media theorist, pointed out in his book Understanding Media that all technology is an extension of people. Ever since apes could make and use stone tools, humans have been looking for ways to extend themselves. The telephone extends the ears, the electric light extends the eyes, the transport extends the legs and feet, the knife and axe hoe and rake extend the hands, the weapons extend the teeth ... AI undoubtedly extends the human "brain" to a large extent.
AI native applications allow AI applications to be more deeply and widely applied in massive scenes in the physical world, acting on all aspects of people's work, study, life and entertainment, extending the human "brain" and "hands, eyes, legs, mouth" and other physiological abilities to a greater extent, and even the four words "life realization" have a new meaning.
Baidu brings "new tickets" to more companies
The value of large model technology has been quickly reached on a global scale, and enterprises from all walks of life are racing to embrace the wave of large model technology, for fear of missing a golden opportunity, it is not too much to describe the value of large models as the "new ticket" of the new era.
Some time ago, many large model players fell into the "arms race" of crazy roll parameters, which may be a detour, because large models to be widely used, it is not enough to have powerful technology, but also need the path of landing, and it can even be said that the most urgent thing for large models at the current stage is also landing.
As Robin Li said: Big model technology and AI native applications are two sides of the same coin, symbiotic coexistence: the progress of the former's four major capabilities of understanding, generation, logic and memory is an indispensable survival soil for AI native applications; Conversely, without the landing and popularization of AI native applications in various scenarios, large model technology will lose its practical significance.
Baidu launched a large number of AI native applications at this time, which is of great significance to the landing of large model technology:
First, set off a wave of AI native application development.
In the past six months, Li Yanhong emphasized that all applications are worth redoing with a large model. "Every product is worth redoing with large model technology", but it must be done with AI native thinking, not tinkering, must be torn down and restarted, and must be rethought, redesigned, redeveloped in product form, interaction logic, functional services and even business models, and new applications instead of "improvement".
Secondly, a clearer landing path for large models is formed.
From PC to Internet to mobile Internet to cloud computing, digital technology has a similar development path: incubation and incubation in B-side professional scenarios, amplification, expansion and cost reduction in C-side consumption scenarios, and finally C-side technology overflow and then return to B-side scenarios to empower thousands of industries to form "B/C collaboration".
Led by leading players such as Baidu, AI big models seem to be on a similar path: germinating in the labs of technology giants and taking the lead in B-side scenarios such as cloud computing - as early as 2019, Baidu launched Wenxin ERNIE 1.0 for the industry. After the emergence of large models on the industrial side, they will be popularized in the form of AI native applications in the form of C-end consumer experts such as search, library, map, and online disk, and eventually will naturally go to thousands of industries on the B-side to help the industrialization of large models.
Finally, further consolidate the leading position of large model technology.
Baidu's advantages in AI technology need not be repeated, but AI technology, including large model technology, will eventually become universal, and if the first mover advantage becomes a barrier, it must seize the opportunity in the scene landing and transform the technical advantage into a business advantage. Li Yanhong has long said that the native application of AI is the key to success or failure, because it is related to the improvement of productivity and whether technology can be truly applied to the real economy. Baidu has also been practicing itself, on the one hand, building AI native applications, allowing large model technology to widely penetrate into Baidu's own business, on the other hand, quickly exporting large model capabilities to the outside world through "windows" such as Baidu Intelligent Cloud, and now has the largest large model ecology in China.
Judging from the news released at the Baidu World Conference, Baidu not only has the top technology of "big killer" such as Wenxin Big Model 4.0, but also has the application landing ability and thinking such as AI native application. Making AI big models accessible to all people and all enterprises is Baidu's commitment to society, and "AI native application" is a constructive suggestion given by Baidu. For all enterprises, it's time to redo all applications with AI native application thinking and big model technology!