Yiwen Understands Baidu World: Is ERNIE Bot 4.0 comparable to GPT-4 model and the way out clear?

"Without rich AI native applications built on the basic model, the basic model has no value." At today's Baidu World 2023 Conference, Robin Lee pointed out the "way out" of the big model

"Without rich AI native applications built on the basic model, the basic model has no value." At today's Baidu World 2023 Conference, Robin Lee pointed out the "way out" of the big model. Robin Lee has said on many public occasions that all products should be redone with AI.

According to NetEase Technology, the application layer of Baidu's mobile ecosystem has already launched an internal "competition", and all products are being reconstructed based on AI native thinking. Today, Robin Lee has finally fulfilled his promise and demonstrated at the conference the new achievements of Baidu search, library, online disk, map, marketing and other business lines under the support of AI.

oneWenxin large models 4.0: four capabilities provide a base for AI native applications

"This is the most powerful Wenxin large models so far, and its comprehensive level is no inferior to GPT4!" On the scene of the conference, Robin Lee made a bold statement. So where does Robin Lee's confidence come from? Since the last ERNIE Bot press conference was criticized by the industry as PPT demonstration and "ChatPPT", Robin Lee has made full preparations this time, demonstrating ERNIE Bot's ability to understand, generate, logic and remember.

Robin Lee first demonstrated his understanding of the model through a case of buying a house. In this case, a person wanted to apply for a provident fund loan in Beijing. The specific question was: "I want to go back to Chengde to buy a house. Can I use provident fund loan? What about the procedures? I work in Beijing." Robin Lee said that the case was very complicated, involving multiple conditions and intentions, However, ERNIE Bot can accurately understand that this is a disordered statement, and also understand that the subtext of "working in Beijing" is "paying provident fund in Beijing", and provides users with the correct solution.

Next, Robin Lee showed the generation ability of ERNIE Bot. ERNIE Bot has generated a group of advertising posters, five advertising copies and a marketing video through a picture of the material. These contents meet the standard size and format of advertising placement and can meet different advertising needs.

In addition, Robin Lee demonstrated ERNIE Bot's logical reasoning ability by solving a mathematical problem. From the description of the topic to the steps to solve the problem, and then to the summary of knowledge points, the logic and thinking of ERNIE Bot is commendable. Robin Lee pointed out that this logic ability is not only reflected in problem solving, but also in route planning of intelligent maps, intelligent assistant handling complex tasks, traffic light control in intelligent transportation systems, etc.

How about the memory ability of ERNIE Bot? Robin Lee tested ERNIE Bot's ability to remember the previous content by asking him to write a novel on the spot and intersperse disturbing questions. ERNIE Bot is able to grasp and maintain consistency in story background, character setting and dramatic conflict. This also means that ERNIE Bot can support other applications that need to be consistent.

Robin Lee concluded: "We have just experienced the four abilities of ERNIE Bot. Of course, most of the time, these four abilities do not appear alone, but complement each other. For example, to create an advertising copy, it needs to understand the creative theme, clarify the creative logic, and maintain consistency from front to back through memory. In solving problems, it also needs to comprehensively apply the abilities of understanding, generation, logic, and memory. Through these examples, we can see that these abilities are How to play a role in real-life scenarios

At the end of the speech, Robin Lee comprehensively demonstrated the four capabilities of the Wenxin large models through an application scenario where a digital doctor helped patients interpret drug instructions. In this demonstration, digital doctors can help patients interpret the medical terminology and medication instructions in the drug instructions based on the drug photos and symptom descriptions provided by patients, and provide correct medication recommendations.

In Robin Lee's opinion, the ability of the Wenxin large models is the basis for the survival of all AI native applications. Similarly, without a rich AI native application ecosystem built on top of basic models, large models are worthless. At present, Baidu search, such as streaming, maps, online drives, and libraries, have all been equipped with new features. When users search for questions on Baidu, the new search will "no longer give users a bunch of links", but through understanding the content, generate multimodal answers of text, images, and dynamic charts, allowing users to obtain answers step by step. When dealing with complex needs, the "multi round interaction" feature can also meet users' more personalized search needs through prompts, adjustments, and other methods.

twoBaidu GBI: Mobile Business Analyst

There are many problems in traditional business analysis that require professional personnel to operate, and the data processing and analysis process is also time-consuming and labor-intensive. In response to this pain point, GBI, the first generative business intelligence product based on large models in China, was also presented to the public at this conference.

It is worth mentioning that GBI has natural language processing capabilities, which can directly understand enterprise requirements and generate corresponding analysis reports. That is to say, corporate executives can directly raise various questions to GBI without the need for professional personnel to operate. GBI Tongsha supports convenient data access, allowing enterprises to connect various data sources to GBI for analysis. GBI also has the ability to learn and customize development based on the needs and industry characteristics of the enterprise, making it an exclusive industry expert for the enterprise.

According to the on-site introduction, the advantage of GBI products lies in rapid response to enterprise needs. For business analysts, data analysis and report writing tasks that take more than ten days to complete can now be completed at the minute level. This increase in speed will reduce decision-making costs for enterprises. In terms of multi-dimensional data analysis, GBI can also assist enterprises in better understanding the market and customer needs.

threeNew AI native application: a personal assistant who understands cloud intelligence

How intelligent can traditional applications reconstructed by AI be? At the conference, Robin Lee announced that "our search, such as streams, maps, online disks, libraries, etc., will meet you in a new face." These applications can indeed become "personal assistants" in users' lives.

At present, Ruliu has become an indispensable work assistant for Baidu employees. It has functions such as real-time recording of meeting minutes, intelligent classification, and summarizing group chat information, which can improve work efficiency. Through the one click travel function of Ruliu, enterprises can more conveniently book air tickets, hotels, and various conference and exhibition affairs, making tedious work easier and faster.

Baidu Wenku has also undergone a brand new restructuring, which could potentially revolutionize users' previous work, learning, and creative habits. Through Baidu Wenku, enterprises can generate high-quality speeches, PPTs, and various document materials. At the same time, Baidu Wenku also has the ability to generate multimodal content and style selection functions, which can generate different styles of content according to user needs, improving the flexibility and efficiency of creation.

When using the latest version of Baidu Maps, users only need to express their needs, and the map can mobilize thousands of service interfaces to help users recommend restaurants, compare information from multiple locations, and provide travel suggestions.

Baidu Netdisk, as a cloud storage service provided by Baidu, has also received a new upgrade. The new version of the online drive not only achieves a transformation from graphical interface interaction to natural language interaction, but also enhances multimodal information understanding. The new version of the online drive introduces AI technology, which enables advanced functions such as understanding video content, extracting abstracts and sentences. Users can quickly and accurately understand video content, improving learning and work efficiency.

fourepilogue

With the changes of the times, "digitization" has become "digital intelligence", and now it has been upgraded to "AI based". Applications reconstructed by AI undoubtedly take the user experience to a higher level.

"China has rich application scenarios, and Chinese users are naturally willing to embrace new technologies. With an advanced basic model, we can build a prosperous AI ecosystem and jointly create a new round of economic growth." Robin Lee said. Looking forward to the future, Robin Lee believes that the future AI native applications must be multi-modal, and the physical world will be reconstructed beyond the information world.