From focusing on technology to focusing on application, what is the first stop for the deployment of large models?

To answer this question, we need to start from the application scenarios of the Internet era. "Search", as the most frequent contact scenario between us and information, has become one of the most extensive and high-frequency reconstruction needs.

Abroad, OpenAI has just released the AI search engine SearchGPT, challenging traditional searches such as Google and Bing; domestically, there are already more than a dozen AI search engines such as Baidu Search, Alibaba Quark, 360 AI Search, Mita AI Search, and Tiangong AI, which have dazzled users with their choices.

Advertisement

Which one is easy to use? The user volume may be a more convincing indicator. According to the report by Zhi Dongxi on August 8, one of the most recognized AI application lists in China, "AI Product List" (aicpb.com), released the July list recently, and Baidu Search's AI intelligent answering function topped the domestic AI product list.

In this AI product list, Baidu occupies 2 out of the top three spots in the domestic list. Among them, the first place is Baidu Search's AI intelligent answering, and Baidu Wenku and Wenxin Yiyan are ranked 3rd and 5th respectively. In addition, the new product "Orange Chapter" launched by Baidu Wenku also landed on the "Global New Product Growth List", becoming the fastest-growing AI product in China.

Baidu started with a search engine and has taken the lead in the AI era with the Wenxin large model. With the support of the Wenxin large model, Baidu Search now not only has the function of AI intelligent answering but can also transform into a medical expert to interpret examination reports and become a PS photo retouching expert. Based on the Wenxin intelligent body platform, Baidu Search has accessed the intelligent bodies of the Wenxin intelligent body platform, which can answer more vertical and professional questions.

How easy is Baidu Search with a large model? What's the difference compared to other AI search products on the market? Zhi Dongxi has tried it out.I. Wenxin's Large Model Reconstructs Search for More Accurate Information, Recognizing Everything with a Photo

In the Internet age, search engines have become an important tool for us to obtain information in our life and work. However, traditional search engines still have pain points in many aspects, such as limited understanding of complex semantic relationships and context, inability to accurately understand user needs, and difficulty in processing non-textual information such as images and audio, which limits the efficiency and quality of our information acquisition.

In the AI era, how does a search engine enhanced by a large model solve these dilemmas?

Firstly, compared to a brand-new AI search application, adding AI capabilities to the existing search engine is more in line with users' usage habits. Taking Baidu search as an example, when you open the Baidu App, you can see that the search box entry has not changed much from before.

1. AI Intelligent Answer: Dynamic Entry, Automatic Scheduling of Intelligent Agents

As usual, I only need to enter my question in the search box, such as "Men's singles table tennis champions of all Olympic Games".

In traditional search, we might get a lot of matching results containing related content, and we need to summarize the required answers by ourselves. Now, Baidu search will automatically call the AI intelligent answer capability, directly "chewing" the knowledge and feeding it to me, quickly listing a complete list.Of course, not all questions require the invocation of large models to be answered, hence the AI response entry is set to be dynamic.

For example, when I search for "Who is bigger, 9.11 or 9.8?", Baidu search will first present traditional search results and provide a button for me to choose.

After pressing "Answer", Baidu will call upon the AI's intelligent response capability using a large model to provide an answer. It seems that this question that has collectively stumped the large models did not pose a challenge to the Wenxin large model.

In addition to general models, AI intelligent responses can also call upon corresponding intelligent agents based on intent recognition. For example, when I search for "When is the best time to visit Singapore with the fewest people?", it automatically calls upon the authoritative Singapore Tourism Board official intelligent agent, and even creates a chart to visually represent the data.

If I want to further inquire, I can also click on the "Chat" button at the bottom to enter into a conversation with the intelligent agent.

The advantage of intelligent agent scheduling is that it can obtain more vertical and effective information.

In contrast, when I use several similar AI search tools to search for this question, the answers I get either do not focus on the issue of "the fewest people", or the basis provided is insufficient to support the answer. Compared to this, the search results obtained based on the intelligent agent are more professional.2. Multimodal Search: Instant Image Inquiry and AI Interpretation of Medical Reports

The search for multimodal inputs is a significant limitation of traditional search engines. With the support of AI, Baidu's multimodal search AI capabilities have also been upgraded based on traditional strengths, not only recognizing images but also interpreting medical reports, creating images, and more.

For example, when I come across something interesting in my daily life but do not know exactly what it is, I can simply take a photo, and Baidu Search's "Recognize Everything" feature can provide me with an answer. In addition, by combining the capabilities of large models, I can continue to ask "how to learn," and I will receive detailed learning steps and skills.

▲ Recognize Everything

Sometimes when we receive the results of an examination or physical examination, we may not understand them and need to consult a doctor, which requires re-registration. In such cases, we can first use Baidu Search to get a general understanding of the situation.

For example, if I upload a serum immunology test report, the "Recognize Everything" feature, which has learned a large amount of medical knowledge based on the Wenxin large model, can quickly interpret for me what abnormalities there are and provide health and consultation advice.

▲ AI Medical Report Interpretation

In addition to recognition, Baidu Search has also integrated generative AI image capabilities and launched the "AI Image Assistant" feature, including photo editing, image extraction, redrawing, and image expansion.The previously viral claymation style transformation can now be achieved within a search engine.

▲Style Transformation Feature

After a trial run, my impression is that compared to traditional search engines, the Baidu search rebuilt with the Wenxin large model has significantly reduced the cumbersome processes of information filtering and summarization, making up for the lack of multimodal search, and providing a more rich and comprehensive way to obtain information.

Compared with similar AI search products on the market, the need for additional downloads of Apps, intelligent body scheduling capabilities, etc., has brought differentiation to Baidu search. The search results obtained based on the intelligent body are more professional and can meet more vertical and segmented search needs.

Secondly, the search box leads to everything, and the intelligent body is always available on call.

Efficient retrieval of information is just one aspect. On the other hand, the search box in the AI era has broken through the traditional constraints and is no longer limited to simple text or image searches.

For example, when I search for keywords such as "watermark removal," "Beijing passport application," and "Olympic blessings," Baidu will automatically recognize the intention and display the corresponding entry points for internal and external functions, allowing you to achieve the desired functions without switching Apps.

▲Search can obtain different functional entry pointsThis is like a window that can lead to any place. With semantic understanding and intent recognition technology, the Baidu search box seems to have become a channel that can lead to everything.

When searching for "watermark removal" in similar AI search applications, although I can get detailed tools and methods, I still need to download or switch to other applications to find the corresponding functions, which is somewhat like "talking about war on paper."

▲ Comparison of similar AI search products

Scheduling large models automatically in search is more of a passive process of using AI capabilities. In work and life, we sometimes encounter more difficult problems that require multiple rounds of interaction to solve.

Among them, the industry generally believes that agents are one of the solutions to this problem. At the 2024 World Artificial Intelligence Conference held in July, Baidu founder, chairman, and CEO Robin Li said that agents are the direction of AI applications he is most optimistic about, and search is the largest entrance for agent distribution.

To meet these in-depth needs, Baidu search has built-in AI assistants and agents, which users can actively use through a simple entry.

For example, when I want to further inquire about the search results, I can click on the AI assistant bar above to switch, and the search interface becomes a chat window similar to ChatGPT. In addition to continuing to ask questions, Baidu also provides relevant agent recommendations at the top.

Another entry is at the bottom of the App, click on the message bar to directly enter the AI assistant interface.AI Assistant Entry

The AI assistant also has the capability to schedule intelligent agents. For example, when I ask "How to refuse someone's request for a loan with high emotional intelligence," the AI assistant not only provides an answer but also suggests further inquiry directions and suitable intelligent agents at the bottom. By clicking on this intelligent agent, you can summon the rejection assistant in the chat box to get more targeted advice.

Intelligent Agent Distribution and Multi-Round Interaction

In the AI assistant interface, clicking on the small square at the top right corner allows you to enter the intelligent agent plaza, which uses various intelligent agent tools that cover various specific needs in work, study, and life.

Intelligent Agent Entry

Some intelligent agents will provide a GUI (Graphical User Interface) during the conversation, allowing users to input their needs more simply without having to learn complex prompt words.

GUI GuidanceThe AI assistants and intelligent entities found in Baidu's search results all originate from the Baidu Wenxin Intelligent Entity Platform. As of July this year, there have been 200,000 developers and 63,000 enterprises that have joined the Baidu Wenxin Intelligent Entity Platform. This platform provides developers with Baidu's ecosystem plus external distribution paths and diverse business opportunities, helping developers to complete the business loop and assist them in "development + distribution + operation + monetization."

Overall, although Baidu's search has integrated a lot of AI capabilities, the entry point is still the original input box, so it does not appear "bulky."

The search box directly leads to various functional entrances, which has greatly improved the interactive experience of Baidu's search compared to traditional search engines and similar AI search applications, saving users a lot of time and effort. The built-in AI assistants and intelligent entities also make its functions more comprehensive.

Conclusion: In the new battlefield of AI search, traditional search manufacturers have an advantage.

In this new battlefield of AI search full of opportunities and challenges, we have seen the great changes brought to the search experience by technological innovation.

However, for AI startups, this track has brought many difficulties such as high index library costs and user retention difficulties, and it is not easy to stand out in this competition. In contrast, traditional search engine developers still have certain advantages with their years of accumulated technology, data, and user base.

As a major manufacturer that started with search engines, Baidu, with its underlying algorithm capabilities, has made Baidu's search more advanced in both search capabilities and interactive design, bringing more intelligent, convenient, and efficient search services.