Author: He Sisi

Editor: Chen Caixian

Source: AI Technology Review

"Tomorrow's meeting requires a PPT. Prepare it when you get home tonight."

Most office workers, especially those new to the workplace, would shudder upon receiving such a notice.

Because this not only means overtime, but also entails a painful and complex execution process: collecting and organizing relevant background information, determining the theme style, finding suitable templates, which takes about one to two hours. Then, entering the production process: should this text be centered or aligned? What font size looks less abrupt? Where is the appropriate place to put this image? How should this section be laid out? After a series of operations, another two to three hours have passed.

The next day, when presenting the report at the meeting, the feedback is mostly "the theme is not highlighted, and the expected effect is not achieved…" In the workplace, such situations seem to have become commonplace, and PPT production has become the most headache-inducing task for office workers.

With the rise of large-scale models, intelligent documents have become the vanguard of AI technology, quietly changing the document industry, and have become an essential "magic tool" for improving work efficiency for many people.

According to statistics, Baidu Wenku, reconstructed based on the WENXIN large model, attracted over 2 million online users on the first day of its public release.

As a company that first advocated the concept of "AI native" in China, Baidu has always pursued the reshaping of products and applications with AI technology. According to AI Technology Review, since March, based on the foundational capabilities of the WENXIN large model, Baidu Wenku has been continuously updating and upgrading its AI functions.

I personally tested it: uploading a document to Baidu Wenku, it only takes less than a minute (about 40 seconds) to generate a complete PPT!

The second-level acceleration of productivity iteration is a leap for Baidu AI and an advancement for domestic intelligent document software.

PPT, seconds, one command, or even just a few words can solve it

Regarding the AI-generated PPT function, let's test it together:

Go to the Baidu Wenku official website, and on the right side, there is a "Document Assistant" port. In the dialog box, enter "/" to display the various functions of the document assistant: AI-generated PPT, Word to PDF, PDF to Word, intelligent editing, document chart… These are all functional modules that have already been launched.

First, let's look at the intelligent PPT generation function.

Users can select documents from Baidu Wenku or upload their own document content.

We found a "Mid-Autumn Festival text planning" document and uploaded it. Then, we entered "/" and selected the "Word to PPT" function, which led to the PPT production process—

The document assistant quickly combined the content of the document, listing a detailed outline in less than 30 seconds, and then produced a 14-page PPT in 40 seconds:

To verify the objectivity of the PPT content, we also uploaded a document we prepared— "Introduction to the Education of Love." From uploading the document to the completion of the PPT production, it took about 40 seconds.

In this round of testing, we analyzed the content of the PPT. Because the theme was related to "love," the corresponding images were matched for the related content. In terms of content, not only was the content of the original document sorted, summarized, and extended, but it also made certain extensions.

After multiple rounds of testing, Baidu Wenku has shortened the time for PPT generation from days to seconds. In other words, Baidu Wenku can complete a day's worth of work for office workers in less than a minute.

Comparatively, if a PPT were to be manually created, the steps would be roughly as follows:

First, clarify the presentation theme and objectives, comprehensively sort out the content of the copy, collect the necessary materials or resources for production, then design the style and layout, and finally produce a high-quality presentation, which usually takes several hours to complete—this is only an ideal assumption for experienced individuals. If it were a beginner creating a PPT, it would probably take "days."

Only when the time spent using the tool is reduced, will the time for creativity and imagination increase.

In addition, during the evaluation process, we found that Baidu Wenku's intelligent PPT generation can be achieved not only by uploading documents but also by using natural language. For example, if you say a sentence or enter a brief and concise text, the same effect can be achieved.

We entered the demand "generate a PPT on the historical development of the Star Wars movie." First, it will help you write an outline, and then generate a PPT based on the outline content. It is worth noting that after the PPT is completed, there are also functions to modify the theme color, modify the font of the entire text, and the color of the entire text. In addition, in terms of image selection, Baidu Wenku is also very particular, selecting images related to Star Wars.

In other words, Baidu Wenku not only helps you automatically generate PPT but also ensures the aesthetics, completeness, and operability of the PPT.

However, the capabilities of Baidu Wenku go far beyond this. It has also put a lot of effort into document chart generation. Even for complex statistical data, Baidu Wenku can handle it. Below is our practical process:

Step 1: We uploaded a document on "The current situation and future development trends of China's intelligent automobile industry in 2023";

Step 2: Select the extremely complex and tedious data that needs to be analyzed, such as "In 2022, the Chinese intelligent automobile market had a total of 119 financing and investment events, with a total financing and investment amount of 62.932 billion yuan. From January 1 to July 24, 2023, the Chinese intelligent automobile market had a total of 47 financing and investment events, with a total financing and investment amount of 15.295 billion yuan. Looking at the distribution of investment rounds, strategic investment accounted for 25.5%, A-round accounted for 23.4%, C-round accounted for 8.5%, angel round accounted for 6.4%, A+ round accounted for 6.4%, B-round accounted for 6.4%, C+ round accounted for 6.4%, Pre-A round accounted for 4.3%, B+ round accounted for 4.3%, seed round accounted for 2.1%, D-round accounted for 2.1%, E-round accounted for 2.1%, G-round accounted for 2.1%, the continued prosperity of the capital market has driven the entire industry chain to benefit";

Step 3: Enter "generate chart" in the dialog box, and in the end, a complete chart can be obtained, and the entire process takes only about 30 seconds.

In addition, at the end of the chart, the document assistant will also provide suggestions based on the generated data in terms of "conclusions, insights, recommendations," and so on.

Without AI tools, the steps for creating a complete chart would be roughly as follows: First, open the traditional Excel spreadsheet, then categorize and fill in various data in the spreadsheet, and finally select and click on the menu bar to "insert bar charts, line charts, pie charts, etc." in order to obtain a chart.

Next is the PPT generation with charts.

For example, when generating a "Financial Analysis PPT for Johnson & Johnson in 2022," the results were quite ideal, involving relevant data charts and analysis of total asset data, operating cash flow data, and so on.

In addition to Word and PPT, Baidu Wenku has also updated its PDF intelligent summarization and Q&A functions. For example, when searching for a "PDF on domestic large model evaluation standards" in Baidu Wenku, it can quickly summarize the content of this PDF for you and also engage in multi-round conversations, such as "How to improve model quality standards and testing methods?" The document assistant will provide corresponding answers based on the document content.

Baidu Wenku PDF intelligent summarization, multi-round conversations

It is worth noting that not only on the PC side, but Baidu Wenku has also been restructured on the mobile side. On the mobile side, you only need to click the blue button and select "AI Create PPT." It is worth emphasizing that the mobile side fully supports voice input, meaning that you can generate a PPT that meets your needs just by speaking. This also means that Baidu Wenku has overcome the limitations of time and space, enabling office workers to work anytime, anywhere, and even to avoid overtime and leave work early.

After the rise of large-scale models this year, Baidu has mentioned multiple times that it intends to "reconstruct all products" using large-scale models. Currently, large-scale models have been fully reconstructed in various products such as search and recommendation, and Baidu Wenku is also a key player in Baidu's competition in the intelligent office field.

Data shows that Baidu Wenku's new AI functions have accumulated over tens of millions of users and nearly a hundred million uses; over 500,000 PPTs have been generated, and over tens of millions of documents have been generated.

Baidu Wenku's Ambition

This year, the popularity of large-scale models is unprecedented. However, whether it's the battle of hundreds of models or the competition of thousands of models, all AI companies will ultimately face the same question: whether to choose a platform-type general AI tool or a customized enterprise solution?

For a startup company, this proposition needs to be carefully considered from the beginning of entrepreneurship. But from Baidu's perspective, as a company that has been laying out artificial intelligence for ten years and investing heavily in technological innovation, walking on two legs is a natural choice, a result of market competition, and a result of technological accumulation.

Take Baidu Wenku as an example. Why has Baidu Wenku gained the favor of many users in a short period of time?

The biggest reason is: large-scale models.

Since the official announcement of WENXIN on March 16, it has achieved rapid iteration within six months. There will be significant improvements in content understanding, memory, and generation capabilities, providing a solid technical foundation for the innovation of application-level products like Baidu Wenku.

Take Word to PPT as an example. Baidu Wenku needs to start calling various capabilities of WENXIN from the moment it receives a document. In this process, it needs to leverage its content understanding capabilities in the early stage to summarize the PPT outline based on the document content. In the PPT generation stage, it needs to call on memory and AI generation capabilities to ensure the completeness and accuracy of the content.

This is often much more difficult than generating a PPT from a single sentence. First, in the process of generating a PPT from Word, users provide very clear requirements and expectations. Second, in terms of understanding, the difficulty of parsing long text in Word is much greater than parsing a single sentence. After parsing, it is necessary to refine the content, create an outline, and fill in the PPT. Most importantly, the entire process needs to ensure the accuracy of content, images, and so on.

From the functionality of generating a PPT from a single sentence to the current functionality of generating a PPT from Word, Baidu Wenku only took about a week, relying on its long-term accumulation in AI, and WENXIN played an immeasurable role in this process.

On the other hand, Baidu Wenku is also constantly feeding back to the large-scale model. It is well known that to improve the capabilities of a large-scale model, it is necessary to continuously feed it high-quality data. Founded in 2009, Baidu Wenku has accumulated 1.2 billion documents over more than ten years, providing a large amount of high-quality professional document data, which has helped improve the training effectiveness of WENXIN, thus forming a positive flywheel effect.

From this perspective, Baidu Wenku and WENXIN are more like hands, feet, and the brain, mutually complementing each other. Without the control of the large-scale model, Baidu Wenku's hands and feet would find it difficult to complete various complex tasks. If it weren't for Baidu Wenku's hands and feet, the brain of WENXIN would have no way to unleash its value.

Since March of this year, based on the foundation of WENXIN, from the initial text document processing to the current intelligent PPT generation, intelligent editing, summarization of lengthy documents, intelligent summarization and Q&A for PDFs, Baidu Wenku has been continuously iterating and innovating its capabilities.

Currently, Baidu Wenku has become the industry's first all-in-one intelligent document platform, capable of handling document creation, complex document processing, and a series of intelligent editing and management tasks.

For users, whether individuals or enterprises, they can quickly generate documents through a new document creation mode, improve work efficiency and creativity. For enterprises, they may also be able to customize and develop specific document templates according to their needs to meet diverse and differentiated requirements. For individuals, there is no need to switch between multiple platforms for document creation and output, which improves work efficiency.

More importantly, Baidu Wenku, empowered by large-scale models, has greatly reduced the barrier to document creation for users and opened up a new paradigm in the traditional office field. From frequent clicking of the mouse and typing on the keyboard to natural language interaction, users only need to input the corresponding keywords or speak to generate the corresponding documents, greatly improving content productivity and quality.

As time evolves and technology advances, the future of Baidu Wenku is full of imagination.

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。