AI Implementation ultimately depends on how to solve user pain points and create user value.

Author | Lian Ran
Editor | Jing Yu

Not long ago, according to Foreign media reported that Dennis Austin, the inventor of PowerPoint, died at home. Netizens have different opinions on this news. Some people think that PPT is a great invention, while others think that PPT wastes a lot of human time.

Thousands of words come together In a word, "May there be no PPT in heaven."

After all , which migrant worker has not suffered the fate of being tortured by PPT?

To make a To present a workable PPT, you must first search for information, organize the materials, and write a speech, then create a new document, and then manually input the content one by one, and then adjust the page layout of each document. Polishing, after a set of procedures, most of the day is indispensable.

It is said that AI will liberate Improve workers' productivity, but there seems to be no AI application that can solve the PPT troubles for workers?

The good news is, Now there is really such a product coming forward.

With the end of August With the opening of Baidu Wenxin Yiyan large model, Baidu Wenku, which is connected to Wenxin large model, has begun to launch new functions one after another. Generate PPT from one sentence, directly generate PPT from Word document, or even summarize the key points of a 10,000-word work document< /span>, all directly help modern workers save a lot of time and energy to truly put their On to more meaningful work.

Geek Park Recent After in-depth experience of these new functions of Baidu Wenku, I have to say that some of the latter's performances even give people quite a surprise. Data shows that after Baidu Wenku launched its AI capabilities, it has experienced nearly 10 million users, generated more than 300,000 PPTs, and over 10 million documents.

In Wen Xinyi With the blessing of the capabilities of Yanda Model,Baidu Wenku's "one-stop smart document" platform dream , it really doesn’t look that far.

01 PPT, just do it

Baidu Library page layout is simple. Click the AI ​​icon on the lower right to enter the exclusive document assistant with built-in assistance The instructions make it easy to get started.

Baidu Wenku App Page screenshot｜Geek Park

Although the Li Jiaqi incident has The most popular search period has passed, but the discussion about Li Jiaqi is still continuing. Many people are "enthusiastic" in making public relations plans for Li Jiaqi. Let's see what plan Baidu Wenku will come up with?

You can see, In just 40 seconds, Baidu Wenku provided a document including measures, clarifications , improvement and commitment, how to interact with consumers and fans, how to cooperate and co-create, etc. A complete and comprehensive thousand-word public relations plan. This speed, let alone thinking, is already beyond the reach of an ordinary typist's typing speed of 50 to 80 words per minute.

Generate a PR The plan only takes 40 seconds, so what if we continue with the PPT?

A few days ago, The joint coffee "Sauce Latte" between Luckin Coffee and Moutai has been widely spread. Why not let Wenku make an autumn drink plan and enter the sentence "You are a coffee brand operations manager and need to create an autumn beverage plan" After the instruction, Wenku quickly gave a list of elements including product planning, marketing strategy, production technology and material procurement, service and personnel training, budget and schedule. Complete beverage plan PPT:

Baidu Library　Generate PPT in one sentence"｜Geek Park

PPT generated from It seems that not only is the structure perfect, but also important details and indicators such as turnover and average passenger flow are also given. It can be said that it is so detailed that it is a bit exaggerated. Just change some specific figures and hand them over directly to Party A. It is not impossible.

Generate PPT in one sentence It has greatly liberated productivity. On September 14, Baidu Wenku web version launched the Word to PPT function again.

In order to test the library in For the accuracy of content understanding and matching, Geek Park selected a previously published beautiful financial report, converted it into word and uploaded it to Baidu Library. In less than 30 seconds, the logic of the manuscript was given, and then it took about one minute to produce a document with accurate content, accurate illustrations, and clear organization. PPT.

Whether Baidu Library is used In two minutes, at least half a day's work of a laborer was completed.

This is how I do PPT Come on, summarizing should be a piece of cake. The author has never had time to read the famous feature-length masterpiece "Ulysses". On the one hand, there is not enough time, and on the other hand, 1.05 million words is a bit prohibitive for working people. Why not let Wenku summarize it first:

Baidu Library 1 Key summary of 10,000-word document｜Geek Park

In addition, Baidu Library It can also help users quickly analyze the key points of the document. At the same time, intelligent editing functions such as document polishing, abbreviation, and tone change can also improve the level of user-generated content.

In data statistical analysis In this field, the library has also achieved certain technological breakthroughs.

Users only need to When reading a document, simply select the data paragraphs in the text, and the library assistant will automatically generate matching statistical charts.

Same as Hua Xizi Take the Li Jiaqi incident as an example. According to DoNews, "The price of Huaxizi's eyebrow pencil reaches 987.5 yuan per gram, which is higher than most beauty brands. For example, Shu Uemura's classic machete eyebrow pencil costs 230 yuan for 3.4 grams, with an average price of 67.6 yuan per gram. According to In Li Jiaqi's live broadcast room, buy one and get two refills for 79 yuan, and the price per gram has reached 329.2 yuan. Another domestic brand, Mao Geping's eyebrow pencil, was found to be 1,222 yuan per gram." Intercept this text and enter it into the document Assistant to create forms:

Baidu Wenku Text Generate form｜Geek Park

In PPT content generation Functionally, the library can also automatically generate and insert related statistical charts based on the text theme and content. This greatly eases the user's difficulty in data analysis.

Even more complex Some data statistics are not difficult for Baidu Library.

With new energy vehicles Taking the statistics of the past two years as an example, "According to China-Singapore Jingwei report, according to data from the National Bureau of Statistics and China Automobile Association, China's automobiles have grown rapidly in the past two years. In September 2020, China's cumulative production and sales of new energy vehicles exceeded 5 million vehicles; exceeded 10 million vehicles in February 2022; by July 2023, the cumulative production and sales of new energy vehicles in China will exceed 20 million vehicles. The second 10 million vehicles took only 16 months, which is very fast. It is expected that the third million should be achieved by July 2024, and the continued high growth of new energy vehicles will become a new driving force for China's economic growth." After intercepting this text and inputting it, the document assistant quickly produced a chart that meets the needs:

Baidu Wenku Text Generate charts｜Geek Park

The above functional breakthroughs, All of them are inseparable from the library's technical accumulation in the fields of text understanding and data analysis. Baidu Wenku can accurately capture and analyze the data information in the text, and then convert it into suitable chart presentation. Compared with traditional manual processes, the efficiency is significantly improved.

Currently, library products It is still undergoing continuous upgrades and has achieved deep synchronous reconstruction of the web page and mobile APP. This provides users with richer and more flexible usage scenarios: users can operate documents via voice or text on the web page, generate PPT, outlines, etc., and can also implement the above functions through voice editing on the APP side; between different devices, Users can seamlessly switch usage scenarios; the mobile terminal supports a variety of voice interaction commands, which reduces the difficulty of use. Users can process documents simply through voice.

In other words, Even if workers don't have a computer with them, they can still process and modify documents at any time, and they no longer have to worry about delaying work.

02 AI, how to truly unleash productivity?

The sudden "explosion" of Baidu Wenku's capabilities is actually inseparable from the long-term thinking and experimentation behind the team's efforts to access large AI models and how products can meet customer needs.

As the first batch One of the products that is connected to the Wenxin Yiyan Big Model. In the early days of integration in March, the Baidu Wenku team believed that the Big Model has great potential in content acquisition and can empower platforms such as Wenku.

But in the subsequent During the user survey, the Wenku team found that the real pain point of users is not just access to content; From creation to acquisition to final sharing, there are some problems in the entire link. For example, if you want to switch between different platforms, the creative threshold is relatively high.

For example, in the last When starting to write a document, not all users will open an editing software to create a new document. Maybe 50% of users will first come to the library to find a reference content to get inspiration, and then write the document. However, When writing a document, you may have to use other office software to edit and modify it. In the end, even if the word document is formed, a series of formatting styles must be beautified.”

And further beautification At the document stage, some of the more advanced editing functions on some office software have barriers to use for users, and secondly, operating them step by step will consume a lot of time, and if you want Word to fully function, such as as a lecture notes Content generation PPT will involve more cross-platform operations, which is somewhat miserable for users.

True Starting from these pain points of users, the team reconstructed the positioning of the library into a one-stop intelligent document platform. The reason why this positioning can be implemented is inseparable from the improvement of the large model capabilities behind it.

Previously, due to natural Due to the technical limitations of language processing NLP, it is difficult to achieve in-depth understanding and processing of documents. Today's Wenxin The one-word large model has stronger content understanding, memory and generation capabilities, Content production can already be guided through simple voice or text dialogue, which provides technical possibilities for positioning and adjustment of libraries.

During the entire transformation after the model, the Wenku team has always focused on the original intention of helping users solve pain points throughout the entire process, starting with text document processing, and then further launching AI capabilities such as PPT production.

AI production PPT Yes One of the most important functions after the reconstruction of Baidu Library,< /strong>In the Wenku AI new function dialogue interface, users only need to enter a sentence topic or simply "move their mouth" to send a voice command, you can quickly generate PPT within 30 seconds. The document generation speed, content accuracy, content richness, and illustration accuracy are all eye-catching.

Compared to one sentence Generate PPT,Baidu Wenku has recently launched a more powerful word ppt generation function span>.

Geek Park Curious The question is, why was the function of converting word to PPT introduced later?

Generally speaking, We may subconsciously think that it is more difficult to convert a sentence into a ppt because the information given is limited; while it is easier to convert a word into a ppt because the word will contain more information, but the opposite is true.

In this regard, Baidu Wang Ying, vice president, head of interactive entertainment and vertical platforms, told Geek Park the truth behind it.

Generate PPT in one sentence Compared with word generation of PPT, the room for play is different for large models. When generating PPT in one sentence, large models have sufficient freedom, and the user's expectations are relatively low, so it is easy to give what is expected.

But word generates PPT The difficulty lies in:

• First , in the process of generating PPT in word, users will give very clear needs and expectations, so the space for play is compressed;
• Secondly, in terms of understanding, it is far more difficult to understand and parse long word texts. It is much more difficult than parsing a sentence. After parsing, you have to refine the content, form an outline, and fill in PPT;
• In addition, in this set of processes, the accuracy of content matching must be ensured.

These all make it much more difficult to generate PPT in word than to generate ppt in one sentence. At present, even in foreign countries, there are no mature solutions in this field. However, Baidu Wenku developed and implemented the function of word to ppt only one week after launching a sentence to generate ppt.It relies on its long-term accumulation in AI technology.

In terms of text processing , the capabilities of the large model behind the library have been greatly improved, especially significant breakthroughs in context understanding, which can achieve precise semantic matching; in terms of images, the library not only has accumulated image search technology, but is also considering deeply integrating it with AI technology. In the future, it may be possible Under the premise of ensuring the protection of intellectual property rights, semantically matching images are automatically generated, which for users will mean further release of productivity.

Library The optimization of functions is closely related to the ability of large models. The Wenxin large model behind the library has a very fast iteration speed and a short model optimization cycle, which provides opportunities for product reconstruction work. On the one hand, in some R&D scenarios, the iteration speed of Wenxin's large model is significantly faster than the development iteration speed of the product side, which brings certain challenges and opportunities to the continuous optimization of the product side. On the other hand, Baidu Wenku has provided Wenxin Model with a large amount of high-quality professional document data. It helped improve the training effect of Wen Xinyiyan and formed a positive flywheel.

In short, with traditional Compared with documents, Baidu Wenku based on large models can now realize full-process intelligent services including intelligent editing of documents, intelligent summary of documents, and document generation PPT. It is a stop that meets various needs of document creation, lowers the threshold of creation, and improves creation efficiency. platform.

Now, based on Baidu The library accumulated with powerful AI technology is leading the document processing industry into a new era of "create by speaking", providing users with a new document creation experience.

For users , the library transformation has very effectively solved their various pain points in the document processing workflow, and realized a one-stop full-process document processing solution. Users no longer need to switch between multiple platforms for creation, output, etc., which improves work efficiency. In addition, Wenku’s AI capabilities have significantly lowered the threshold for users to create, enabling “creation by speaking.”

Wang Ying told Ji Ke Park, after the new function was launched, Baidu Wenku’s payment rate and users’ willingness to pay have been improved. In the future, Baidu Wenku will still rapidly iterate products, and new AI functions will also be launched one after another.

Essentially , the initial usage pattern of users in Baidu Wenku is actually people looking for documents/information, that is, people looking for information. Now, with the access to Wenxin Yiyan's large model capabilities, Wenku is using a language interactive interface to achieve a state where machines can find "information" and beautify it before delivering it to users. This can be seen from generating PPT from a sentence, to converting Word to PPT.

To further complete To meet user needs and improve the matching of image and text information, on the one hand, it relies on the continuous evolution of large model capabilities, and on the other hand, it depends on the library team, how to use "prompt words" to find people in a more convenient and concise way, which not only helps Users can complete their work more perfectly, expand their thinking boundaries, and grow at work.

And if you can At this point, Baidu Wenku's product is not only a useful To C product, but may even become an office-level application equally favored by B-end users.

　No matter whether they understand large models or not, users are willing to pay for products that can save time and create value for themselves< /span>." The attitude of consumers just proves Baidu Wenku's "one-stop intelligent document platform" The transformation was correct and initially successful.