Reporter | He Jipai
On November 12, Robin Lee opened the Baidu World Conference with a question.
"The super application that everyone expected has not yet appeared. Someone asked whether this global big model craze in the past 24 months was a new technological revolution or a new round of foam?"
The annual World Congress is Baidu's most important strategic and product launch event, and this year's theme is set as "Applications Come". In the one hour speech, Robin Lee mentioned the word "application" 42 times.
Robin Lee said that he would create millions of "super useful" applications. At the meeting, he showed a lot of products and scenes.
Among them, the "Free Canvas" that he spent a considerable amount of space introducing is a brand new tool based intelligent agent jointly developed by Baidu Wenku and Baidu Netdisk.
This new intelligent productivity platform is associated with a billion level user pool, two flagship products, and a business foundation.
Universal whiteboard
Baidu defines free canvas as a groundbreaking and fully modal new species of AI creation.
Combining Wenku and online storage, we need to run from the starting point of content production to the end point of content consumption
Wang Ying, Vice President of Baidu and Head of the Library and Cloud Storage Business Units, told 21CRBR reporters that free canvas is currently the top priority of our work.
As its name suggests, users can quickly generate multimodal content by dragging and dropping materials on a canvas like interface.
Robin Lee mentioned that users are free to input, edit, create and share.
Bunko and Netdisk bridge the gap between public and private domains, allowing users to access conference audio from Netdisk or documents included in Bunko.
This tool supports over 100 mainstream formats, and users can directly select content or use annotations to tell AI how to utilize materials.
Finally, users can share, view, edit, or store on their personal cloud drive with just one click, which is compatible with external sharing platforms such as Xiaohongshu and WeChat Moments.
In other words, the user is the commander, giving instructions to the AI to deploy and deploy materials, indicating where to hit.
Robin Lee called it a "universal whiteboard" and proudly said that what he did was not futures, but immediately available cash. The product has been tested and reserved for the public, and the number of reservations on the first day of opening has reached 200000.
What is the inspiration for the free canvas? What aspects of content production and creation will it serve?
Wang Ying told 21CRBR that as early as March last year, the Wenku team had an idea to start from the beginning of content creation and "tinker" with a cross modal large model container.
The underlying thinking is that with AI capabilities, there are currently two marginal products. One is editors like WPS Office, which are precise but have high creative costs; The second is an AI editor, which has a low threshold but is difficult to accurately generate content.
By coupling these two points with a new product form, her thinking gradually became clear when she reported upwards in February and March this year.
The Wang Ying team has made it clear that they need to consider the three aspects of input, processing, and output, breaking down the boundaries of format, modality, and public/private domains. One link, one drag, one loop, one-stop processing, users can operate smoothly enough.
Users do not use tools for the sake of using them, but to meet their needs. When hungry, they need to eat; when thirsty, they need to drink water; and when writing reports, they need to find the core starting point. Tools serve their needs
Wang Ying emphasized that the perception of users should not be reversed. They don't need so many tools. It's like going out alone, not knowing how to bring a bunch of axes and knives, definitely hoping for All in One
Old tree sprouts
The emergence of the free canvas also means that the two trumps in Robin Lee's hand have opened the integration.
Betting on AI is original, and the library is the main front. It is one of the products that Robin Lee has called for many times and "AI reconstruction is the most thorough".
It's natural for Wenku and Netdisk to be together, "Wang Ying believes. These two products are essentially different facades for communicating with users, with the same underlying AI capabilities.
She already has a complete methodology for Wenku.
After taking over the library two years ago, she began to make drastic changes to it.
Since August last year, Wenku has released over a hundred multimodal AI capabilities, including intelligent PPT generation, intelligent drawing books, and more. Among them, the intelligent PPT market has captured 80% of the market share.
At present, Wenku's AI MAU has exceeded 70 million, reaching the top tier in China. As of October, the cumulative number of AI users has exceeded 230 million.
Adhering to end-to-end approach and MoE (Mixed Expert Model), coupled with engineering capabilities to make up for deficiencies, we have not taken the wrong path since day one, "said Wang Ying.
It is not an easy task for a team of only a few hundred people to revitalize a 15-year-old old product in a short period of time and achieve a positive commercial cycle.
In the past, technology was just about providing requirements to products, but now it's different. What big models can do depends on inspiration, not something you can figure out right away
Wang Ying told 21CRBR that technology needs to be coordinated with products, working together to find solutions and moving forward in both directions.
The major functional iterations of the library are conducted on a monthly or even weekly basis.
The technological leadership is limited to six months at most. The team needs to quickly build its own product advantages within this six months, widen the gap, and in turn, through technological iteration, obtain more data and feedback to give the model a greater advantage
Wang Ying thought very clearly that only by repeating this cycle can we make it thicker and take it to the next level.
Integrating Value
Not long ago, Baidu held a review meeting internally, and Robin Lee took the stage to speak and share many thoughts.
According to his judgment, strategically speaking, Baidu needs to acquire but also give up, seeking both speed and caution. At present, there is still too much value in AI that has not been unleashed.
As of the beginning of November, the daily average adjustment amount of Baidu Wenxin's big model exceeded 1.5 billion times, an increase of about 30 times compared with a year ago, far above the benchmark of Robin Lee's view that the big model "can become".
The so-called 'product integration' is no longer a literal meaning. Without big models, libraries and cloud storage will only have simple integration of functional items, and there will be no change in product boundaries, "said Wang Ying.
Empowered by a large model, the library has transformed from a retrieval platform to a creative platform, and the cloud storage has transformed from a storage warehouse to a consumable one. These two products leverage their original advantages while simultaneously integrating and restructuring, jointly supporting every content production and consumption action of users.
Many users say that the things written by Wenku using AI do not have an 'AI flavor', and this is just our first step. After introducing private data from cloud storage, we hope that users' AI creations can truly integrate their thoughts and express their personal wishes
Wang Ying revealed to 21CRBR that the integration of Wenku and online storage will be figured out in the first month, and a plan will be formed in the second month. New releases will be made gradually this month, and by the end of the year, the foundation construction will be basically completed.
Obviously, the free canvas is just the first step taken. However, whether this step can be taken accurately is extremely crucial.
Wang Ying stated that using this basic feature to connect the two products can bring benefits to users, as well as increase payment and retention rates.
To put it bluntly, if users feel that the product is worth using, their loyalty increases, and they are more likely to accept new forms of charging, further driving commercialization.
In terms of development indicators, the team will comprehensively focus on user scale, revenue, and profit to assess the health of business growth.
More ideas are on the way to queuing up for implementation.
Our aim is for models to think like humans and products to work like humans. As long as it is in this direction, it is what the team needs to do, "said Wang Ying.