Why has Feishu become the common choice among domestic large model unicorns?

中国 Question diagram | Visual China has been the most watched track in the field of Chinese technology in the field of science and technology in China in the past year. Especially after entering 2024, the popularity of the entire industry can only be described as "

hot

": In terms of financing, the craze from 2023 to the present has gradually reached its peak, and the valuations of many top

startup companies have soared to billions of dollars. ; At the business level, emerging startups and Internet giants are engaged in fierce competition around the capabilities and prices of basic large models, as well as the market share determined by the first two. More intense than the price war is the collision of business routes. Some companies adhere to the toC business model and focus on providing products and services directly to end users. Others are turning to the toB market to provide customized solutions and technical support for other companies. The remaining companies have begun to establish cooperative relationships with Internet giants, seeking broader development space and deeper resource integration. The fierce collision at multiple levels from technology to application to commercialization has made the entire large model industry full of gunpowder. But in this environment, almost all large-scale startups have agreed on one thing - tacitly choosing Feishu as their collaboration tool without official promotion. Recently, the author went to Guangzhou and saw a Feishu customer advertisement in the arrival hall of Baiyun Airport. Nine of the most popular ExcellentAI innovation companies were listed, including the large model "Qilin": MiniMax, Yue The Dark Side, Wisdom Spectrum AI, Zero One Thing, Baichuan Intelligence, and Stepping Stars. “Choosing Feishu was a natural choice, and I didn’t go through a complicated process. It may be because the founding teams of large model companies are generally young and are more adaptable to collaboration tools like Feishu. At the same time, there are too many Internet companies in these companies. "My classmates are used to Feishu's experience, so this has become an inevitable problem for them," said a classmate who switched jobs from the Internet to a large model company. Using Feishu seems to be a choice that does not require much thinking for large model companies. The “challenges” behind the glamorous appearance of big model leaders Before analyzing Feishu’s attraction to big model head start-ups, we need to understand the latter’s actual needs. If you want to use three words to summarize the working status of the large model industry this year, it would be "fast, fast, fast."

FlagEval Large Model Evaluation Capability List (Objective Evaluation)

In order not to fall behind in the field of large models, thousands or even tens of thousands of GPUs are often used for training, and they are performed at an increasingly faster frequency renew. Regarding basic large models, we must not only catch up with and surpass advanced foreign models in terms of performance, but also compete fiercely with other domestic competitors in terms of application implementation. Many companies have also entered the commercialization stage in advance, trying to open up the market first and find customers in the toC and toB markets to realize commercial value. At the same time, the arduous task of promoting "scientific research, application exploration and commercialization" has allowed large-scale startups that have been established for only 1 to 3 years to enter an extremely rapid expansion period, with larger teams and more projects. , more complex commercialization attempts have directly caused a surge in collaboration needs. Let’s talk about the “people” challenge first. There are many “super individuals” (employees with extremely outstanding professional abilities) in the R&D teams of large model startups. Take Dark Side of the Moon, one of the big model "unicorns", as an example. Its early core team were all born in the 1990s and came from Tsinghua University. Just three months after its establishment, it was already rated as the most likely to become the "Chinese version of OpenAI" One of the candidates. The entire large model industry's reliance on "super individuals" poses new challenges to companies from early recruitment to later management incentives. Faced with these "super individual" employees, it is impossible to evaluate talents based solely on results as a single dimension. Secondly, for start-up companies that are highly young and full of "super individuals", traditional methods of attendance and solidification processes also appear to be incompatible. Large-model startups urgently need a more flexible and innovative way to coordinate and improve "people" efficiency. From a management perspective, the large model industry is still in a stage of rapid development and requires continuous exploration and practice. This process requires the joint efforts of all functional departments of the company to "cross the river by feeling the stones." Rather than executing the boss's instructions, it is more important to give full play to the "creativity" and "initiative" of each employee and respond quickly to the market and technology. changes to push the project forward.

If the production and research of large models is already very difficult, then the commercialization challenges that large model startups need to face today can be described as "even more difficult." How to continuously improve capabilities at the product R&D level, how to plan new functions that individual users really need, how to work with enterprises to explore the application potential of large models, and how to adjust one's own development strategy based on real feedback from the market; this series of new questions The emergence of the Internet has made the overall challenges faced by large-model startups more and more "complex".

Take Zhipu as an example. In early 2024, it officially announced that it had 1,000+ large-scale model applications and had carried out in-depth "co-creation" with 200+ companies. This "co-creation" process is actually a process in which Zhipu and customers jointly "excavate" the needs and application prospects of large models in various industries. This has resulted in a large number of front-line relationships between Zhipu and customers, sales and development. Communication and collaboration needs with management.

Obviously, at this critical time when large model technology is gradually maturing and more and more attention is paid to application implementation, large model startups are pursuing full-chain innovation from theory, method, technology, product to market, plus the industry itself The uncertainty brought about by rapid development and the huge challenge of pushing a new technology to all walks of life have created a strong sense of urgency lingering in the hearts of large model startups.

Why is Feishu the unanimous choice?

Why Feishu? The author speculates from the characteristics of large model companies, which can be summarized into three points:

1. Rapid iteration of tools drives the iteration of the organization: In the process of rapid expansion, large model companies are in urgent need of mature methodologies and tools to improve efficiency and even iterate internal operations. mechanism. It just so happens that Feishu has accumulated the methodologies of thousands of advanced enterprises in the past few years of service, and has brought positive changes to the organization through rapid tool iteration.
2. contextnotcontrol: In order to match the rapidly changing business, building an agile organization is the goal of emerging companies such as large models. Allinone's model adds to the rapid flow of information and achieves efficient organizational synergy. So-called contextnotcontrol, today’s large model companies are goal-driven and more like distributed organizations, stimulating everyone’s potential.
3. High product flexibility and high openness: Feishu’s multi-dimensional tables, open platform capabilities, etc., have strong room for openness and creativity. This feature exactly matches the characteristics of large model companies. This type of company has a high concentration of production and research, is good at using tools, and has built many applications in-house through Feishu's open capabilities. It is said that Dark Side of the Moon has used the Feishu integration platform to build dozens of enterprise-owned applications, integrating them with other functions on Feishu to jointly improve efficiency.

As an example of information flow, ordinary IM and collaboration tools can realize the flow of information within their respective product scopes, while Feishu's system can realize multiple different functional products, specific projects and approvals, processes and knowledge accumulation. Open up the space. This ability to accelerate information transmission is more effective in complex scenes. For example, traditional CRM software is good at analyzing customers' historical transaction data, but does not have powerful communication and office functions. They may become information islands during the long-term use of enterprises, resulting in inconvenient use and low process efficiency. Feishu has enhanced the import and interaction capabilities of sales data, making data management more efficient. With more macro and clear data presentation and insight capabilities, it helps sales staff make timely decisions and effective follow-up, improving efficiency and performance.

In addition to accelerating various collaboration processes within the company, Feishu's "information-centric architecture" feature also changes the dilemma of business experience following individuals and being unable to use individual wisdom to drive group development. Just like the delivery project we mentioned above, the coreinformation of the entire project has been collected into the knowledge base and connected with a page tree structure that is much clearer than the folder system. Even new employees who have not handled this project can learn and summarize the implementation process from historical documents afterwards. By overall improving the efficiency of knowledge dissemination within the enterprise and transforming information and experience into a systematic knowledge base, large model startups can efficiently circulate and utilize this knowledge. This systematic process not only promotes thinking collision and innovation among employees, but also feeds these innovative results back to the company itself, becoming an important driving force for corporate development. If choosing Feishu at first was a judgment based on "feeling", then considering key issues such as overseas travel and safety, it seems that Feishu has become a must-have choice for large model companies. On the one hand, in the long run, almost all large model companies have a need to go overseas. Considering the various collaborative relationships across countries and regions and strict cross-border compliance requirements, it seems that only Feishu, which has served many companies to achieve global layout, can satisfy. On the other hand, extreme security requirements also exclude some collaborative software options for large model companies. Feishu is currently extremely meticulous in terms of terminal security, data confidentiality label protection, and data leakage prevention. One of the reasons why many large-scale startups like to use Feishu is that it has very "detailed" management permissions for various information. It can even limit copying and copy creation scenarios on demand, and the right to share information with others can be determined based on the scenario. There are differences and so on. Only big model companies? In fact, if you broaden your horizons, the advanced companies that choose to use Feishu are far more than just startups that focus on large models. From autonomous driving to embodied intelligence, to the upstream and downstream of AI innovative companies, Feishu seems to have been the common choice of those in the "innovationindustry" at the forefront of this society. Including this website itself, we are also using Feishu. Feishu's advanced collaboration capabilities not only attract companies, but also individual users who represent the most advanced productivity. Just like those "old users" of Feishu in large-scale startups, they are expanding Feishu's influence and territory outward like a "torch passing". The most influential open source AI knowledge base project "WaytoAGI (Road to General Artificial Intelligence)" in China built on Feishu Knowledge Base is the best example.

WaytoAGI Figure Since its establishment in April 2023, "WaytoAGI (Road to General Artificial Intelligence)" has built an extremely large knowledge base system, covering various technology introductions, AI industry news analysis, and AI application practical operations. It has received more than 1.5 million views, triggered tens of thousands of exchanges between users, and allowed at least hundreds of thousands of Feishu users to have a deeper understanding of AI and make their own attempts to apply AI. The fact that a group of people with lofty ideals on the Internet can achieve such results not only proves the excellence of Feishu's entire collaboration system, but also proves that "people who understand large models are likely to be using Feishu." In a company that already has a huge number of "fans" Under the premise, Feishu has not stopped its own evolution. While serving customers of large-scale startups, Feishu has also begun to accelerate the introduction of AI capabilities into its products. As early as the end of last year, Feishu had embedded "Feishu Intelligent Partner" in its entire set of products, which can work with users in business scenarios such as content creation, content summary, data analysis, scenario construction and system construction. Enterprises can even choose suitable underlying large models based on business needs, such as Baichuan Intelligence, MiniMax, and Zhipu AI.

Take the daily work summary, the most common

job for employees, as an example. It can be directly generated by the intelligent partner. Some PDF files with more complex content and a huge number of words can be sent directly to the smart partner, and the core content can be extracted and summarized in the form of questions and answers. You can even describe your needs and uses in words, and let the smart partner build the required multi-dimensional table system. From excellentcustomers, to seniorusers, to the upgrade of all AI-related capabilities, perhaps we can say: in China, everything related to AI is becoming more and more closely connected with Feishu .

The above is the detailed content of Why has Feishu become the common choice among domestic large model unicorns?. For more information, please follow other related articles on the PHP Chinese website!