On the eve of the big model App Store moment, DingTalk, which has 700 million users, wants to “kill” the app

In 2023, when large models are roaring by, there are already more than 200 large models on the market.

According to statistics, in the past year, the world's 50 most popular AI tools have been visited a total of 24 billion times, and the traffic has increased more than 10 times. But in fact, this is not the peak of the AI ​​explosion. AI tools have a great impact on most people. It may start to take shape this year.

Recently, many developers are looking forward to the launch of OpenAI's GPT Store. In fact, they are looking forward to AI's "App Store moment". It belongs to the AI ​​native application ecosystem, and the ideas for application use and development methods will be different.

Unexpectedly, the first person to hand in this answer sheet in 2024 will be DingTalk, which has more than 700 million users.

DingTalk officially released the 7.5 major version update today. The blockbuster feature "AI Assistant" and the upcoming AI Assistant market are actually an exploration of the productization form of AI Agent.

Thus, the year of AI Agent explosion kicked off.

The birth of an AI super assistant

In November last year, Alibaba announced the first batch of four strategic-level innovative businesses, and DingTalk was among them. Alibaba Group CEO Wu Yongming once said that DingTalk has gained unprecedented imagination due to the arrival of the AI ​​era.

In DingTalk's definition of AI assistant, DingTalk AI assistant refers to an AI application that can interact with natural language and complete complex tasks based on the understanding of user and enterprise data and business.

The AI ​​assistant launched in DingTalk version 7.5 released today is actually the actual implementation of AI Agent. It can belong to either an individual or an enterprise. Each person and each enterprise can customize a personalized and exclusive super assistant.

The personal super assistant can help you open various function settings, including quickly opening functions such as screen mirroring, or setting up entrances such as incoming call ringtones, simplifying function paths and making product use easier; or it can help you summarize information, write work summaries, and write Documents and other work, or operations such as initiating meetings, sending schedules, creating to-dos, etc., to improve work efficiency.

Not long ago, after more than a year of development, the personal version of DingTalk was fully launched. It updated many AI assistants such as Luban Camera, Smart Portrait, Graffiti Painting, and Legal Assistant, and provided AI portraits, digital avatars, AI painting, and AI lawyers. Waiting for service. The on-site demonstration used the graffiti drawing function of DingTalk Personal Edition to create story scripts and picture books.

In the scenario of home-assisted homework, when you are in the family group @ AI assistant, let the AI ​​home assistant correct and grade the composition. It can automatically identify typos and good words and sentences in the composition, fully understand the structure of the article and give modification suggestions. . After completing the correction, it can also be polished with one click, making it easier to have a conversation with your child about homework.

For part-time workers, it can tell you a summary of unread messages, clear red dots of work matters with one click, create agents, and create schedules, such as discussing tomorrow's customer visit plan. Interestingly, the group document also supports quick reading, which automatically helps you summarize long articles sent by others. The main purpose is to save worry and be efficient.

At the press conference, DingTalk President Ye Jun repeatedly emphasized that reducing costs, increasing efficiency, and improving quality are an eternal topic for corporate development. This is the value provided by enterprise super assistants. One of the more prominent applications is intelligent question and answer AI.

DingTalk hopes to realize real value in inter-enterprise management, customer consultation or knowledge sharing among enterprise employees through intelligent question-and-answer AI.

Zhang Xiaolong, who plays the role of Dr. Wen, also came to the press conference and released the tenth-level scholar-human-machine challenge. Zhang Xiaolong and the on-site guests randomly asked three questions, allowing the well-known UP leader Yuan Ri'an of the "Tenth Level Zhen Huan Scholar" to compete with the intelligent question and answer AI of the "Tenth Level Zhen Huan Scholar" and DingTalk CEO Ye Jun on the same stage.

Among the three questions, UP leader Yuan Ri'an finally got two points, and DingTalk CEO Ye Jun barely got one point. However, the "Tenth Level Zhen Expert" intelligent question and answer AI did not overturn and reached three points. In answering questions such as "Zhen Huan is here" In questions such as "What are the titles in the palace? Can you answer them in Chinese and English at the same time?", the overall performance can be said to be remarkable.

In addition to intelligent question and answer AI, DingTalk also has more private domain AI operation tools. Whether it is AI-generated event posters, AI-generated lottery activities, or AI-generated group chat summaries, they can better help consolidate bloggers and fans, buy The emotional connection between home and seller.

The capabilities of enterprise super assistants are not limited to generation, but can also do more.

During the live demonstration, when smart travel AI is used, you can book air tickets and hotels near the destination with one click. You only need to enter text requirements, and DingTalk will help you complete it as much as possible.

The AI ​​Agent that was frequently mentioned at the press conference is actually an intelligent agent with the capabilities of perception, memory, planning and action.

This is the biggest difference between DingTalk AI Assistant and chatbots such as ChatGPT, because it already has cross-application task execution capabilities, and the AI ​​Assistant can be seamlessly integrated with DingTalk's rich third-party applications and enterprise-built applications.

It is worth mentioning that DingTalk also specifically demonstrated intelligent data AI, which can realize corporate travel analysis by coupling native application data, low-code application data, or customer-owned data.

If you are a boss or other management, you might as well ask the AI ​​assistant to count the number of business trips last month by department, easily know the details of business trips, make business management more transparent, and help you make better management decisions.

So how to create a dedicated AI assistant? Specifically, it can be divided into three steps:

  1. You need to customize an exclusive name and avatar for the AI ​​assistant;
  2. Select a specific knowledge base, document or upload file to train the AI ​​assistant;
  3. To configure the required AI capabilities for the AI ​​assistant, you can directly select the official capabilities or configure custom capabilities.

This customization capability includes enterprise self-built systems, third-party applications, DingTalk ecological applications, etc. It also includes configuring open APIs for major e-commerce, video, news, search and other platforms to achieve open connection capabilities.

Compared with other companies, DingTalk has rich scenarios, corporate accumulated knowledge and data; it also has clear customer needs. This is something that all basic large models do not have, and will also be the core differentiation of DingTalk's AI Agent.

Therefore, DingTalk plans to launch the AI ​​Assistant Market (AI Agent Store) in April 2024, with the goal of becoming a low-threshold, high-frequency and open AI assistant platform. Ye Jun believes that AI Agent has become the best entrance to AI applications at the moment, and 10 million AI super assistants will be created on DingTalk in the next three years.

This product will inevitably be compared with OpenAI's upcoming GPT Store in the future, but it represents two routes.

The best closed-source large models rely on their powerful capabilities to attract a large number of developers to build applications, while the best open-source large models build an AI application ecosystem in a more targeted manner based on the needs of a large number of C-side and B-side users accumulated in the past.

It seems that the story of iOS and Android is repeating itself. There is another detail in this story that is often ignored. The iPhone and Android were actually released in the same year, and the impact of the latter on the entire market, especially the significance of domestic manufacturers today, even exceeds The former.

The second half of the big model, the eve of the outbreak of AI Agent

One year after the explosion of large models, whether it is a general large model or an industry large model, whether it is closed source or open source, self-research or "shell", we are bound to face a problem: how to convert the computing power of large models into productivity. Start to move from the stage of spelling out parameters to the stage of spelling out applications.

Microsoft CTO Kevin Scott believes that what makes PCs, the Internet, and smartphones great are not the things that exist with the birth of the platforms, but the new things that are re-created by people based on these.

This point of view also applies to the large model market. Only AI native applications built on large models can bring the greatest value to everyone.

From the GPT Store released by OpenAI at the end of last year to the AI ​​super assistant released by DingTalk today, we can already vaguely see the main line of the AI ​​industry in 2024 – this will be the year of the explosion of AI Agents in various fields. Ordinary consumers and enterprises The needs of more specialized segments will be addressed in completely different ways.

The root cause of this change is actually a fundamental change in the underlying logic of human-computer interaction.

As can be seen in DingTalk version 7.5, the AI ​​super assistant is already the entrance to 30 high-frequency functions. Users interact through natural language conversations and get results directly, instead of finding entrances to different functions through application menus, which greatly shortens the level of interaction. , which is the biggest difference from the mainstream graphical user interface (GUI) from PC to mobile phone.

Based on natural language interaction, it is more intuitive and efficient. August de los Reye, chief UI design director of Microsoft, also regarded this interaction method as the future in 2008, but the conditions at the time were obviously immature.

When this interaction method becomes the mainstream UI in the future, it also means that the concept of application will be redefined. AI native applications and current apps are two completely different species.

DingTalk Al Assistant is an exploration of future application forms. At the same time, both individuals and enterprises can build their own AI applications according to their own needs, so that everyone can have an AI assistant like Jarvis, which is the greatest potential of this product form in the future.

When a person who doesn’t understand code can develop an AI assistant, then for enterprise-level users from all walks of life, just like the case mentioned above, the cost reduction and efficiency improvement may be exponential.

AI allows everyone to become a super individual, and it can also open up the two channels of an organization. As Ye Jun, CEO of DingTalk, said:

Doubling the cost is innovation, and reducing the cost tenfold is generational difference. DingTalk will drive the popularization of AI, lower the threshold for enterprises to become intelligent, and ensure that every enterprise will not miss out on the AI ​​era.

When Open AI's GPTs were released, one GPT was born almost every minute, and nearly 100,000 user-customized GPTs have been gathered on GPThunter alone.

And this is just the tip of the iceberg of the AI ​​native application ecosystem. There are still massive C-side and B-side demands that are in urgent need of more professional and personalized AI solutions.

For example, the AI ​​assistant released by DingTalk today combines the real usage needs of 700,000 companies. At a time when large model products at home and abroad are chasing GPT-3.5 and GPT-4, it is possible to gain insights into these real needs. It will become the advantage of domestic manufacturers in the second half of large models.

In fact, while we marvel at the rapid evolution of GPT again and again, domestic large models are no longer unmatched. Alibaba Cloud Tongyi Qianwen Qwen-72B has recently surpassed Llama2 to reach the top of the Hugging Face open source large model rankings, entering the first echelon of global large models.

Since DingTalk connected to Tongyi Qianwen's large model in April to start its AI drive, in August it opened its intelligent base (AI PaaS) to ecosystem partners and customers, and today it has launched its AI assistant ecosystem, the pace of its large model productization has been It’s as fast as Open AI and Microsoft.

On the eve of the explosion of AI native applications, when we discuss who will be the Android of the big model era and who will bring AIGC's "App Store moment", Chinese technology companies such as Alibaba are already strong competitors.

And we might actually be ready to create and use tools in entirely new ways.

Why does the next smartphone need to be a mobile phone, and why does the next super app need to be an app?

# Welcome to follow the official WeChat public account of aifaner: aifaner (WeChat ID: ifanr). More exciting content will be provided to you as soon as possible.

Ai Faner | Original link · View comments · Sina Weibo