DingTalk launches AI search: unlike Baidu and Secret Tower, it also collects domestic large-scale models of “Dragon Balls”

Yesterday, OpenAI announced the termination of API services to China.

Today, Tongyi, seven domestic large-scale model manufacturers, including MiniMax, Dark Side of the Moon, Zhipu AI, Zero One Thousand Things, Baichuan Intelligence, and Orion Star, gathered for a press conference.

There seems to be no shortage of drama in the AI ​​industry. Of course, the gathering of these big models is not to "siege Guangmingding", but to officially announce a cooperation with DingTalk to build the most open AI ecosystem in China.

In version 7.6 released by DingTalk today, DingTalk returns the choice of large models to users.

Users can switch AI large models according to their own needs. In addition to the default general meaning, the first batch of six large models can be selected: MiniMax, Dark Side of the Moon, Zhipu AI, Orion Starry Sky, Zero One Thing, and Baichuan Intelligence. DingTalk President Ye Jun described these as the "seven dragon balls" that summon the dragon.

What will work look like in the age of AI? You may be able to see some answers in the AI ​​search and AI assistant functions in DingTalk version 7.6.

In the view of DingTalk President Ye Jun, as the industry moves from model innovation to application innovation, exploring the application scenarios of large models will be DingTalk’s responsibility.

APPSO has previously put forward a point : We need more AI applicationists who do not make large models to transform large model capabilities into productivity, thereby affecting people's work and life.

Now, more and more large model and AI product companies are accelerating this process.

DingTalk’s AI search is different from Perplexity

For example, the AI ​​assistant released previously aims to solve the problem of bloated and scattered functions of DingTalk. The AI ​​search released by DingTalk today focuses on solving the problem of information fragmentation on DingTalk. This is exactly the same as the commonly used general AI search routes such as Perplexity and Secret Tower. different.

Specifically, DingTalk AI search has six major features: proprietary personalized search, sensing information changes, natural language input, directly generating answers, in-depth digging and questioning, and tracing content sources.

For example, Ye Jun said that if you want to know about the recent major progress in globalization work, you only need to use AI search. It can use the understanding, reasoning, generation and other capabilities of large models to list customer progress, product information based on daily data and information. Iteration, market strategy, cooperation progress, etc.

In addition, the AI ​​summary will also have built-in references to prevent AI hallucinations, and generate DingTalk brain maps to make the structure clearer.

Or, as the president of DingTalk, Ye Jun receives countless customer feedback every day. Now, with AI search, you can summarize and analyze customer needs with one click, helping to improve work efficiency. Including when asking the department that undertakes customer projects, the results given by DingTalk AI search are clear at a glance.

The ability of AI search to build knowledge networks can also be applied to scenarios such as writing weekly reports and task management. This means that the way users process information will change from the original "time flow" style to a matter-centered one, allowing users to be more effective. , focus more intently on important things.

This is also the biggest feature of DingTalk AI search, which is to integrate, logically, and network knowledge to achieve comprehensive knowledge, thus becoming everyone's "book of answers."

An interesting case was also shown at the press conference. For example, if you ask AI Search how many people Ye Jun promised to drink coffee last week, DingTalk AI Search can integrate the group chat records and get the promise in less than half a minute. 3 times, involving the answers of 9 students.

This also reflects the significant improvement in semantic understanding, logical reasoning and information integration capabilities of AI assistants with the support of large models.

Currently, DingTalk AI search has opened invitation testing. Now users can click on the search box at the top of DingTalk APP/PC to apply for internal testing.

The upgraded AI assistant can access more workflows

In January this year, DingTalk launched an AI assistant product, allowing everyone and every company to create their own super assistant. As of the end of May, the total number of DingTalk AI assistants reached 500,000.

Since the DingTalk AI assistant market was launched more than a month ago, more than 700 AI assistants have been launched.

Whether it is work reporting, meeting coordination, life entertainment, or music creation, AI assistants can provide all-round support. Especially in terms of product capabilities, DingTalk has significantly upgraded the thinking system, perception system and action system of the AI ​​assistant.

  • Thinking system: AI assistant has stronger memory and reasoning planning capabilities
  • Perception system: Perceives changes in the scene and automatically performs specified tasks based on the changes
  • Action system: Calls richer tools to achieve multi-Agent collaboration and anthropomorphic operations

After the user authorizes it, the AI ​​assistant can remember relevant information, habits, preferences, etc., including name, position, superior-subordinate relationship, and work task progress. It also supports user-defined memory settings, so that the results generated by the AI ​​assistant can be thousands of times. People have thousands of faces.

In other words, DingTalk AI Assistant is expected to become your “second brain” in work scenarios, understanding you better than you do yourself.

When performing specific tasks, the AI ​​assistant's enhanced reasoning and planning capabilities can deeply think about and rationally dismantle tasks like a real person. For example, it can quickly generate personalized summary weekly reports based on different people's schedules, documents, meeting minutes and other memories. .

With the ability of multi-Agent collaboration, users can make decisions in a workflow or group chat and let multiple different AI assistants collaborate to complete tasks together.

For example, in the on-site demonstration at the press conference, with the collaboration capabilities of multiple AI assistants, users can create a reservation itinerary for an egg-fighting game through a simple @ number in the egg-fighting old friend group, and summarize the winning rate of the previous egg-fighting game. , use the music creation master AI assistant to create music, etc.

For another example, ask the DingTalk AI assistant in charge of itinerary planning about the weather in Hangzhou tomorrow. After using the online query function, the AI ​​assistant not only knows the weather, but also humanely summarizes your work plan for the next week.

Ye Jun emphasized the need to use AI to help break down corporate data walls and upgrade existing SaaS to AI assistants through DingTalk low-code.

In this process, users only need to talk to invoke AI and let AI run errands for them to complete complex operations, truly liberating everyone's work.

Workflow can improve the accuracy of AI in processing complex and multi-link tasks. By orchestrating the AI ​​execution process, multi-step tasks can be completed automatically and step by step, and when necessary, you can access the website or call various tools to complete the task.

At present, anthropomorphic operation and multi-agent collaboration have been integrated into the workflow. Users can directly configure document creation, schedule posting, to-do and other nailing functions in the workflow, as well as more than 20 third-party services such as weather query, route query, and OCR recognition. Richer tools can be integrated by accessing API interfaces or DingTalk connectors.

Ye Jun said that the upgrade of DingTalk's AI capabilities will be further opened to ecological partners and customers through Assistant API and Inside API, providing scenario-based intelligent services to truly promote AI into applications, collaboration, operations, etc. in the scene.

DingTalk said that DingTalk is committed to making AI inclusive and making AI accessible to everyone, so it provides free credits. If you need more advanced product features or want a customized solution, you will have to pay for a consultation.

What kind of AI search do we need?

When I first entered the content industry, search engines such as Google were important channels for collecting information and selecting topic materials. But since last year, I have used less and less traditional search engines. AI searches such as Perplexity and Tiangong provide a more efficient information processing experience.

AI search greatly improves the speed of information retrieval and shortens the process of filtering out our massive search results.

However, the information provided by AI search may not be reassuring. We have previously reported that more and more AI-generated content is included in search engines, and what is ultimately presented to users is likely to be junk results of AI superimposed on AI.

Improving the efficiency of information acquisition is the essence of technological improvement. AI provides new solutions, but the demand for obtaining high-quality information has not yet been fully met.

In office scenarios, internal information management is a somewhat neglected scenario. A McKinsey survey shows that the typical knowledge worker spends more than a quarter of their time searching for information. The information here is not only external, but also quite complex information accumulated internally.

After former Google engineer Arvind Jain founded the cloud data company Rubrik, he found that because data was scattered in a large number of different software, his work efficiency was delayed due to the time it took to find the correct information.

Jain believes that finding information is the biggest challenge facing people's productivity, so he started to found Glean, a company focused on enterprise AI search.

▲Arvind Jain

Many companies are now exploring the application of AI in enterprises. In addition to improving the efficiency of repetitive and low-knowledge-density work, another aspect that can have a significant impact on organizational effectiveness is the management of corporate digital assets.

DingTalk did not do universal search and instead used AI search to solve the problem of information dispersion within the application. It was also a timely supplement to the current mainstream AI search.

When more and more information is stored in office applications, such as group chats, meetings, to-dos, documents, logs… How to evolve from the logic of people being driven by massive information in the past to a people-centered processing of information may be the answer for many enterprises A key step to achieve intelligence through AI.

After DingTalk's AI assistant supports memory and adds multi-agent collaboration and other capabilities, it actually brings more possibilities to the AI ​​search experience. For example, after AI sorts out the required document information, notifications, collaboration and task distribution with other people can all be connected.

Although DingTalk did not disclose many details at the press conference, AI search + AI assistant may indeed form a more complete workflow within the enterprise .

According to ReportLinker's forecast, the global enterprise search market is expected to reach US$6.9 billion by 2028, with enterprises' demand for efficient search and knowledge management increasing.

In fact, the biggest innovation in application experience brought by large models is also reflected here. Based on the huge knowledge base, what users have to do changes from searching to asking questions.

In addition to publicly searchable knowledge bases, can individuals or companies also form such a complete knowledge base in more subdivided scenarios in different industries?

In fact, we may also see similar logic in the recently released Apple Intelligence and Hongmeng native intelligence. The personal information of the device is integrated into a personal knowledge base, and users can retrieve and call functions using natural language at any time.

Perhaps this is a coincidence with the logic of AI information processing. DingTalk is also one of the first applications of Huawei's native Hongmeng that is most adapted.

In the native Hongmeng system, you only need to say a word to Xiaoyi, and the intent framework can understand the user's intention to hold a meeting on DingTalk, directly find the contacts in DingTalk, and then start a DingTalk meeting.

Today, a large number of mobile terminals have almost become the memory bank of our lives, and the same is true for enterprises. The data scattered on different devices is, to some extent, the digital twin of the organization.

Compared with personal data, internal data within an enterprise needs to shift from fragmentation and discreteness to structure and systematization to form an exclusive knowledge asset library. The organization and structure of information are the key to realizing enterprise intelligence. In this scenario, calling general knowledge from the public domain becomes relatively less important.

Whether in general search or enterprise search, we need a more intelligent and efficient search tool that can not only process massive data, but also accurately extract high-value information.

This tool must have the following characteristics:

1. Intelligent data integration: Able to integrate and analyze scattered data to form structured information.

2. Precise information screening: It has an efficient algorithm and can quickly find the most relevant and valuable content in massive information.

3. Contextual understanding: Ability to understand and analyze user needs and provide more targeted search results.

4. Dynamic update: Update the information database in real time to ensure the latest and most accurate information.

5. Security and privacy: Ensure data security and user privacy, and prevent information leakage.

As Ye Jun, CEO of DingTalk, said, “In the AI ​​era, the knowledge of all mankind is easy to search, but the knowledge accumulated by enterprises or individuals is difficult to find.”

We are in an era of unprecedented information explosion. Rapid acquisition and use of high-quality information will determine the future.

Although traditional search engines may be eliminated, the need for efficient acquisition of high-net-worth information will never disappear, and the exploration of this need will continue to drive more technological advancements and innovations.

Authors of this article: Li Chaofan, Mo Chongyu

# 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