Technology Experience Officer | DingTalk slash '/' actual test: a private customized robot appears in group chat, and a small program can be generated by taking photos

DingTalk posted a slash at the 2023 Spring DingTalk Summit, demonstrating four application layer scenarios with large model capabilities. After DingTalk is connected to Qianwen's large model, users can activate more than 10 AI capabilities by inputting "/" to realize conversation or photo generation applications, one-click creation of Q&A robots that can automatically learn documents, automatic creation of document copywriting, and intelligent extraction of videos. Features such as meeting and group chat message summaries.

Recently, DingTalk slash “/” has begun targeted invitation testing. The reporter summarized some product features by trying DingTalk slash “/”.

The AI ​​robot joins the group chat and takes photos to generate a small program

In DingTalk group chats, the “/” smart assistant can automatically generate chat summaries, create Q&A bots, and build applications with zero code.

Faced with the massive chat records in the group, climbing up the stairs is time-consuming and laborious. Using the "/" smart assistant, users can easily know the context of the event and grasp important information.

For example: When discussing content topic selection in a group, new members of the group need to quickly know key information about the topic selection direction. They can use the "/" assistant to generate a content summary with one click and quickly understand the key points of the discussion.

In DingTalk group chat, users can also use "/" to create an AI question and answer robot. It has the ability to automatically learn document knowledge and does not need to manually set questions and answers before it can start working.

The reporter’s experience found that the Q&A robot can automatically learn the content of DingTalk documents, and can also “feed” it multiple documents so that it can automatically add and update Q&A content. After creating a question-and-answer bot, every group member can tag it in the group, ask it questions, and get intelligent replies.

For example, the reporter fed the robot multiple documents in advance for the robot to learn, and asked the intelligent robot questions about the band Mayday, and the robot could accurately tell the band members and related information.

In addition, companies can use it to train their own small models in many scenarios, such as community operations, customer service Q&A, new employee training, product Q&A, legal consultation, insurance consultation, etc., both internally and externally. Can be used externally. Imagine if there was such a Q&A bot in the community, then administrators would be spared the trouble of answering repetitive questions, and community members would not be embarrassed because no one answered their questions.

In a group chat, by passing "/", you can also use natural dialogue to generate applications, such as voting applications. You can also automatically create a low-code application by taking pictures of hand-drawn forms and blackboard writing. The generated application can continue to use dialogue. How to modify application options and fields.

For example: You can enter "initiate group voting" in the group chat to initiate a group voting. What should we have for dinner tonight? And install the generated group voting into the current group to quickly collect opinions and feedback from group members, support viewing voting results and remind everyone to vote.

You can also build applications in the group, quickly draw handwritten forms, and take photos to identify and generate applications, so that you can easily and efficiently collect information in the group and automatically complete data summary. For example, generate an "Equipment Inspection Form", draw a handwritten form or paper form, and perform photo identification.

After the image recognition application is generated, the user can continue to modify its options and fields through dialogue, such as changing the device model input box to a drop-down box, or deleting an unnecessary field in a row, or changing a field. This is done by talking to the AI.

After the adjustment is satisfactory, the user can install the application to the current group with one click. Members in the group can quickly fill in the form and view the form details. They can also click on the group shortcut bar application such as "New Product Feedback Sampling" to directly submit the form information. Group members can click on the operation bar at the top of the page to enter the "Data Management Page" to view all data submitted in the form.

One key to wake up "/" can write, draw and draw

According to reports, the AI ​​of DingTalk Documents is built based on the Tongyi large model, including Wenshengwen, Wensheng pictures, Wensheng tables, smart summaries and other capabilities. Through the DingTalk document "/" intelligent assistant, you can perform AI intelligent creation, rewriting, summarizing, polishing, translation, continuation, etc., covering most document usage scenarios for users, and is a small assistant for inspiration, creation, and organization.

The “/” smart assistant has literary capabilities. For example, it can wake up the “/” smart assistant and ask it to help write a promotional copy for Mayday’s concert with the theme of [Noah’s Ark]. At this time, the reporter did not give specific information. , the smart assistant wrote "Draft 1", which briefly introduced the band and theme of the concert, as well as the venue location, ticket purchase method and other information.

The reporter thought that the content of "Draft 1" was not rich enough, so he woke up the smart assistant to "continue writing". At this time, Slash added some text content to describe the atmosphere and intermediate links of the concert.

The reporter noticed that the ticket purchase method and event address are inaccurate and need to be modified. Just select the modified part and type in the correct content to replace it. You can also ask the "/" smart assistant to provide a more optimized transportation schedule based on the surrounding traffic conditions.

Of course, some parts of the text generated by the intelligent assistant are not very accurate. The reporter adjusted the position of the paragraph and generated the following text, which is in line with expectations. It can be seen that the intelligent assistant still has relatively little understanding of the connotation of the [Noah's Ark] concert. If you want to get perfect copywriting, you still need to polish it manually.

Clicking Generate Summary can also compress the above copy into a short summary.

At the same time, the "/" intelligent assistant can also brainstorm, such as brainstorming online car-hailing topic planning. The "/" intelligent assistant can give relevant suggestions including user needs, comparison of different platforms, innovative service methods, etc. And be able to rewrite the tone.

Moreover, for the first time, DingTalk has implemented Wenshengtu into the work scene. It supports AI drawing in documents and chat scenes. It can be used to draw landscapes, characters, real objects, etc., and add ink, cartoon, impressionism, advertising illustration, and Ji. Expressions of various painting styles such as brilliance and sticker art.

For example, if you enter the text "There is a bee on the peony flower", you can get pictures of the same scene, and you can have multiple choices.

The "/" intelligent assistant also has the function of writing a literary form. For example, the "/" intelligent assistant is asked to generate a recruitment interview schedule, which includes the department, interviewer's name, interview time, interview results, and the next round of interview time. The following graph was obtained.

Write at the back:

In addition to the above capabilities, DingTalk slash "/" is also a powerful working partner in meeting scenarios.

On June 1, Alibaba Cloud announced new progress in the Tongyi model, and the new AI product "Tongyi Listening" focusing on audio and video was officially tested. According to reports, the Tongyi Listening Large Model has been specially enhanced in audio and video scenarios, integrating more than ten AI functions, and can efficiently complete the transcription, retrieval, summary and organization of audio and video content. In addition to "good listening", it can generate meeting minutes with high accuracy and distinguish different speakers. It also has "high understanding" and can divide audio and video into chapters and form summaries in one second, summarize the full text and each speaker's views, and organize to-do items. and focus.

After testing, the DingTalk meeting after being connected to Tongyi Listening can achieve summary capabilities with no time limit and no word limit. The technical implementation method is to segment the content based on real-time speech to text. Summarize, and then process and polish it as a whole. For example, for a two-hour, 30,000-word meeting, it will intelligently divide chapters according to the content, such as 30 pieces of 1,000-word content, to process the meeting minutes, and finally integrate similar content based on the information to generate A summary of several major topics.

More DingTalk slash "/" functions and product usage logic require more user testing experience.

The above is the detailed content of Technology Experience Officer | DingTalk slash '/' actual test: a private customized robot appears in group chat, and a small program can be generated by taking photos. For more information, please follow other related articles on the PHP Chinese website!