1 Rapid creation of intelligences
● Click the "Create Smartbody" button on the homepage and "My Creation", and you can get your own Smartbody in 5 simple steps:
○ Step 1: Fill in the name and profile of the smart body and use AI to generate an avatar, or upload a photo as the smart body avatar.
Step 2: Fill in the detailed settings (system prompt), reply suggestions, opening remarks and other content. You can click the brush button on the upper right corner of the AI to automatically generate the corresponding content, or you can refer to the detailed settings provided by the example, appropriate modification to meet your needs.
Step 3: Upload and configure the knowledge base of the intelligent body. When the large model encounters a relevant problem, it will automatically determine whether it needs to call the knowledge base to help answer the user's question, and the knowledge base will significantly reduce the probability of the model's "hallucination".
○ Step 4: When you finish editing each form, the right side will automatically update the configuration of the smart body (expect a delay of about 3-5s), and you can instantly view the effect of the smart body after the configuration is updated.
Step 5: When you are satisfied with the effect of the smart body, click the upper right corner of the release, in the release page to complete the public configuration and final release. After review and approval, you can use it in the public number.
2 Detailed Settings
summarize
● Detailed settings are central to the creation of intelligences and determine the form in which the intelligences will answer the user's questions.
● Sometimes the detailed settings automatically generated by the AI do not meet your requirements, and you can fine-tune the base to try to set up the intelligences. Sometimes a very small fine-tuning can make a big difference to the effect of the smart body.
● You can also add a persona description to the Detailed Settings to have the Intelligence play you or another character.
paradigm
It is recommended to use a structured approach to describe the roles, workflows, and principles of the intelligences, which can be fine-tuned based on the following detailed setup templates for "Customer Service Intelligence" or "Digital Split Intelligence".
Customer Service Intelligence
# RoleYou are a professional and responsible XXX public customer service, able to quickly and accurately respond to all kinds of questions raised by customers on the public website, based on the "XXX" knowledge base.## Skills### Skill 1: Answering customer questions- As soon as a customer asks a question on the public website, the answer is found in the "XXX" knowledge base and responded to.- Try to select coherent content from the knowledge base for your answer, and don't split various pieces from the knowledge base.- Extracted from the knowledge base is just organized and not directly exported. For direct operational requests, extract the appropriate solution to the user's problem directly from the knowledge base, and recommend the appropriate public article at the end of the answer### Skill 2: Handle unanswerable questions- If you have a question that cannot be answered, direct the user to qq group 1234567 to ask a question.## Restrictions- Respond only to questions posed by public customers and not to other extraneous content.- Outputs are prioritized by reference to the "XXX" knowledge base, and only when no answer can be found are outputs produced by other means, e.g., by using the model's own capabilities.
digital doppelganger type intelligence (DDI)
# Role- You are a xxx style blogger :xx.- Speak to the user in the first person.## Personnel- Setting 1 ......- Setting 2 ......## Expression Habits- Colloquialization and informality: The use of colloquial and informal expressions makes the text more lively and grounded.- Humor: Humorous language and metaphors are interspersed throughout the text, making the reading experience easier and more enjoyable.- In-depth analysis: despite a colloquial language style, has a deep understanding of the relevant field and is able to provide detailed technical analysis and industry insights.## Skills### Skill 1: Answer questions1. As soon as a user asks a question on the public website, the relevant answer is found in the "xxxx" knowledge base and responded to.2. Try to select as many answers as possible that are coherent from the knowledge base, and do not break down the various pieces from the knowledge base.3. Extracted from the knowledge base just to organize, not direct output. For direct operational requests, extract the appropriate solution to the user's problem directly from the knowledge base and recommend the appropriate public article at the end of the answer.### Skill 2: Handle unanswerable questions1. If you have a question that cannot be answered, direct the user to xxx to ask the question.## Binding- The output is prioritized to the "xxxxxx" knowledge base, and only when no answer can be found are other methods of output applied, e.g. through the model's own capabilities.- Whatever point of view is expressed should be answered from the blogger's point of view.- Knowledge that is beyond what the role can answer, such as unknown and absent from the knowledge base, or asking for knowledge that is beyond the role (Hegelian philosophy of natural law, etc.) can optionally be generated with modeling capabilities.- Try to keep your response to 300 words or less. But in order to ensure the accuracy and completeness of the reply, you can also reply with more than 300 words. (Note: the public number can only give a reply after the model outputs all the contents together, the output text content is too long may prolong the user's waiting time)- You have to generate secure responses.- We are able to respond quickly and accurately to all types of questions posed by our customers on our public website, based on the "Digital Life Kazik" knowledge base.- Use emoji to add interest and intuition, but make sure they are closely related to the content of the question.
3 Knowledge base
summarize
● A knowledge base is a collection of documents. Multiple documents can be included under one knowledge base, and multiple knowledge bases can be included under one intelligence (recommended to be less than 10, more than 10 may result in inaccurate knowledge base calls).
● After receiving a user's question, the hybrid large model will determine whether the user's question intent requires going to a certain knowledge base for relevant information. If it is needed, the model will compare the similarity of the information provided by the user with the documents in the knowledge base and find the most relevant content to assist the model in answering the user's question.
Creating a Knowledge Base
● Create a knowledge base - select public articles - authorize public numbers, you can automatically pull the public number history of group articles, and you can choose "daily update".
● After all the files have been uploaded successfully, the meta ware will start parsing your files in the background, you can leave the editor and come back to test the smart body effect when the document parsing is finished.
How is the knowledge base named?
:: Intelligent bodies determine whether or not to invoke the knowledge base by using the "description". If you find that the intelligent body does not call the relevant knowledge in the dialog, you can adjust it by modifying the description of the knowledge base.
○ If the knowledge base has done any categorization of content, it can be written:
■ A knowledge base about XXX to retrieve the contents of the knowledge base when XXXX.
○ If the knowledge base is not categorized by content, it can be written in such a way as to increase the probability of the knowledge base being called:
■ XXX's public articles, answering XX's related questions and questions related to XXX's public articles. The output of the answer content of the intelligent body is prioritized to refer to the "XXX" knowledge base, and other methods are used to output only when the answer cannot be found.
● Uploading one file per article to the knowledge base works better than combining multiple articles into one large file.
How to restructure the answer?
● You can check the box "Structured answer format" to get a more organized output.
Note that this option only works for intelligences that carry only the knowledge base without plugins and workflows.
When checked, the intelligent body will give more neatly formatted answers (including clear subheadings and subpoints), and tend to have a serious and formal output style. However, this option may weaken the detailed settings for persona and language style, so it is not recommended to check this option if you want the intelligences to output answers in a relaxed and informal format.
(Turns off structured output switch)
(Turns on structured output switch)
⚠️ Note
● Through repeated debugging, if you find that the intelligences still cannot call the knowledge base, you can directly add this paragraph to the detailed settings to improve the probability of the knowledge base being called:
○ 1. When there is only one question in [Question] and the question cannot be answered based on the content of the document/link, please reply with "I don't seem to have found anything related to [Core Question] in the document/link you sent in." Please reply "I don't seem to find anything in the document/link you sent in that relates to [core question]", where [core question] is a distillation of the core content of the question, and do not try to answer the question based on your own knowledge.
○ 2. When there are multiple questions in [Question] and some of them cannot be answered based on the documentation/links, first point out that the questions cannot be answered based on the documentation/links. If it can be answered based on the model's own knowledge, then try to answer it based on its own knowledge, otherwise refuse to answer the question, or give some suggestions to guide the user to provide more information.
○ 3. Regardless of whether there is one or more questions in [Question], for questions that can be answered based on the content of the document/link, it is sufficient to give the answer based on the content of the document/link.
4 Release
After creating a smart body, you can publish the smart body to WeChat and let the smart body receive your users.
Public numbers are categorized into service and subscription numbers.
● The Service can proactively push intelligent body openers with preset question suggestions.
● Subscriptions can't be actively pushed, only passive responses to user questions.
○If the smart body does not generate a response within 15 seconds, the subscription number will prompt the user to reply "Continue" to complete the response generation.
○If an intelligent body replies with more than 550 words, the subscription number will send the first 550 words and then prompt the user to reply "Continue" to get the rest of the answer.
○If the user asks a new question after being prompted to reply "Continue", the smart body will answer the new question and discard the answer to the original question.
Post to the public backend
A public number can only be tied to one smart body.
After posting to a public website, keyword replies on the WeChat public platform will be turned off by default. You can turn this feature back on and it does not affect the normal use of smart body replies.
Step 1: Click on WeChat Public Channel - De-Authorization
Step 2: Scan the QR code to complete the authorization
Step 3: Publish
● After completing the configuration process, you can check the WeChat Access and click Publish to submit to the review process.
● After the approval, you can click the WeChat public number logo in the smart body card in My Creation - My Smart Body, and open the public number by scanning the QR code with your cell phone WeChat.
● Note that the public scope of the release page only affects the visible scope of the intelligences in the metaverse and the metaverse, and does not affect the use of the microsoft public.
Configure to the public menu bar
Configure the applet link
You can also set jump to applet in the menu. When you click the button, you will jump to the associated applet page directly.
Configuration process
First, go toWeChat Public Platform, click Content & Interaction - Customize Message - Add Menu.
If you are the first time to configure the jump small program, WeChat public platform will prompt you to associate small program, according to the following process to open this function.
After successful opening, click Associated Applets to search and select the applet.
After completing the above steps, go back to the custom menu setting interface, click Select Applet, you can find the applet you just associated.
You can get the path of your smart body applet in Tencent MetaWare [Workbench-My Creation-Experience-Tencent MetaWare-Small Programs], and just copy it into the path column.
Once the configuration is complete, click Save and Publish. The publish will take effect for all users within 24 hours.
Configuring Menu Prompt Messages
You can add buttons to the public menu, and clicking on them gives a text prompt that allows you to have a conversation on the chat page.
Configuration process
First, go toWeChat Public Platform, click Content & Interaction - Customize Message - Add Menu.
Just enter the menu name, select Send Message, Text, and enter the text content you want to send. You can also configure graphic messages, pictures, videos and other message forms.
Once the configuration is complete, click Save and Publish. The publish will take effect for all users within 24 hours.
Configuring Intelligent Bodies to Public Articles
Operation Path: Article Edit - Insert Applet - Applet Link - Set Title / Display Method
How to get the link to the small program: WeChat open your small program smart body, copy the link in the upper right corner and paste it into the box
You can customize the display format of text/image/application cards; and customize the card title and cover style.
