In the past two years, the development of large models continues to be hot, so much so that many capital and scholars believe that AI has a bubble, the fundamental reason is still because the large model has not yet appeared to be able to combine with the industry to practically implement the application.
I'm not concerned about bubbles or not, I'm just concerned about what big models can do to help me. The closest way that big models can help users is with AI intelligences.
In this article, I will use an AI intelligent body platform to easily build an AI assistant according to demand. At present, there are more intelligent body platforms on the market, such as Baidu's Qianfan large model platform, the intelligent body platform of Smart Spectrum Qingyin, Coze and so on, and this article takes the domestic version of Coze as an example.
1. Demand
My consistent view on AI has been that it is our super assistant, not meant to replace us, but to allow us to do things faster and better.
This post to implement a requirement such asLittle A.It is an operation, one day the leader sent him a picture with the content of the article, and asked him to rewrite the article according to the content of the article in the picture, and then give the article a title, and then a cover picture, and finally in the publication to the public number.
The following uses the Smart Body platform to implement an AI assistant to helpLittle A.Easily accomplished tasks.
2. Analyzing demand
To realize the aboveLittle A.The following 6 steps are required for the needs of the
- According to the image link, the text content in the image is recognized by OCR.
- Call the big model and let the big model rewrite it, based on what's already there.
- Call up the big model and have the big model come up with a catchy headline based on the content.
- Call the big model and let the big model write the cue words for generating images based on the content.
- Call the relevant plugin to generate images based on the prompt word for generating images.
- Finally, the title, content, and cover art from the output of the above steps are output together.
Realize an AI smart assistant based on the above requirements next timeLittle A.Just enter a link to an image, and the intelligent assistant will give you the title, content, and cover image.
The final workflow is roughly shown below, and the step-by-step disassembly continues below.
3. Understanding the AI intelligent body platform
The Intelligent Body Platform is a platform where everyone can use and publish intelligent assistants. Published smart assistants can be used by yourself or published to a store for others to use. Take Coze for example (/), after registering, you can see the following interface:
Intelligentsia actually is:An assistant that can help us automate a series of tasks by utilizing the superbrain that is the Big Model.. It is somewhat similar to the RPA automated bots, but the previous RPA bots were only automated without intelligence and could not understand natural language.
There are 3 core concepts of intelligences：
- plug-in (software component): if an intelligence needs to accomplish a task, it needs a variety of tools.Plug-ins are tools. This tool can be developed on your own or you can use one published by someone else.
- workflow: If you want an intelligent body to accomplish complex tasks, only tools are not enough, you need to add thesequence of processes. Then workflow is where this process is choreographed.
- repository: Intelligent bodies sometimes need industry knowledge, such as legal knowledge, Q&A knowledge deposited within the organization, etc., to accomplish their tasks. That knowledge base isProvide relevant knowledge to intelligencesThe place.
The Intelligent Body Platform has 2 core concepts：
- store: Since it's a platform, it's a stage for supply and demand. There are bound to be all kinds of stores in it, smart body stores, plugin stores, workflow stores, etc.. All the stuff in these stores are posted up by others.
- (an official) standard: Since the idea is to make it accessible to everyone, there are bound to be some plugin standards, API standards, etc., so that you can publish your idea to the store if you implement it according to the platform's standards. This idea can be a plugin, workflow, smart body.
4. Creating Intelligent Bodies
Intelligent body platforms on the market meet almost the same in creating intelligent bodies. The left side is the persona area, the center is the creation area, and the right side is the debugging area. Each module in the creation area, when tapped, will enter a new interface for selection or editing.
A's requirements favor a fixed process, so he uses a "workflow" to implement it. Various plug-ins and big models can be used in the workflow.
5. Creating workflows
5.1 Understanding workflow
The core of the workflow has:Beginning, End, Node, choreographed with one thing in mind:As long as the nodes are connected, then the output of the previous node can be used as an input to the subsequent nodes。
A node in a workflow that can bePlug-ins, big models, knowledge basesAnd so on.
The overall feel of the workflow is like a chain with N nodes in series, as follows:
- commencement: Each workflow has a start node, and the start nodeOnly the input, which is used to accept user input.
- close: Each workflow has an end node, the end nodeOnly output, the output can be converted to the final answer result presented to the user. The output of any of the previous nodes can be combined as the output of the end node and presented to the final answer result.
-
nodal: A workflow is any node that consists of any number of nodes, each of which has inputs and outputs, and each of which can be connected to any of the nodes. Start and end are also a special kind of node.
- Inputs: any preorder node connected to this node can be used as an input to this node.
- Outputs: each node can have outputs and can pass the outputs to the back-order nodes connected to it.
5.2 Implementing workflows
The following workflow is utilized to implement the requirements of Little A in a total of 6 parts.
①, accept image links, use plug-ins to convert images to text
Select Create Plugin on the left to create your own plugin, or select a plugin in the store that someone else has done well. There are many regular plugins for general functions in the store, usually we search for them first, and if they are really not available, then we create them ourselves.
Here we can find ready-made plugins for recognizing text from images.
Each plug-in has its own output. After selecting a plug-in, the output of the preorder node is selected as the input of this node, and the output of this plug-in is also visible.
②, rewrite content using large models
The big model is generally chosen when content needs to be generated. The core of the big model is to understand natural language and generate content.
Click on the large model on the left, then in the large model edit box, select the specific model, inputs, cue words, outputs, and so on.
③. Utilizing large models to title content
Go ahead and add the big model, have it write the title, and remember to connect the relevant nodes!
④. Use the big model to generate the cue words needed for the picture
Go ahead and add the big model, have it write a prompt word that generates an image, and then pass the output to the image generation plugin.
⑤. Use the plug-in to generate images from the prompts.
Go to the store and find the relevant plugin (here it is generating images based on the cue word) and use the plugin to generate image links.
(vi) Integration of title, content, and cover image
Eventually, the output from multiple nodes in the preorder is summarized in the end node.
In the end node, you can select the output of the preorder node, and then, based on the output, modify the following format to generate the final answer content.
5.3. Test workflow
Once the workflow is created, it's time to try it out. Click on Trial Run in the upper right corner, enter the image link address, and wait for each node to finish running.
The results of the trial run are as follows, and you can see that it basically complies with theLittle A.The needs of the
6. Publish to store
Once the workflow has been created, you can both return to the SmartBody creation interface, select the workflow you just created, and then add an opening statement to preview and debug the SmartBody.
When debugging the smart body, you can see the input process of each node in between, and if the whole process is fine, you can publish the smart body to the store.
At this point it has been realized that a system that satisfies theLittle A.Demanding intelligences la!
Of course, there are all kinds of plug-ins and workflows for intelligences, so as long as you have an idea, you can implement all kinds of fun and useful intelligences.
7. Summary
This article mainly introduces how to use the intelligent body platform to build a private customized intelligent assistant, I hope it will be helpful to you!
The end of this article! Welcome to pay attention to, add V (ylxiao) exchange, the whole network can be searched (programmers half a cigarette)
Link to original article:/s/INdSD6wosqIALxiJcDpi1Q
