present (sb for a job etc)
AgentBuilder is an intelligent body (Agent) platform launched by Baidu based on Wenshin's big model, which supports developers to select different types of development methods according to their own industry fields and application scenarios, and to build product capabilities in the era of big model. Developers can develop intelligent bodies (Agent) at low cost through prompt orchestration, and at the same time, the Wenshin Intelligent Body Platform will also provide intelligent body (Agent) developers with corresponding traffic distribution paths to complete the commercial closed loop.
For the commercialization section, we recommend reading theKiller traffic portal: using AI intelligent body to get external traffic for websites and public numbers in the long runThe
Wenxin Intelligent Body Core Components
in a word
Wenxin Yiyin is a large language model for knowledge enhancement, based on the Flying Paddle deep learning platform and Wenxin knowledge enhancement large model, continuous fusion learning from massive data and large-scale knowledge with knowledge enhancement, retrieval enhancement and dialog enhancement technology features. Wenshin Yiyan has five major capabilities, namely, literary creation, business copywriting, mathematical and logical reasoning, Chinese comprehension, and multimodal generation.
Prompt
AI Model Prompt Words. When training models with supervised or unsupervised learning, Prompt can help models better understand the intent of inputs and respond accordingly. In addition, Prompt can also improve the interpretability and accessibility of the model. In layman's terms, it provides a "hint" or "guideline" to the AI model to help it better understand and accomplish the task.
visual organizer
The platform provides a way to quickly build a business flow to complete the development of intelligent bodies through a visual interface and drag-and-drop kit connection combinations; the platform has built-in the following capabilities in this mode, which can be used out-of-the-box.
Model: The platform provides a large model of Baidu Wenshin Yiyin, which developers can call directly.
Data: Own data can be uploaded to support the model to index and refine the data to serve users.
Framework: The platform has a built-in conversational C-suite template style that supports serving users in conversational scenarios.
Components: The platform has built-in C-suite components, such as the welcome bubble language component, which developers can configure with a single click.
in a word
Wenxin Yiyin is a large language model for knowledge enhancement, based on the Flying Paddle deep learning platform and Wenxin knowledge enhancement large model, continuous fusion learning from massive data and large-scale knowledge with knowledge enhancement, retrieval enhancement and dialog enhancement technology features. Wenshin Yiyan has five major capabilities, namely, literary creation, business copywriting, mathematical and logical reasoning, Chinese comprehension, and multimodal generation.
Prompt
AI Model Prompt Words. When training models with supervised or unsupervised learning, Prompt can help models better understand the intent of inputs and respond accordingly. In addition, Prompt can also improve the interpretability and accessibility of the model. In layman's terms, it provides a "hint" or "guideline" to the AI model to help it better understand and accomplish the task.
visual organizer
The platform provides a way to quickly build a business flow to complete the development of intelligent bodies through a visual interface and drag-and-drop kit connection combinations; the platform has built-in the following capabilities in this mode, which can be used out-of-the-box.
Model: The platform provides a large model of Baidu Wenshin Yiyin, which developers can call directly.
Data: Own data can be uploaded to support the model to index and refine the data to serve users.
Framework: The platform has a built-in conversational C-suite template style that supports serving users in conversational scenarios.
Components: The platform has built-in C-suite components, such as the welcome bubble language component, which developers can configure with a single click.
cable length (= 1
In the visual orchestration development model, the platform provides a series of development kits, which are building blocks of business processes that consist of inputs, outputs, and configuration parameters that define their functionality. Chain is one of the most important types of kits, which allows us to combine multiple kits into a single plug-in. For example, we create a chain that accepts input from a user, then formats the user's input as a prompt word via a prompt word template, and then inputs this prompt word into a large language model, which ultimately outputs the result. We can also combine some chains together to build more complex chains.
What you can do with the platform
Open to multiple subject types: Based on the concept of AI for everyone, the Wenshin Intelligent Body Platform supports enterprises, media, government and other types of organizations and individual developers to move in.
Open to a wide range of industry types: Wenxin Intelligent Body Platform supports developers of various vertical industries to move in and build a large model-based industry matrix ecology.
Multi-competency developers: The Wenxin Intelligent Body Platform supports the onboarding of developers with different capabilities.
- With development ability: the platform provides a full set of technical solutions for development, and supports developers to develop intelligent bodies/plug-ins independently.
- Rich data is available: the platform provides a low-code technical solution to provide data to accomplish smart body/plugin creation.
- Inspired ideas: the platform provides a zero-code solution to quickly transform imaginations into intelligences/plug-ins.
Advantages of the platform
- Powerful big model capability: Relying on the Wenxin Yiyin big model, it has good performance in content creation, mathematical and logical projection, Chinese comprehension, multimodal generation and other aspects;
- Diversified ways to link users: developers can choose different types of development methods, template components, etc. to access, including zero-code/low-code intelligent body, data class/capability plug-ins, to provide C-users with more high-quality services;
- Multi-scenario to reach users: traditional search & AI search dual-engine distribution; Wenxin Yiyin App call plug-ins; intelligent body & plug-ins can be accessed in the experience center, distribution to C-users, but also for enterprise-level developers.
Introduction to Intelligent Body
An Intelligent Body, whose English name is Agent, is an intelligent entity that has the ability to think and act actively based on a large language model and has the following characteristics:
- Ability to think and act proactively
Intelligent bodies are not only able to respond passively to commands, but are also capable of active thinking and decision-making.
- Sensing and understanding needs
Intelligent bodies are able to understand the user's needs, which usually involves understanding natural language.
- Deconstructing goals and shaping planning
Intelligent bodies are able to break down complex tasks into smaller, more manageable steps and develop plans to achieve them.
- memory skills
Intelligentsia possess a degree of memory capacity to store and recall previous interactions, knowledge, and experiences as a means of guiding current decisions and behaviors.
- Using Tools and APIs
Intelligentsia are able to utilize a variety of external tools and application programming interfaces (APIs) to perform tasks and access information.
- Decision-making and action
Ultimately, intelligences are able to make decisions and take action based on the above processes.
Intelligent Body Type
- Zero-code intelligences: Expressing intent, providing behavioral descriptions, introducing datasets, tools, and other capabilities to create intelligences through prompt editing.
- Low-code intelligences: Quickly build business flows by drag-and-drop, combining components such as big models, datasets, tools, and other components to complete intelligent body development.
AI Plugin Introduction
AI plug-in is a carrier that combines the emerging content and services of the big model, and is a solution for the extension of the big model application scene. If the big model is an intelligent center brain, plug-ins are the ears, eyes and hands. Plug-ins combine the AI capabilities of the big model with external applications, which can not only enrich the capabilities and application scenarios of the big model, but also utilize the generative capabilities of the big model to complete previously unattainable tasks, and flexibly and conveniently link users. For example:
- Information Enhancement
Such plug-ins can help users get more timely and professional information, such as Wenxin Yiyin access to Baidu search plug-ins, so that Yiyin can search the whole network of real-time information; In addition, there are also plug-ins to help users retrieve information in specialized areas, looking for a house, looking for a car, looking for the law, looking for stocks...
- interaction enhancement
Help Yiyan understand multimodal inputs such as PDF, images, voice, etc., and help Yiyan generate multimodal outputs such as mind maps and videos. For example, plug-ins that support users to upload documents and conduct Q&A based on documents, etc.
- Service Enhancement
Such plug-ins can help users automate some common tasks, such as booking flights, sending emails, managing schedules, creating questionnaires, etc.; it can also be the use of modeling capabilities to greatly enhance the experience of existing services, for example, you can ask the model to generate interview questions based on the user's resume and JD information, combined with the TTS/ASR, to create a real and unique mock interview for the user.
How plug-ins work
- Plugin Registration: The developer registers the manifest file of the plug-in to the One Word plug-in library, and after the verification is passed, One Word can use the plug-in to process the user query.
- Plugin Trigger: The parsing scheduler module will use the generated API to call the plugin service. After the plugin service finishes processing, the return json data is summarized by a word to return the results.
- Plugin Parsing: The trigger scheduling module of the one-word plugin system will recognize the user query and will select which plugin to use, as well as generate an API for calling the plugin, based on the natural language description of the plugin's API interface and parameters in the manifest file.
For example, a user selects the weather plugin on the platform and enters, "What's the weather in Beijing today?" The model will first call the weather plugin based on the user's intent and parse the time (today) and location (Beijing) information in the query. The model first calls the weather plugin according to the user's intent, and parses the time (today) and location (Beijing) information in the query, and then inputs it into the weather API interface provided by the developer in a json structure, obtains the weather information returned by the interface, and then generates a user-oriented answer after the big model makes linguistic embellishments.
Plug-in Type
- Data class plugin: Applicable to developers with high-quality data, submit existing data, you can quickly create data class plug-ins; can be based on the model's understanding of the data and the ability to process, to provide users with various types of query services.
- Capability Class Plugin: Applicable to developers who already have AI capabilities or models in various categories, provide the service domain name of the capability, and then quickly create a capability plug-in; without repeated development, it can be invoked in various traffic scenarios.
Multi-scenario service users
- Baidu Search: Based on the user's Q&A query, content and services can be distributed in both classic search and conversational AI scenarios.
- Wenxin a word: Baidu's newly launched Traffic Scene can be based on the content understanding and generation capabilities of the big model to deeply mine, understand and meet user needs.
- Experience Center: Intelligent bodies enter the experience center of the Wenxin Intelligent Body Platform, which can be distributed in the C-end multi-traffic scenarios, as well as for enterprise-level developers to expand the use group.
