This morning, @shyamal from the OpenAI implementation team open-sourced the GithubSwarmThis OpenAI official multi-intelligence body framework. I have to say that the community impact gained by the official OpenAI downfall is just not the same, and a large number of parsing articles have already appeared in WeChat groups and circles of friends.
This Multi-Intelligence Framework has really taken the key to Multi-Intelligence and made it very clear that there are two cores defined inside Swarm「Agents」cap (a poem)"Handoffs", the core of multi-intelligence is on top of this Handoffs.After a brief look at examples I think this multi-intelligent body framework is not good enough, coincidentally, I am very familiar with the cloud native technology, borrow the development of cloud native history, to give a brief review of this [Swarn] framework: from the history of the development of the cloud native containers, the equivalent of docker swarm and k8s, we need the intelligent body framework should be k8s such a framework. A framework, if you are a cloud native technology familiar with the students easily know what I'm talking about.
Single Agent piece, simple encapsulation of the prompt word and the use of function calls can be completed business, OpenAI on a /api/chatcompletions interface to help us get it done, a large number of Agent products on the market are stuck in a single Agent, but the "Handoffs" piece, Swarm did do a very elegant. I think his design is not as good as our multi-intelligence framework, OpenAI's [Swarm] is docker swarm, our multi-intelligence framework is k8s, what I need is to orchestrate the intelligences like k8s orchestrates the containers, we have just released the multi-intelligence industrial design products to the public on September 26th, please see the details:The First Multi-Agent Driven Industrial Design Platform Enabled by AI Agent Foundry of Smart Use Research Institute Successfully Released。
The central challenge of multi-intelligence is the problem of communication between different intelligences. How and what information is transmitted is important. Collaboration between multiple intelligences only needs to be invoked when necessary. Look at our diagram of intelligence collaboration:
When our multi-intelligence application receives a request from a user, the design concept borrowed from Semantic kernel is to realize a "goal-oriented" AI application, which means that it can help to identify goals and then find ways and steps to achieve those goals. In the "goal-oriented" approach, goals are first identified and then broken down into a series of tasks to be performed by a Planner. These tasks can be performed one by one to achieve the final goal. This process is natural for humans, but relatively complex for machines. With the power of LLM AI, we can make this process easier.
The intelligent body that receives the user request is called the routing intelligent body, and it is responsible for routing to the task intelligent body that performs the task. The Planner of our intelligence framework is also similar to the "Handoffs" of OpenAI's Swarm, which handles the handoff logic, and our Planner is more perfect than Handoffs. OpenAI's Swarm is still in the experimental stage, and we expect it to be developed into a multi-intelligence orchestration framework like k8s:
This framework is written in python, do you think it's a good choice to write multi-intelligence applications in python? I personally think that to do application development, Python is not a good choice, Python is used more, because this wave of artificial intelligence is dominated by algorithmic engineers, they are accustomed to using the programming language is just Python, with the increase in demand for complex scenarios of artificial intelligence applications, the right of control gradually to return to the hands of the application developers, for the complexity of the high level, the need for long-term maintenance of the Application systems still need to use c#, java and other business systems development programming language to dominate.