Hold on a moment… Don’t worry, nobody is taking your job, just yet!
Today’s post is inspired by the remarkable AutoGen framework developed by Microsoft. The concept of allowing Language Models (LLMs) to engage in conversations with each other is such an ingenious idea!
If you are not familiar with agents and how they work:
In essence AutoGen is a platform that simplifies the creation of conversable agents capable of solving tasks through inter-agent conversations. With AutoGen, developers can easily construct various forms and patterns of multi-agent conversations involving Language Models (LLMs), humans, and tools.
Think of ChatGPT as a single agent that assists you in obtaining answers to your questions. Now, having multi-agents is like having an entire team at your disposal — not just for getting answers but also for getting things done.
It effectively emulates how human teams collaborate to solve problems.
Moreover, as a human, you still have the option to be part of the team. Naturally, you can choose to let the team autonomously execute tasks either with or without your feedback.
The illustration below highlights a human feedback instance — such as “No, please plot % change!” — in the midst of the conversation. This directed the agents to different outcomes.
If the above illustrations don’t give you enough clue, check out the example below.
This example is from the AutoGen Github, and it’s definitely eye-opening🤩to experiment with. After setting up the two agents — the user assistant and the user proxy — you can kickstart a task by initiating a conversation between them.
Here’s the output:
As you can see, the human user wants to know “who would be interested in reading an Arxiv paper”, providing a link to the paper.