CS294-3-Autogen
CS294/194-196 Autogen
Finally back… So sorry for the updating absence these days.
Introduction
In nowadays GPT models, which is known as Generative Large Language Models, is extremely powerful in generating new tokens, including generating texts, images, or even videos. We see the hope of AGI (Artificial General Intelligence) coming in the next few years. However, for real-world scenario, there is still plenty of room for improvement in enhancing the breadth and depth of problem-solving for large language models.
- For depth: LLM’s reasoning and reinforcement learning may be the final way out! (We won’t discuss this topic now, you can skip to my another blog focusing on LLM reasoning instead)
- For breadth: Well, let’s dive deeper into this problem.
Assume you want to develop a small program for your study (a program for arranging daily plans, etc). You can you GPT-4o or Deepseek by using prompt of “Please help me develop a python program to arrange my daily routines and tasks automatically.”
Then models may reply a response like this:
I use Next-chat applications by filling my
gpt-api-key.
It’s great, but when the task is becoming more complex, it’s likely for models to make mistakes. For example, if you ask only one AI model to write a brand new operating system, it won’t get a satisfied response.
So what’s next? Enhancing the single performance won’t get too much progress compared to the increasing cost of computing resources. Thus, multi-agent is here to implement and solve large-scale developing projects!
Today, we gonna introduce Autogen[1], a framework for creating multi-agent AI applications that can act autonomously or work alongside humans. This framework is developed by Microsoft.