Microsoft AutoGen

AutoGPT shall step aside now, it’s big brother has arrived and is here to take names

Throughout this publication, I have time and time again pointed out that the non-determinism of LLM feedback is a problem for agent-to-agent communication.

Now it seems that with AutoGen Microsoft has been building something big in the background. And I have to admit, AutoGen looks admirable.

Satya is likely the most effective tech CEO right now.

As one of the most active AI companies with a reputation for creating industry-dominating software, being an early investor in OpenAI, and also the creator of API-driven Cognitive Services, Microsoft has open-sourced a multi-agent collaboration framework that has the potential to change the world.

What do you think? Let me know in the comments.

Executive Summary

Microsoft proposes with AutoGen a framework for the simplification of the orchestration, optimization, and automation of LLM workflow. Within these workflows, agents can be configured to use LLMs, humans, tools, or a mix of those elements. These agents are customizable and conversable leveraging the strongest capabilities of the most advanced LLMs, like GPT-4. Microsoft states that workflows can be used to solve a variety of tasks, such as code-based question answering, that these workflows are easy to design and implement, and that they can be used to build complex multi-agent conversation systems.

Sounds interesting? Let’s dive in