The Facts
In the last few weeks, there have been some huge announcements about LLM “agents” — the result of hooking up language models to tools. The most eye-catching announcement was ChatGPT Plugins, but there have been a few others worth noting:
Fixie announced $17M of funding on March 30th to build a platform for LLM agents
Adept announced a $350M fundraiser on March 15th for their automation platform
Langchain continues to grow at an amazing pace (17k stars and counting) as the premier open-source framework for building LLM agents
Why it Matters
As LLMs get more powerful, they’re getting better at logic, reasoning, and planning:
GPT-4 is great at breaking down complicated problems into tractable steps — so the only limitation is the model’s ability to actually execute each of those steps. OpenAI Plugins, Fixie, Langchain and Adept are all attempts to give models the tools needed to complete more complex tasks.
As LLMs can directly solve more complex problems, they become a new form of operating system, where complex problems can be solved with natural language.
My Thoughts
LLM agents represent a novel form of computing — by granting new skills to language models, the number of tasks they can complete is amplified. I think agents are the clear path to automating more tasks with AI.