What if AI agents could write their own tools/plugins? Introducing 🧰 Toolkit: the easiest way to create & discover AI Plugins for 🦜📎LangChain and ChatGPT Just describe a plugin in natural language, and get working LangChain Code. ChatGPT coming soon!

We'll soon be shipping an open-source version of this functionality that you can use locally to generate new tools. We'll also have the agent recursively run and improve them till they work as desired. Feel free to follow along here:

On the directory side, we're working on a way to register and save plugins, and turn on the entire library of plugins within your AI agents with just a few lines of code

The overall theme we've been exploring for the past few weeks is AI agents that can create their own tools in self-directed ways. This is both exciting, and poses serious safety concerns that need to be considered.

From the safety side, my position is that I'm concerned about what these auto-recursive agents are able to do. I believe that scaffolding agents with additional tooling is both an exciting opportunity, and also likely to lead to unintended risks & outcomes.

Similar to humans, giving agents additional tools increases both their capabilities, and their overall intelligence. Think of how tools as simple as paper have profoundly increased our intelligence as a human species.

I believe that with innovative tool-access alone, we'll be surprised by what models at the level of GPT-4 + Tools can do.

These new tool-enabled agents allow for a reimagining of a lot of the ways we use computers, and collaborate with one another. Stay tuned for more thoughts on the UX & infrastructure needed to support this approach.

Here's some thoughts on where we're heading this next little bit with Toolkit, and the future of tool-enabled AI agents. Stay tuned for some exciting new things we'll be shipping:

Wow this is insane Just tried it and was able to create a langchain plugin in seconds 🤯

wtf haha so cool @pharmapsychotic @makeitrad1 @huemin_art @Rahmeljackson