Autonomous LLM agents with human-out-of-loop

Author of Gorilla Execution Engine (GoEx) here! The key idea behind GoEx: 1. "Post-facto validation" for assessing LLM actions after execution. 2. "Undo" button and “Damage confinement” abstractions to manage unintended actions & risks!

GoEx (code below) inherently supports Gmail/Slack/Dropbox/Spotify/Github, try it out to build your own LLM agents!

Blog: https://gorilla.cs.berkeley.edu/blogs/10_gorilla_exec_engine... Code: https://github.com/ShishirPatil/gorilla/tree/main/goex

Congratulations on the release! I have interests to integrate concepts of GoEX to my applications. What's some of the main focus for GoEX to improve, and make GoEx enterprise-grade and how far is the system away from getting there?

Congrats!! Such brilliant ideas and amazing work! Can't wait to see it being deployed in more applications. Also, second to what polarbear-c said, a side-by-side comparison with other LLM agent systems would be great!

Very interesting to see GoEX the concept of “post-facto LLM validation," and concept of damage confinement, allowing reversible actions and limiting risk in real-world applications. Can't wait to try out the demos

Curious to see what are the common failure modes that the GoEX have, so that I can understand how the post-facto LLM validation has compared to those of pre-facto LLM validation

Exciting work! Congrats! Hope to see a side-by-side comparison for LLM agent systems to see more clearly the advantages and limitation that GoEX offers.

Excited to see what kind of agents can be built given an execution framework.

Exciting contributions to LLM tool uses; Can't wait to try it in Slack.