From the source
An experiment testing how well LLMs can fix their mistakes when given feedback in plain English, using a simple calendar API scenario.
The author built a chatbot arena using Keras, JAX, and TPUs to interact with multiple LLMs simultaneously.
