We introduce a framework for automated skill discovery for language model-based agents in open-ended environments, and show its potential toward self-evolving system.

A novel framework of training a contextualization module to help the decision-making of LLM agents achieves the super-human performance in the WebShop benchmark.

A novel benchmark that can serve as a unified testbed for mobile device control agents on performing practical daily tasks across diverse device configurations.

A newly proposed distribution (over a Riemannian manifold of the diagonal Gaussians equipped with Fisher information metric) empowers learning a hyperbolic world model.