Reproduce “How to Motivate Your Dragon: Teaching Goal-Driven Agents to Speak and Act in Fantasy Worlds” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Prithviraj Ammanabrolu, Jack Urbanek, Margaret Li, Arthur Szlam, Tim Rocktäschel, Jason Weston
The authors train an agent for LIGHT that can pursue motivations through game actions and dialogue. They pair motivations with quest demonstrations and use separate policies for action and speech. In zero-shot tests on 211 held-out demonstrations, Adaptive meets both goal criteria in 0.303 of episodes, versus 0.103 for Scratch; this measures game goals and a learned speech proxy.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.