BT

Facilitating the Spread of Knowledge and Innovation in Professional Software Development

Write for InfoQ

Topics

Choose your language

InfoQ Homepage News Minecraft Welcomes Its First LLM-Powered Agent

Minecraft Welcomes Its First LLM-Powered Agent

This item in japanese

Bookmarks

Researchers from Caltech, Stanford, the University of Texas, and NVIDIA have collaboratively developed and released Voyager, an LLM power agent that utilizes GPT-4 to engage in Minecraft gameplay. Voyager demonstrates remarkable capabilities by learning, retaining knowledge, and showcasing exceptional expertise in Minecraft.

Voyager operates autonomously, continuously exploring the virtual world, acquiring diverse skills, and making groundbreaking discoveries without any human intervention. Voyager's innovation lies in its automatic curriculum that optimizes exploration, an ever-expanding skill library for storing and retrieving complex behaviors, and an iterative prompting mechanism that incorporates environment feedback, execution errors, and self-verification for program enhancement.

Voyager consists of three key components: an automatic curriculum for open-ended exploration, a skill library for increasingly complex behaviors, and an iterative prompting mechanism that uses code as action space.

Utilizing blackbox queries to interact with GPT-4, Voyager circumvents the need for model parameter fine-tuning. The skills developed by Voyager are both temporally extended and interpretable, resulting in rapid compound growth of the agent's capabilities and mitigating catastrophic forgetting.

According to Jim Fan, one of the researchers of the project, GPT-4 experiment in Minecraft is a good place to start when creating effective AI agents. Autonomous agents with broad capabilities are the next step in artificial intelligence. They are motivated by curiosity and survival to explore, plan, and learn new abilities in open environments.

Compared with baselines, Voyager unlocks the wooden level 15.3x faster in terms of the prompting iterations, the stone level 8.5x faster, the iron level 6.4x faster, and Voyager is the only one to unlock the diamond level of the tech tree.

An unparalleled attribute of Voyager is its ability to utilize the learned skill library in a fresh Minecraft world to solve novel tasks from scratch, a feat that other approaches struggle to achieve when generalizing.

Lifelong learning agents are AI models designed to acquire knowledge and skills continuously throughout their operational lifespan. They possess the ability to adapt, learn, and improve as they encounter new information and experiences. Lifelong learning agents excel in retaining and transferring knowledge, allowing them to handle diverse tasks and domains effectively. Their capacity for continuous learning makes them valuable in various fields, including gaming, robotics, healthcare, and education.

With Voyager, Minecraft enters a new era of innovation, laying the foundation for future advancements in embodied lifelong learning agents.

About the Author

Rate this Article

Adoption
Style

BT