#68 · Primary category: Robotics & Embodied AI Simulation

GITM

artificial-intelligence ghost-in-the-minecraft gitm llms minecraft reinforcement-learning

Ghost in the Minecraft: Generally Capable Agents for Open-World Environments via Large Language Models with Text-based Knowledge and Memory

Project last updated:06/05/23

GitHub Stars

643

Forks

25

Contributors

2

License

Other

Why we included this project

GITM is a research framework that gets a large language model to play Minecraft using only text-based knowledge and memory, with no GPU training at all. That last point is the interesting one: the authors report unlocking the full Overworld technology tree and a 67.5% success rate on the ObtainDiamond task, all on a single CPU node. For anyone working on embodied agents or long-horizon planning, the codebase is small enough to read through and see how the LLM, a structured knowledge base, and a memory module combine into step-by-step actions. It is a research artifact and a source of ideas rather than something you would deploy, since it is tied to Minecraft and to specific LLM API choices.

Articles for this project

No articles for this project yet.

To suggest a topic or contribute an article, contact us.

Related projects in this category