Voyager: An Open-Ended Embodied Agent with Large Language Models
What this work does
An LLM proposes exploration tasks, writes executable skills and refines them from environment/error feedback; validated skills enter a growing retrieval library.
Connection to discovery
A model of persistent skill accumulation and endogenous task generation that can be transferred to scientific tool use.
Evidence
Inspected skill-commit conditions and Minecraft tech-tree/map-coverage comparisons, including the skill-library ablation.
Scope and limitations
Minecraft has fixed mechanics and strong pretrained priors. Code-memory accumulation is distinct from updating the LLM's weights and from discovering physical laws.
THE RESEARCH CONVERSATION
Discussion & reactions
Share a question, an insight, or a replication experience. Sign in with GitHub to join the conversation.