Voyager: An Open-Ended Embodied Agent with Large Language Models

https://arxiv.org/abs/2305.16291

Automatic curriculum, skill library of executable code, iterative prompting with environment feedback, execution errors, and self-verification; GPT-4. 3.3x unique items, 2.3x distance, tech-tree milestones up to 15.3x faster; 63 items in 160 iterations; GPT-4 15x cost of GPT-3.5; GPT-3.5 code gen 5.7x fewer items; hallucinated tasks (copper sword) and nonexistent API calls.

LLM-written skill libraries (Voyager)