BabaIsAI

Rule-manipulation puzzle benchmark testing compositional reasoning.

Install:

pip install -e ".[babaisai]"

Paradigm:

Single-agent

Stepping:

SINGLE_AGENT

Environment	Description
BabaIsAI-Default-v0	Default rule-manipulation puzzle

Citation

@inproceedings{cloos2024babaisai,
  author       = {Solim LeGris and Wai Keen Vong and Brenden M. Lake and Todd M. Gureckis},
  title        = {Baba Is AI: Break the Rules to Beat the Benchmark},
  booktitle    = {ICML 2024 Workshop},
  year         = {2024},
}