New benchmark tests AI's ability to combine problem-solving strategies

Original: ClassicLogic: A Knowledge-Driven Benchmark of Classic Puzzle Games for Evaluating Compositional Generalization

Source: arxiv.org ↗