ARC-3 Benchmark Focuses on Causal Modeling and Adaptation

The ARC-3 benchmark tests qualitative AGI properties including exploration under uncertainty, adaptation without instructions, and causal world modeling from limited data, though it is not intended as a definitive AGI proof.