Read the original at HF Daily Papers
Researchers introduce WorldCast, a benchmark containing 320 manually designed cases across eight categories, to evaluate twelve models on multi-step world state prediction tasks.
Carried by: HF Daily Papers. First seen: .