New benchmark tests AI reasoning in realistic problem-solving

Original: GIM: Evaluating models via tasks that integrate multiple cognitive domains

Source: arxiv.org ↗