Human-like object concept representations emerge naturally in multimodal large language models