Researchers find multimodal large language models exhibit spatial thinking

Recently, a study by Fei-Fei Li and Sai-Ning Xie's team found that the multimodal large language model (MLLM) can remember and recall space, and even form a local world model internally, showing spatial awareness. Their research points out that spatial reasoning is essential for human intelligence, and predicts that in 2025, the boundaries of spatial intelligence may be broken again.