Today's best AI needs orders of magnitude more data than a human child to achieve visual competence.

The paper introduces the Zero-shot World Model (ZWM), an approach that substantially narrows this gap. Even when trained on a single child's visual experience, BabyZWM matches state-of-the-art models on diverse visual-cognitive tasks – with no task-specific training, i.e., zero-shot.

The work presents a blueprint for efficient and flexible learning from human-scale data, advancing a path toward data-efficient AI systems.

Full Twitter post: https://x.com/khai_loong_aw/status/2044051456672838122?s=20

HuggingFace: https://huggingface.co/papers/2604.10333

GitHub: https://github.com/awwkl/ZWM