Stanford AI Laboratory proposes zero-shot world models, narrowing the gap between AI and human children's visual learning data.

ME News report, on April 15 (UTC+8), the Stanford AI Lab (StanfordAILab) recently pointed out that the amount of data required for today’s most advanced AI models to achieve visual capabilities is far beyond that of human children by several orders of magnitude. To narrow this gap, researchers proposed the Zero-shot World Model (ZWM, zero-shot world model) approach. The method has made significant progress: the BabyZWM model achieved performance comparable to a benchmark that is not specified, when trained using only first-person perspective data from a single child. (Source: InFoQ)