Data scarcity. Training large AI models requires correspondingly large datasets. The indexed web contains about 500T words of unique text, and is projected to increase by 50% by 2030. Multimodal ...