SCALING LAWS FOR NATIVE MULTIMODAL FOUNDATION MODELS

Summary from the TLDR AI newsletter

Controlled pretraining experiments that explore how text, images, and video should be combined when building multimodal models from scratch.