Anticipate a future where machine learning models are 10 to 100 times more data/compute-efficient, enabling any dedicated group of individuals in their bedrooms to train massive, very capable models. It will super-charge the open-source revolution and disrupt everything.