The way we will create a future where powerful AI is open-source and accessible to all is by making AI radically more efficient, both in terms of inference compute and (more importantly) in terms of training data requirements. That is what

→ View original post on X — @fchollet