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