> we have no way of knowing in advance what the capabilities of current AI systems will be if we are able to scale them by 10x, 100x, 1000x, and more. Scaling

> we have no way of knowing in advance what the capabilities of current AI systems will be if we are able to scale them by 10x, 100x, 1000x, and more. Scaling experiments are routinely performed (the results are not encouraging). To say we know nothing about this is wrong.