On a fast enough timeline, I think it’s possible that compute-efficiency is one of the necessary breakthroughs to be had. We consistently see small models achieving what it had taken very big ones to achieve earlier, after all.
On a slow enough timeline, I think the compute gap will narrow.
On a fast enough timeline, I think it’s possible that compute-efficiency is one of the necessary breakthroughs to be had. We consistently see small models achieving what it had taken very big ones to achieve earlier, after all.
On a slow enough timeline, I think the compute gap will narrow.