This makes me think the first to make a really successful model will be the ones to pony up to get human feedback at scale from the smartest humans instead of what they do now.