4 ms·
Maybe they think it's possible to train a better, more efficient model on the chain of thought outputs of the existing one, not just matching but surpassing it?
by schmorptron 2y ago
Maybe they think it's possible to train a better, more efficient model on the chain of thought outputs of the existing one, not just matching but surpassing it?