Fetching the paper…

U-shaped and Inverted-U Scaling behind Emergent Abilities of Large Language Models · Around