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

2025-01-22 · ICLR 2025 Poster · anchor

Findings