Report
What does a model’s rise on a creativity leaderboard actually measure?
The authors say their paper examines long-tail performance using 15 benchmarks and 26 open-weight models.
TLDR
The authors of a new paper ask what a higher rank on a model creativity leaderboard, such as LM Arena’s creative-writing ranking, measures—and what has improved. They say their work points to long-tail performance as a shared target, drawing on 15 benchmarks and 26 open-weight models.
Combined views
—
2 Sources, first seen ago
