Alex Fang

Reusing Pre-Training Data at Test Time is a Compute Multiplier

Datasets, Documents, and Repetitions: The Practicalities of Unequal Data Quality

Language models scale reliably with over-training and on downstream tasks