Parallel Structures in Pre-training Data Yield In-Context Learning

Re-assign community