Canonical event

Scaling instruction finetuning improves generalization across large language models

Google researchers published results scaling instruction finetuning across many tasks, model sizes and chain-of-thought data, with public Flan-T5 checkpoints.