Training
Scaling Instruction-Finetuned Language Models
Google·October 20, 2022
Hyung Won Chung, Le Hou, Shayne Longpre
View on arXivTL;DR
Shows that fine-tuning on many tasks phrased as instructions broadly improves zero-shot ability.
Why it matters
Helped establish instruction tuning as a core step in making models genuinely useful.