Palash Chaudhary

Efficacy of Machine-Generated Instructions

2023 · arXiv (Cornell University)

Large "instruction-tuned" language models (i.e., finetuned to respond to instructions) have demonstrated a remarkable ability to generalize zero-shot to new tasks. Nevertheless, they depend heavily on human-written instruction data that is often limited in quantity, …