[Submitted on 15 Nov 2023]

Study reveals why in-context learning fails on complex specification-heavy tasks and how fine-tuning can help