The paper compares methods for selecting high-quality data for instruction fine-tuning of LLMs and finds that selecting the 1,000 longest responses consistently outperforms sophisticated methods.

https://arxiv.org/abs//2402.04833

YouTube: https://www.youtube.com/@ArxivPapers

TikTok: https://www.tiktok.com/@arxiv_papers

Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-papers/id1692476016

Spotify: https://podcasters.spotify.com/pod/show/arxiv-papers