This paper argues for controlling randomization in evaluating large language models, showing that coupled autoregressive generation can yield different rankings than vanilla methods, despite fewer required samples.

https://arxiv.org/abs//2502.01754

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