The paper introduces time vectors, a method to customize language models to specific time periods. By finetuning the model on data from a single time period and subtracting the weights of the original model, time vectors improve performance on text from that time period. Interpolating between time vectors allows for better performance on intervening and future time periods. The findings are consistent across different tasks, domains, model sizes, and time scales, suggesting that time is encoded in the weight space of finetuned models.

https://arxiv.org/abs//2312.13401

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