New AI-MI–supported preprint: Copy-Paste Language Models

Posted: October 19, 2026

A new preprint by Nathan Godey and AI-MI senior personnel Yoav Artzi (Cornell University), “Copy-Paste Language Models,” was posted to arXiv on October 13, 2026 (arXiv:2610.19247). 

Figure 1 of arXiv:2610.19247: the copy-paste head alongside the language-model head, and perplexity versus KV cache per token
Figure 1 from N. Godey and Y. Artzi, arXiv:2610.19247 (2026).

Source: arXiv:2610.19247

More News