arXivJan Tempus, Philip Whittington, Craig W. Schmidt, Dennis Komm, Tiago PimentelThu, May 21, 2026, 10:59 AM PDT
score 14.8
New algorithm builds better text tokenizers using mathematical optimization
Original: Tokenisation via Convex Relaxations
Source: arxiv.org ↗
Writing ELI5 summary…