notesum.ai

Published at October 30

All or None: Identifiable Linear Properties of Next-token Predictors in Language Modeling

stat.ML
cs.AI
cs.CL
cs.LG

Released Date: October 30, 2024

Authors: Emanuele Marconato1, Sébastien Lachapelle2, Sebastian Weichwald, Luigi Gresele3

Aff.: 1University of Trento & University of Pisa; 2Samsung - SAIT AI Lab, Montreal; 3University of Copenhagen

Arxiv: http://arxiv.org/abs/2410.23501v1