arXiv - CSCL: "'Rarely' a problem? Language models exhibit inver…" - Qoto Mastodon

arXiv - CSCL @arxiv_cscl@qoto.org

'Rarely' a problem? Language models exhibit inverse scaling in their predictions following 'few'-type quantifiers. (arXiv:2212.08700v1 [cs.CL])

http://arxiv.org/abs/2212.08700 #arXiv #NLProc

Dec 20, 2022, 03:06 · · arxiv-cscl · · ·

Sign in to participate in the conversation