作者位次 #7/10ACL 2026Poster· Natural Language Processing
Language Models Learn Universal Representations of Numbers and Here’s Why You Should Care
Michal Štefánik · Timothee Mickus · Marek Kadlčík · Bertram Højer · Michal Spiegel · Raúl Vázquez · Aman Sinha · Josef Kuchař · Philipp Mondorf · Pontus Stenetorp
摘要
Prior work has shown that large language models (LLMs) often converge to accurate input embedding for numbers, based on sinusoidal representations.In this work, we demonstrate that these representations are in fact strikingly systematic, to the point of being almost perfectly universal: different...