BriefGPT.xyz
May, 2023
在 GPT-3 和 GPT-4 中测试词义的因果模型
Testing Causal Models of Word Meaning in GPT-3 and -4
HTML
PDF
Sam Musker, Ellie Pavlick
TL;DR
本研究通过HIPE理论评估了GPT-3和GPT-4的词汇表示，结果发现GPT-3未编码所假设的因果结构，但发现GPT-4编码了这种结构，为评估大型语言模型的表征能力做出了贡献。
Abstract
large language models
(LLMs) have driven extraordinary improvements in NLP. However, it is unclear how such models represent
lexical concepts
-i.e., the meanings of the words they use. This paper evaluates the lex
→