跳到正文
原文
Anthropic:Transformer Circuits(可解释性研究)·· 13 小时前AI 评分42

Anthropic 解读分布式表征:组合与叠加是两种相互竞争的策略

Distributed Representations: Composition & Superposition An informal note on how "distributed representations" might be understood as two different, competing strategies — "composition" and "superposition" — with quite different properties.

AI 导读

Anthropic 的可解释性研究笔记指出,"分布式表征"实际包含"组合"与"叠加"两种不同且相互竞争的策略,二者在泛化能力和可线性计算的函数上性质迥异,且存在根本性权衡。

来源:Anthropic:Transformer Circuits(可解释性研究) · transformer-circuits.pub