Your Transformer Can Hold Two Thoughts at Once: Evidence of Linear Superposition in LLMs

  • 类型:arxiv
  • 标识:2609.29845
  • 链接:https://arxiv.org/abs/2609.29845
  • 主分类:engineering
  • 形态:position
  • TLDR:While Large Language Models (LLMs) rely on highly non-linear components, in this work we demonstrate that they exhibit fundamental linearity: when inputs from distinct text streams are linearly combined, the model outputs a superposition of the individual next-token distributions. We term this the Superposition Linearity Hypothesis. We provide evidence that superposition is an intrinsic property of the Transformer architecture rather than an emergent consequence of training; in fact, we observe that it tends to diminish as pretraining progresses. However, we demonstrate that linearity can be s
  • 待LLM分类:否
  • 来源文件:
  • /inbox/tom/_candidates/2026-09-26-agent-rag-longcontext-candidates.json