Cross-layer superposition

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

• Superpositionconcept0.839
  Phenomenon where models represent more features than dimensions via almost-orthogonal directions.
• Cross-layer superposition is a fundamental challenge for dictionary learning.claim0.824
  Features smeared across layers cannot be fully disentangled by SAE on a single residual stream.
• Superposition Hypothesisframework0.802
  Core theoretical framework: neural networks represent more features than neurons by encoding features as directions in superposition
• Superposition in Neural Networksconcept0.799
  Theoretical model of how neural networks encode more features than dimensions, informing linear representation work.
• Superposition of Simulacraconcept0.768
  The state in which a dialogue agent maintains multiple possible characters simultaneously, refined as the conversation proceeds
• Superposition in Residual Streamconcept0.755
  The phenomenon where the residual stream communicates many more features than its dimensionality by encoding information across overlapping subspaces
• Simulacra in Superposition Frameworkframework0.750
  The more nuanced second metaphor: LLM as simulator maintaining a superposition of possible simulacra across a multiverse of characters
• Memorization in Superpositionconcept0.747
  Specific phrases or sequences memorized via binary features in superposition, enabling narrow pattern matching despite few neurons