Decompose parameters, not activations

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

• Parameter Decomposition (vs Activation)concept0.835
• Activation decompositionconcept0.800
  The conventional approach (e.g., SAEs, transcoders) of decomposing activations into interpretable features.
• VPD decomposes parameters, not activations, flipping the standard SAE / activation-patching paradigm.claim0.797
  Core proposition of the paper: a substrate-level critique of existing interpretability methods.
• Sparse autoencoders don't provide a comprehensive solution because they decode activations, not parametersclaim0.791
  Critique of activation-based interpretability methods.
• Activationsconcept0.757
  Internal representations of the model on which probes operate; the method uses activations to rank datapoints.
• Activation Compressionconcept0.753
  Key capability: covariance pooling compresses gigabytes of activations into compact stable embeddings without large labeled datasets.
• Parameters did not change; the binding did.quote0.746
  Emphasizes that anchoring is a binding process without weight updates.
• Activation-based interpretability does not immediately explain the computations that gave rise to activations; understanding parameters is necessary for deeper insightclaim0.735
  Motivates shift from studying model activations ('thoughts') to understanding parameters ('the computations themselves').