VPD enables manual model editing through direct parameter manipulation.

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

• The ability to make precise edits demonstrates that VPD identifies real computational machineryclaim0.846
  Claim that editing success validates VPD's decomposition.
• Manual model editing through parameter manipulationconcept0.816
  Application enabled by VPD: direct manipulation of weight matrices for interpretable model modification.
• VPD decomposes parameters, not activations, flipping the standard SAE / activation-patching paradigm.claim0.794
  Core proposition of the paper: a substrate-level critique of existing interpretability methods.
• VPD scales to a 4-layer 67M-parameter model trained on The Pile.finding0.792
  Empirical demonstration of VPD on a mid-scale transformer, establishing feasibility.
• The VPD-based edit has similarly low off-target effects as uninterpretable fine-tuning methodsfinding0.781
  Performance comparison showing subcomponent editing is comparable to fine-tuning in preserving off-target behavior.
• VPD (adVersarial Parameter Decomposition)concept0.778
  Core methodological framework introduced in this paper; decomposes weight matrices into rank-one interpretable subcomponents using adversarial ablations.
• VPD is a meaningful step toward bottom-up interpretabilityclaim0.776
  Positioning of VPD as advancing the paradigm of explaining computation in the model's terms.
• VPD can be arbitrarily applied to any neural network architectureclaim0.759
  Claim of generality, highlighted as a key strength.