framework
active
framework:p-annealing-sae

P-Annealing SAE

Evaluated architecture.

Neighborhood — ranked by edge-count

Related by similarity (8)

cosine ≥ 0.65 · no typed edge

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

  • Nested SAEframework0.786
    Hybrid SAE jointly training turn-averaged inner and per-token outer partitions.
  • Decoupled SAEframework0.784
    Hybrid with separate turn-averaged and residual per-token SAEs.
  • Interpretability method criticized in this paper for shattering manifolds into atomic pieces, obscuring overarching semantic structure.
  • SAE Latentsconcept0.761
    Interpretable features extracted by sparse autoencoders used as steering targets in this study
  • SAE featuresconcept0.757
    The individual, supposedly monosemantic directions learned by SAEs; argued here to fragment manifolds into disconnected pieces.
  • Interpretability tool underlying top-feature, details, steering tools.
  • Adding a multiple of the SAE latent decoder vector to token activations to causally test each latent's role in misalignment
  • BatchTopK SAEframework0.746
    Evaluated architecture.