SAE Latents

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

• Auto-Interpretation of SAE Latentsmethod0.842
  Using GPT-4o or o3 to automatically generate interpretations of SAE latents from top-activating examples
• SAE Latent Steeringmethod0.829
  Adding a multiple of the SAE latent decoder vector to token activations to causally test each latent's role in misalignment
• Concreteness Filtering of SAE Latentsconcept0.790
  Pre-filtering step excluding abstract latents where off-topic detection is harder
• Sparse Autoencoders (SAE)method0.782
  Interpretability method criticized in this paper for shattering manifolds into atomic pieces, obscuring overarching semantic structure.
• SAE featuresconcept0.776
  The individual, supposedly monosemantic directions learned by SAEs; argued here to fragment manifolds into disconnected pieces.
• Latent SOO Metricmethod0.756
  Metric measuring the mean MSE between self and other-referencing activations across all hidden MLP/attention layers
• Latent Structuresconcept0.745
  Hidden or underdeveloped structures existing 'between the lines' of a configuration that can be enhanced and developed through harmony-seeking computation.
• Ten SAE latents (out of 2.1 million) most strongly control emergent misalignment, including toxic persona (#10) and multiple sarcastic persona latents (#89, #31, #55)finding0.743
  Model diffing identifies a small, interpretable set of latents responsible for emergent misalignment