Larger SAEs contain features for concepts not captured in smaller SAEs, indicating improved coverage.

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

• Feature splitting occurs: smaller SAE features split into multiple finer-grained features in larger SAEs.claim0.809
  Observed across SAE scales, e.g., 'San Francisco' split into 11 features.
• SAEs can surface features relevant to meta-cognitive monitoring, not just object-level content representationclaim0.797
  Extension of mechanistic interpretability findings to the metacognitive domain
• SAE features tend to shatter manifolds into many small and apparently-unrelated pieces, obscuring the overarching semantic structure.claim0.784
  Core critique of sparse autoencoders: they break the geometric structure of representations, making it harder to see the big picture.
• SAE features can be grounded in clinical taxonomy (abnormality, age, sex, medication) to benchmark monosemanticity and entanglement.claim0.781
  Claim that feature grounding enables interpretability metrics.
• Our SAEs' features are more interpretable than neurons.claim0.775
  Automated interpretability and specificity ratings show SAE features are clearer than MLP neurons.
• Features can be used to steer large models.claim0.767
  Clamping feature activations causally alters model behavior in interpretable ways.
• SAE features that the model self-describes as more emotional tend to be more persistent than variance-matched SAE features.claim0.757
  Novel finding that agentic self-evaluation of emotionality correlates with feature persistence
• SAE features generalize to images despite training only on text, indicating out-of-distribution robustness.claim0.757
  A promising property for interpretability analysis off-distribution.