Goodfire SAE API

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

• Goodfireinstitute0.817
  AI research company; authors' affiliation; develops tools including EVEE and publishes research on genomic foundation models.
• SAE featuresconcept0.741
  The individual, supposedly monosemantic directions learned by SAEs; argued here to fragment manifolds into disconnected pieces.
• Goodfire Ember Contrastive Searchmethod0.738
  API method used to identify latents differentially activated between on-topic and off-topic prompt-response pairs
• SAE Feature Conditional Firing Persistence Metricmethod0.720
  P(feature fires at t+100 | fired at t) minus P(feature fires at t+100 | did not fire at t), used because SAE features are binary unlike probe activations
• SAE feature firing probability persistence metricmethod0.709
  Persistence metric for SAE features: P(fires at t+100 | fired at t) minus P(fires at t+100 | did not fire at t)
• Sparse Autoencoders (SAE)method0.707
  Interpretability method criticized in this paper for shattering manifolds into atomic pieces, obscuring overarching semantic structure.
• Sequential SAE Activation Analysismethod0.707
  Token-level analysis of OTD and backtracking latent activations aligned at correction points across episodes
• Sparse Autoencoders (SAE) activation-based paradigmframework0.681
  Standard interpretability approach that VPD critiques and proposes an alternative to.