claim
active
claim:temporal-feature-analysis-augments-rather-than-replaces-sae-tools-novel-codes-remain-compatible-with-autointerpTemporal Feature Analysis augments rather than replaces SAE tools; novel codes remain compatible with autointerp.
§7.2 assertion.
Source paper
extracted_from(2025) · Lubana, Ekdeep Singh · Rager, Can · Hindupur, Sai Sumedh R. · Costa, Valerie +12
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (4)
finding
- Fig. 11.
- Table 3.
- TFA explains similar variance as SAEs (Story 0.73, Web 0.79, Code 0.75) versus Pred-only 0.29–0.40.supportsTable 2.
- Table 1.
Related by similarity (8)
cosine ≥ 0.65 · no typed edgeEntities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.
- Can latent codes from SAEs or Temporal Feature Analyzers capture the in-context graph phenomenon?question0.808§6.3 question.
- Interpretation of smooth predictive trajectories.
- Surprising finding that the two evaluation methods diverge in their relationship with persistence
- Explains weaker datamix SAEs.
- Interpretability tool underlying top-feature, details, steering tools.
- Standard interpretability approach that VPD critiques and proposes an alternative to.
- Out-of-distribution generalization of SAE features.
- Authors' hedged position on SAEs.