claim
active
claim:many-negative-findings-about-saes-instability-brittle-steering-hard-automated-interpretation-reflect-a-mismatch-between-direction-based-interpretation-and-nonlinear-geometryMany negative findings about SAEs (instability, brittle steering, hard automated interpretation) reflect a mismatch between direction-based interpretation and nonlinear geometry
Geometric explanation for prior negative results.
Source paper
extracted_from(2026) · Usha Bhalla · Thomas Fel · Can Rager · Sheridan Feucht +8
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (2)
claim
- Explanation of SAE instability.
- Explanation of brittle steering.
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.
- A critical failure mode identified in the paper demonstrating risk of naïve concept steering
- If agentic self-steering evaluation proves robust, it might be used to better explain and interpret SAE features in generalhypothesis0.789Speculative claim about scaling introspective access to general SAE feature interpretation
- Surprising finding that the two evaluation methods diverge in their relationship with persistence
- Authors' hedged position on SAEs.
- Shows gating effect is specific to the self-referential computational regime, not a general feature effect
- Forward-looking claim about the potential of model introspection as an interpretability tool
- Forward-looking claim about the broader utility of the self-steering evaluation method
- Claim that feature grounding enables interpretability metrics.