hypothesis
active
hypothesis:sdl-reconstruction-error-might-be-reduced-by-improved-sdl-training-or-new-decomposition-methodsSDL reconstruction error might be reduced by improved SDL training or new decomposition methods
Structured errors imply potential solutions.
Source paper
extracted_from(2025) · Lee Sharkey · Bilal I. Chughtai · Joshua Batson · Jack W. Lindsey +25
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (1)
finding
- SAE reconstruction error direction and norm are largely linearly predictable from the activation vectorassociated_withsupportsErrors are structured not random.
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.
- Core assessment of SDL.
- Open question A.1.1a.4.
- Ethical implication about the nature of AI training experience if the thesis holds
- Quantifies SAE reconstruction damage.
- Author interpretation.
- Limitation statement.
- Justification for why the model organism is a realistic test case for studying steering.
- A promising property for interpretability analysis off-distribution.