claim
active
claim:one-step-at-a-time-replacement-bounds-how-far-transcoder-mechanisms-can-divergeOne-step-at-a-time replacement bounds how far transcoder mechanisms can diverge
Partial mitigation of unfaithfulness.
Neighborhood — ranked by edge-count
Papers (1)
paper
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.
- Discussion.
- Number of attention head paths between transcoder features grows exponentially with layer separationclaim0.727Why checkpointing is needed.
- Defines the faithfulness category.
- Why should we believe our transcoders learn the same mechanisms as the original underlying model?question0.724Motivating question.
- Empirical result demonstrating VPD's efficiency advantage in parameter decomposition.
- Footnote 2.
- Cross-layer transcoder features form a mix of local, global and intermediate features by decoder normfinding0.716Preliminary decoder norm analysis.
- Circuit simplification theory.