hypothesis
active
hypothesis:a-combination-of-crosscoders-and-flexible-decoding-matryoshka-iterative-codebook-could-resolve-the-steering-classification-tensionA combination of crosscoders and flexible decoding (Matryoshka, iterative codebook) could resolve the steering-classification tension
Proposed remedy.
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Limitations section.
Frameworks (2)
framework
- CrosscoderssupportsLearn a shared feature space across two models; basis of the work.
- Matryoshka SAEssupportsNested dictionary approach that inspires the nested loss.
Methods (1)
method
- Flexible decoding technique mentioned as possible remedy.
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.
- Translates steering vector via weighted average of shared decoder vectors.
- Demonstrates averaging multiple prompt pairs reduces noise; optimal subset selection further improves performance.
- Core mechanism statement.
- Explanation of brittle steering.
- Practical finding for optimizing steering setup.
- Addresses skeptical alternative that reports reflect only conversational content
- A block-sparse code should hold the curve manifold together where a directional code fragments it.hypothesis0.750Prediction tested on InceptionV1.