method
active
method:tuned-lensTuned lens
Baseline lens that skips ahead to output.
Neighborhood — ranked by edge-count
Papers (2)
paper
Methods (3)
method
- Logit LensextendsUnsupervised interpretability technique that projects activations through unembedding matrix; provides comparison point for NLA approach.
- Jacobian LenscitesGurnee et al.'s interpretability method used to identify the J-space workspace.
- Template lensanalogous_toMulti-token extension via whitened mean-activation templates.
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.
- Parameter updates that reduce mismatch dr; another anchoring variant in UCCT.
- Tuned lens skips to output answers and is best next-token predictor while J-lens has highest KL.finding0.736Comparison.
- First post-training stage; shown to suppress only Impolite persona while boosting others
- The patient, hand-guided adjustment of shape and dimension to each unique condition in a building; requires materials that make it economical and easy.
- Method of rotating dataset examples to show gradual response falloff and orientation tiling across a neuron family
- The literature documenting how fine-tuning can compromise safety alignment even without malicious intent
- RL-trained phrase-proposal lens for arbitrary-length concepts.