method
active
method:oracle-lensOracle lens
RL-trained phrase-proposal lens for arbitrary-length concepts.
Neighborhood — ranked by edge-count
Papers (1)
paper
- Verbalizable Representations Form a Global Workspace in Language Modelsanalogous_tointroduces
Methods (1)
method
- Template lensusesMulti-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.
- Gurnee et al.'s interpretability method used to identify the J-space workspace.
- Unsupervised interpretability technique that projects activations through unembedding matrix; provides comparison point for NLA approach.
- The view of AI as a question-answer system optimized for correctness, often inherited from supervised learning.
- Supervised method training models to answer questions about activations; NLAs differ by being unsupervised.
- Using interpretability for scientific knowledge discovery from trained models.
- Baseline lens that skips ahead to output.
- Density-based clustering used within spectral coarse-graining approach.