method
active
method:logit-lensLogit Lens
Unsupervised interpretability technique that projects activations through unembedding matrix; provides comparison point for NLA approach.
Neighborhood — ranked by edge-count
Papers (9)
paper
Thinkers (1)
thinker
- nostalgebraistintroducesAuthor of LessWrong post deriding GPT-3 evaluations, advocate of ecological evaluation.
Frameworks (1)
framework
- Three-step approach to identify roles of network components.
Findings (1)
finding
- Cited to mechanistically support why the contemplative prompt changes what post-training-shaped final layers allow through
Concepts (3)
concept
- Residual StreamaboutProposed pathway flowing through layers at each position; calculates K/V values that feed horizontal information flow.
- Emotion vectorsaboutLinear representations of emotion concepts extracted from activations.
- predictive integrationimplementsThe mid-to-late layer computational process that converts routed perturbation signals into explicit predictions
Methods (4)
method
- Natural Language Autoencoders (NLAs)associated_withCore unsupervised method for generating natural language explanations of LLM activations through a verbalizer-reconstructor pair trained with RL.
- Jacobian LensextendsGurnee et al.'s interpretability method used to identify the J-space workspace.
- Tuned lensextendsBaseline lens that skips ahead to output.
- Baseline from Minder et al. combining Patchscopes, logit lens and steering.
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.
- Training-free monitor that token-forces an eval-aware sentence and computes its probability at each CoT sentence boundary.
- Primary self-report measure: probability-weighted expected value over all ten digit-token logits, yielding a continuous rating that preserves full distributional signal
- Logit lens decodes probe directions to cheating/illicit/dishonest tokens though absent from synthetic datafinding0.743Probes encode cheating concept.
- Samples from interpolated logits of original and abliterated models at α.
- Dot product of summed decoder vector with unembedding vector, following Hanna et al.
- Prior method extrapolating beyond the original model in logit space to surface rare behaviors.
- Computing each feature's linear effect on output token logits via path expansion through MLP output weights and unembedding matrix
- RL-trained phrase-proposal lens for arbitrary-length concepts.