method
active
method:automated-interpretability-pipelineAutomated interpretability pipeline
Claude-based explanation of features from max-activating examples.
Neighborhood — ranked by edge-count
Papers (2)
paper
- Understanding and Steering Llama 3 with Sparse Autoencoders - Goodfireintroducesmentions
Concepts (1)
concept
- Interpreter agentsextendsProposed AI scientists that hypothesize and experiment on features.
Methods (4)
method
- Using Claude 3 Opus to generate feature explanations and predict held-out activations.
- Stringent test distinguishing a feature from its 10 nearest embedding neighbours with one true and three distractor examples, repeated ten times.
- Scores explanations by distinguishing activating from distractor samples; weakly penalises polysemanticity.
- Detection ScoreusesLLM judge classification accuracy measuring interpretability; threshold 0.8.
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.
- Method using large language models (Claude) to generate and test explanations of features at scale
- The capability to explain model predictions; a central theme of the paper, with disruption profiles as vehicle.
- Socio-philosophical open question.
- Detection-based LLM judging of latent descriptions using gpt-4o-mini.
- Proposed paradigm for evaluating interpretability work through empirical falsifiability rather than benchmarks or user studies
- Advantage of DiffLogic CA over NCA — learned rules are pure binary logic circuits that can be visualized and analyzed
- Programming technique to restructure a fine-grained Linda program for efficiency by replacing live data structures with passive ones and coarser-grain processes.
- The field aimed at understanding what neural networks have learned; characterized as pre-paradigmatic in this paper