method
active
method:automated-interpretability-pipeline-using-llmsAutomated interpretability pipeline using LLMs
Using Claude 3 Opus to generate feature explanations and predict held-out activations.
Neighborhood — ranked by edge-count
Papers (1)
paper
Methods (2)
method
- Automated interpretability pipelinerelated_toClaude-based explanation of features from max-activating examples.
- Automated Interpretability (LLM judge)related_toDetection-based LLM judging of latent descriptions using gpt-4o-mini.
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.
- Claude 3 Opus ratings aligned with human judgment of feature descriptions.
- Method using large language models (Claude) to generate and test explanations of features at scale
- Claude rates coherence and behavior on 0-100 scale, called centiClaude.
- The capability to explain model predictions; a central theme of the paper, with disruption profiles as vehicle.
- Caveat on metric.
- Socio-philosophical open question.
- Establishes that the observed linear structure is not merely a representation of text probability
- Simple interpretability methods scale to 2.8T-parameter models and long-context agentic rolloutsclaim0.740Scaling claim.