method
active
method:automated-interpretability-llm-judgeAutomated Interpretability (LLM judge)
Detection-based LLM judging of latent descriptions using gpt-4o-mini.
Neighborhood — ranked by edge-count
Papers (1)
paper
Frameworks (1)
framework
- SAEBenchusesEvaluation suite of eight metrics across four capabilities.
Methods (2)
method
- LLM judge evaluationrelated_toUsing Claude Sonnet 4 as a grader to categorize model responses according to predefined criteria.
- Using Claude 3 Opus to generate feature explanations and predict held-out activations.
Datasets (1)
dataset
- The PileusesTraining corpus used for the 67M-parameter model tested with VPD.
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
- Evaluation framework using an LLM (GPT-4.1-mini) to score trait expression and coherency
- Alternative data attribution approach using an LLM as a judge; compared against the probe-based method.
- Baseline comparison for data attribution; outperformed by probe-based approach.
- An LLM-based classifier that returns 1 if response contains a clear subjective experience report and 0 otherwise
- Caveat on metric.
- The capability to explain model predictions; a central theme of the paper, with disruption profiles as vehicle.