framework
active
framework:automated-interpretabilityAutomated Interpretability
Method using large language models (Claude) to generate and test explanations of features at scale
Neighborhood — ranked by edge-count
Thinkers (1)
thinker
- Steven BillsextendsDeveloped automated interpretability approach using LLMs to explain neuron activations
Methods (1)
method
- Importance Scoringassociated_withWeighted Spearman correlation that corrects for sampling bias in automated interpretability evaluation
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.
- The capability to explain model predictions; a central theme of the paper, with disruption profiles as vehicle.
- Proposed paradigm for evaluating interpretability work through empirical falsifiability rather than benchmarks or user studies
- Using Claude 3 Opus to generate feature explanations and predict held-out activations.
- Advantage of DiffLogic CA over NCA — learned rules are pure binary logic circuits that can be visualized and analyzed
- Cases where subspace interventions change model behaviour through parallel pathways rather than the target feature
- 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
- An interpretability paradigm that explains computation in the model's own terms, rather than imposing top-down abstractions; VPD aims to realize this.