concept
active
concept:monosemantic-neuronsMonosemantic neurons
Neurons responding to a single feature.
Neighborhood — ranked by edge-count
Papers (1)
paper
- Toy Models of Superpositionmentions
Concepts (1)
concept
- Polysemantic Neuronassociated_withrelated_toA neuron that responds to multiple unrelated inputs, posing a major challenge for circuit-level interpretation
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.
- Interpretability property where a latent feature represents a single semantic concept; benchmarked across architectures.
- Features that correspond to a single semantic concept and are effective for steering behavior.
- The phase diagram explains why neurons are sometimes monosemantic and sometimes polysemanticclaim0.792Theory contribution.
- The sparse set of 28 neurons at layer 18 identified as responsible for Fourier feature computation across all cyclic tasks
- Precise characterization of why polysemanticity poses a combinatorial obstacle to circuit analysis
- We hypothesize that polysemantic neurons may be resolvable by unfolding networks or training to avoid polysemanticity.hypothesis0.756Forward-looking proposal for how the polysemanticity challenge to circuits research might be overcome
- Central claim of the paper, supported by detailed feature analysis, human evaluation, automated interpretability of activations, and automated interpretability of logit weights
- Identified gap linking polysemanticity challenge to disentangled representations literature