question
active
question:why-is-it-that-neurons-sometimes-align-with-features-and-sometimes-don-tWhy is it that neurons sometimes align with features and sometimes don't?
Motivating question.
Neighborhood — ranked by edge-count
Papers (1)
paper
- Toy Models of Superpositionmentions
Findings (1)
finding
- Neuron-level phase change.
Claims (1)
claim
- Explanation for neuron interpretability.
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.
- Extends convergence argument to brain-machine alignment
- Superposition hypothesis: neural networks represent more features than dimensions using almost-orthogonal directions.hypothesis0.770Explanation for why dictionary learning can recover many more features than dimensions.
- First of three speculative claims forming the foundation of the circuits research agenda
- Universality hypothesis to be tested across families.
- Precise characterization of why polysemanticity poses a combinatorial obstacle to circuit analysis
- Claim from footnote 3, acknowledging neuron-level interpretability while arguing subcomponents are better.
- The phase diagram explains why neurons are sometimes monosemantic and sometimes polysemanticclaim0.750Theory contribution.
- Superposition hypothesis: networks represent more features than neurons as almost-orthogonal directionshypothesis0.750Central hypothesis.