finding
active
finding:neurons-in-abs-model-have-a-primary-feature-with-large-weight-plus-secondary-features-with-small-weights-resembling-language-model-neuronsNeurons in abs model have a primary feature with large weight plus secondary features with small weights, resembling language model neurons
Mirrors real networks.
Neighborhood — ranked by edge-count
Papers (1)
paper
- Toy Models of Superpositionmentions
Claims (1)
claim
- Toy model predictions are consistent with empirical observations on polysemanticity in real networkssupportsValidation; geometry and dynamics less certain.
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.
- Working definition adopted by the paper.
- Without superposition.
- Demonstrates that the Arabic feature is not aligned to any single neuron
- Sloman's critique of mainstream neural network theories.
- The central hypothesis of the paper; the platonic representation hypothesis itself
- Data-induced competition over resources.
- Core framing hypothesis; analogy not to be taken literally.
- Superposition hypothesis: neural networks represent more features than dimensions using almost-orthogonal directions.hypothesis0.758Explanation for why dictionary learning can recover many more features than dimensions.