method
active
method:co-activation-probability-toy-modelCo-activation probability toy model
Variant toy model where latents from a common pool have co-activation probabilities, yielding symmetric exclusive features.
Neighborhood — ranked by edge-count
Papers (1)
paper
- Insights on Crosscoder Model Diffingintroducesmentions
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.
- Supports the co-activation explanation of symmetry.
- Pearson correlation of feature activations across 40M tokens used to measure feature similarity and universality across models
- Validates toy model.
- Toy model predictions are consistent with empirical observations on polysemanticity in real networksclaim0.728Validation; geometry and dynamics less certain.
- Success in cognitive science motivates the approach.
- Core statistical quantity in the paper's formal model: probability of two observations occurring within a time window
- Conceptualization of pain perception as inference over hidden nociceptive causes, from Eckert et al. 2022
- Choosing among candidate models based on model evidence.