framework
active
framework:sae-latent-allocation-modelSAE latent allocation model
Reduces SAE optimization to allocating N latents across features.
Neighborhood — ranked by edge-count
Papers (1)
paper
Methods (1)
method
- Derives regimes.
Frameworks (3)
framework
- Models allocate capacity among power-law-distributed data manifolds.
- Activations are sums of sparse features living in subspaces.
- Quantization (quanta) model of neural scalinganalogous_toFeatures analogous to quanta.
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.
- Interpretable features extracted by sparse autoencoders used as steering targets in this study
- Adding a multiple of the SAE latent decoder vector to token activations to causally test each latent's role in misalignment
- Using GPT-4o or o3 to automatically generate interpretations of SAE latents from top-activating examples
- Pre-filtering step excluding latents naturally activated by each prompt to ensure genuine off-topic steering
- Interpretability method criticized in this paper for shattering manifolds into atomic pieces, obscuring overarching semantic structure.
- Standard baseline SAE operating on individual token activations.
- Hybrid SAE jointly training turn-averaged inner and per-token outer partitions.
- The individual, supposedly monosemantic directions learned by SAEs; argued here to fragment manifolds into disconnected pieces.