claim
active
claim:the-sae-scaling-model-mirrors-the-quanta-model-of-neural-scaling-with-features-as-quantaThe SAE scaling model mirrors the quanta model of neural scaling with features as quanta
Link to prior work.
Source paper
extracted_from(2025) · Michaud, Eric J. · Gorton, Liv · McGrath, Tom
Neighborhood — ranked by edge-count
Papers (1)
paper
Frameworks (1)
framework
- Features 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.
- Claim that feature grounding enables interpretability metrics.
- Sweeping number of features and training steps to find compute-optimal SAE configurations.
- Automated interpretability and specificity ratings show SAE features are clearer than MLP neurons.
- Discrete case.
- Unsure.
- Extension of mechanistic interpretability findings to the metacognitive domain
- Pivotal shift that created the faithfulness problem.
- Loss follows a power law in the number of latents.