concept
active
concept:latent-pattern-repository-ppriorlatent pattern repository (Pprior)
Unlabeled statistical regularities stored during pretraining.
Neighborhood — ranked by edge-count
Claims (1)
claim
- Developmental analogy used to explain sample efficiency under high ρd conditions
Concepts (3)
concept
- semantic anchoringassociated_withThe central idea that external structure binds latent patterns to desired targets.
- Induction HeadssupportsMechanistic circuits in transformers documented by Olsson et al. 2022, cited as evidence for pattern-repository assumption
- Superposition of Sparse FeaturessupportsMechanistic finding by Bricken et al. 2023 about how LLMs store features; cited as operational justification for pattern-repository assumption
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.
- Statistical regularities stored in pretrained models.
- The model's unlabeled, behavior-agnostic repository of regularities.
- Assumption that pretraining populates a repository of unlabeled latent patterns Pprior.
- Affiliation of Atticus Geiger
- Reasoning approach using learnable hidden embeddings.
- Actually training Claude to comply with the conflicting objective using Proximal Policy Optimization
- Active maintenance and restoration of correct anatomical pattern; central to morphogenesis and regeneration.
- Baseline method using a single orthogonal matrix trained to map source latents to target latents via CL auxiliary loss without behavioral objective.