framework
active
framework:multi-dimensional-linear-representation-hypothesisMulti-dimensional linear representation hypothesis
Activations are sums of sparse features living in subspaces.
Neighborhood — ranked by edge-count
Papers (1)
paper
Frameworks (2)
framework
- Linear Representation Hypothesisrelated_toThe hypothesis that models internalize concepts as approximately linear directions in representation space; used to interpret MDS injection behavior
- SAE latent allocation modelextendsReduces SAE optimization to allocating N latents across features.
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.
- Hypothesis that information may be encoded in arbitrary non-linear subspaces of a neural network
- Open question A.1.1a.2.
- The idea that features are encoded as directions in activation space.
- The established finding that transformer LLMs encode many interpretable concepts as linear directions in activation space
- Machine-learning hypothesis (Huh et al. 2024) proposed as a candidate framework to unify with the Platonic space model.
- Position on LRH.
- Speculative interpretation of breakdown.
- The central object of study — the idea that a concept like truth is encoded as a direction in the LLM's latent space