finding
active
finding:on-a-tinystories-example-tortuosity-is-77-109-for-saes-and-activations-versus-2-8-for-the-predictive-codeOn a TinyStories example, tortuosity is τ≈77–109 for SAEs and activations versus τ=2.8 for the predictive code.
Fig. 5.
Source paper
extracted_from(2025) · Lubana, Ekdeep Singh · Rager, Can · Hindupur, Sai Sumedh R. · Costa, Valerie +12
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Interpretation of smooth predictive trajectories.
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.
- Fig. 26.
- Fig. 10.
- Fig. 7c over 50 GPT-5 stories.
- Basic SAE performance metrics.
- Fig. 7d,e.
- SAEs assume sparsity (number of concepts) is time-invariant, drawn i.i.d. from a fixed distribution.claim0.755Corollary 4.1.1.
- MDL result at 20% distortion.
- Fig 2 right.