finding
active
finding:pca-of-llama3-1-8b-layer-19-activations-shows-smooth-curved-geometries-for-age-temperature-day-color-years-and-othersPCA of Llama3.1-8B layer-19 activations shows smooth curved geometries for age, temperature, day, color, years, and others
Evidence that manifolds exist in LLM representations.
Source paper
extracted_from(2026) · Usha Bhalla · Thomas Fel · Can Rager · Sheridan Feucht +8
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Manifold structure is present in representations and causally influences downstream behaviorsupportsInterpretation of steering results.
Frameworks (1)
framework
- Linear Representation HypothesiscontradictsThe hypothesis that models internalize concepts as approximately linear directions in representation space; used to interpret MDS injection behavior
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.
- Primary visual evidence for linear truth representations in large LLMs
- Optimal activation capping layers for Llama 3.3 70B are layers 56-71 (out of 80) at 25th percentile capfinding0.786Specific implementation finding for Llama capping parameters
- Connects this study's results to Schrimpf et al. 2021 and Caucheteux et al. 2022/2023 findings on brain-LLM alignment.
- Key empirical result showing that optimizing for behavioral outputs and fitting representation geometry produce the same path in activation space.
- Case of misalignment showing that the truth direction is not always shared between a dataset and its negation in smaller models
- One of the most promising cases; approximately corresponds to the 2/3 layer of LLaMA3.1-8B.
- The specific Fourier feature periods identified confirm base-10 rather than modular computation
- At Llama layer 26 (~81% depth) the predictive component does not dominate the latent norm.finding0.773Fig. 14.