finding
active
finding:at-llama-layer-26-81-depth-the-predictive-component-does-not-dominate-the-latent-normAt Llama layer 26 (~81% depth) the predictive component does not dominate the latent norm.
Fig. 14.
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
- Layer choice discussion.
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.
- Connects this study's results to Schrimpf et al. 2021 and Caucheteux et al. 2022/2023 findings on brain-LLM alignment.
- Math and code tasks show strongest mid-layer anchoring on LLaMA (S ≈ −1.65 at layers 8-12)finding0.820Task-specific E3 finding showing compositional reasoning requires deeper processing
- App. E.
- E3 result establishing the Goldilocks zone at mid-layers for LLaMA architecture
- Core E3 finding validating S as a predictor of anchoring effectiveness
- LLaMA-3.1-8B: Sbmax = -1.896 ± 0.211, AUSN = -2.119 ± 0.198, peak layer ℓ* = 10 (median)finding0.801Seed-pooled geometry-only statistics (per-dev z units).
- Validates representational drift theory: later layers specialize for next-token prediction, increasing dr
- Layer-wise emergence pattern supporting hierarchical development hypothesis