Sbmax mean -1.896 ± 0.211

Geometry summary peak anchoring score averaged over seeds.

Source paper

extracted_from

(2025) · Edward Yi Chang · Kaya, Zeyneb N. · Ethan Chang

claim

community

Few-shot anchoring & latent structure
members_of
How minimal examples disambiguate and recruit latent arithmetic/reasoning interpretations in LLMs
Neural activation geometry and behavioral prediction
members_of
Quantifies relationships between layer-wise activation statistics (Sbmax, AUSN) and task performance metrics in LLMs, bridging internal representation geometry to behavioral outcomes.

cosine ≥ 0.65 · no typed edge

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

LLaMA-3.1-8B: Sbmax = -1.896 ± 0.211, AUSN = -2.119 ± 0.198, peak layer ℓ* = 10 (median)finding0.770
Seed-pooled geometry-only statistics (per-dev z units).
Larger Sbmax associated with smaller θ50 in E3 sweepfinding0.749
Geometry-to-behavior correlate within E3.
peak anchoring Sbmaxconcept0.722
Maximum layer-wise anchoring score across layers.
Geometry summaries (Sbmax, AUSN)method0.705
Peak anchoring (Sbmax) and normalized area under the S(ℓ) curve (AUSN) used to summarize trajectory.
Layer-wise geometry summaries (Sbmax, AUSN) predict internal few-shot thresholds θ50claim0.696
Claim that geometry-to-behavior correlates exist
AUSN mean -2.119 ± 0.198finding0.693
Normalized area under S(ℓ) averaged over seeds.
For all three SAEs (1M, 4M, 34M), average active features per token <300, and reconstruction variance explained ≥65%.finding0.680
Basic SAE performance metrics.
Mean piece diameter 15 cm, mean edge length 60 cmfinding0.679
Derived from the sample statistics to estimate cutting requirements.