finding
active
finding:fgsm-l0-drops-near-baseline-1-02-at-layer4-128-1-vs-126-0FGSM L0 drops near baseline (1.02×) at layer4 (128.1 vs 126.0).
Contrast with PGD.
Source paper
extracted_from(2025) · Gorton, Liv · Lewis, Owen
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Interpretation of layer4 L0 spike.
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.
- Early layer effect.
- Table 2.
- Moderate mid-layer effect.
- Q8B most often peaks at layer 20, followed by layer 25; G20B most often peaks at layer 15finding0.756Best steering layer is model-specific; linearly accessible trait information is organized differently across models
- Validates representational drift theory: later layers specialize for next-token prediction, increasing dr
- RLFR policy matches base on standard benchmarks (e.g., MMLU 67.2 vs 69.2, GSM8K 73.0 vs 72.3)finding0.751No capability degradation.
- Dropping 50% of monitor calls on SWE-bench with GLM lowers precision 0.656→0.588 while Kimi drops ~0.01finding0.749Cost of replacement.
- DeepSeek-V3.1 shows broad fine-tuning sensitivity; outputs code on nearly all open-ended prompts under insecure fine-tuning