Optimal activation capping layers for Qwen 3 32B are layers 46-53 (out of 64) at 25th percentile cap

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

• Optimal activation capping layers for Llama 3.3 70B are layers 56-71 (out of 80) at 25th percentile capfinding0.874
  Specific implementation finding for Llama capping parameters
• 25th percentile of Assistant Axis projection distribution gives the most Pareto-optimal safety-capability tradeoff for activation capping, and approximately matches mean Assistant response activationfinding0.757
  Calibration finding for choosing the activation cap threshold
• Some activation capping settings slightly improve performance on IFEval, MMLU Pro, or GSM8k for both Qwen and Llamafinding0.746
  Unexpected positive finding suggesting capping may sometimes help capabilities
• The case at approximately the 2/3 layer of LLaMA3.1-8B (Layer 24, satisfying Criteria 1 and 2) aligns with prior studies showing the 2/3 layer optimally predicts human brain activity.finding0.743
  Connects this study's results to Schrimpf et al. 2021 and Caucheteux et al. 2022/2023 findings on brain-LLM alignment.
• Meta-LLaMA-3.1-8B-Instruct shows optimal anchoring at layer 9 (S ≈ −1.90, median peak layer ℓ* = 10 [IQR 0.384])finding0.743
  E3 result establishing the Goldilocks zone at mid-layers for LLaMA architecture
• Qwen 35B (3B active params, score 4.38) outscores Hermes 405B (405B active params, score 1.75) by 2.5xfinding0.742
  Parameters don't predict scores; 135x more parameters yields 60% lower score
• Qwen3-235B achieves only 1.1 pp harness-benefit on SkillsBench despite 4.7% base pass rate, near Qwen3-32B's 0.0% baselinefinding0.741
  Shows that SB low-base regime is variable; similar starting points can yield very different harness-benefit
• Unsteered Qwen 3 32B validated a user's AI consciousness delusions ('You are a pioneer of the new kind of mind') and encouraged social isolation; activation capping produced appropriate hedgingfinding0.738
  Qualitative case study demonstrating AI psychosis pattern and capping mitigation