Qwen3-30B-A3B-Instruct (MoE) shows several attention layers with high-contribution heads rather than a single localized layer

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

• All 32 attention heads at layer 3 achieve 100% localization accuracy for injections at layer 2 (5-way classification, 20% chance)finding0.802
  Striking mechanistic finding that injection creates universally detectable perturbation in residual stream immediately downstream
• Only three attention heads in a specific layer independently govern persona and style formation in Qwen2.5-7B and Llama-3.1-8Bclaim0.794
  Striking sparsity finding that enables surgical intervention
• Qwen 3 32B is most likely to hallucinate human personas (names, birthplaces, years of experience) when steered away from the Assistantfinding0.788
  Model-specific difference in how steered personas manifest
• Attention heads with positive projection on reflection direction are sparse and located mostly in deeper layers of DeepSeek-R1-Qwen-1.5Bfinding0.781
  Structural finding about which attention heads control reflection behavior
• Qwen3-30B-A3B-Instructconcept0.777
  Extended evaluation MoE model showing distributed persona localization across multiple layers
• Gemma-3-12b-it shows multiple sharp direction shifts in persona vectors across layers, differing from the single-transition pattern in Qwen and Llamafinding0.776
  Suggests architectural variations influence persona localization pattern
• Layer 27 (last layer) has largest projection magnitude on the reflection direction among all attention head layers in DeepSeek-R1-Qwen-1.5Bfinding0.775
  Attribution finding suggesting the last layer directly controls reflection keyword generation
• In Llama 3 70B, each next-token prediction at token 101 draws on 64,000 independent attention streams (8 heads × 80 layers × 100 prior positions), each carrying a 128-dimensional signalfinding0.773
  Quantitative argument for the richness of quasi-psychological connections enabled by attention streams