Random head ablation yielded no reduction in trait scores, while Style Modulation Head ablation caused rapid decline

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

• Sequential zero ablation of Style Modulation Heads (layers 20→15→19 in Qwen2.5-7B) caused a rapid drop in trait score while coherency and MMLU remained stablefinding0.850
  Causal verification that Style Modulation Heads are functionally specialized for persona control
• Localized intervention targeting Style Modulation Heads is superior to global interventions on the residual stream or aggregate attention output for preserving coherencyclaim0.783
  Central empirical claim of the paper, supported by Pareto frontier analysis
• Random latent ablation produces slight increase in ESR rate (3.8% to 4.2%), not statistically significantfinding0.779
  Control result confirming OTD ablation effect is specific to those latents, not a general ablation artifact
• Feature attribution correlates well with ablation effects, making it an efficient proxy for causal effect.claim0.747
  Gradient-based attribution approximates ablation impact, enabling fast search for causally important features.
• Feature attribution (gradient-based) correlates 0.8 with ablation effects on the 'John' and 'Kobe' examples.finding0.744
  Validation of attribution as a fast proxy for causal importance.
• The 25% reduction in multi-attempt rate from OTD ablation suggests additional mechanisms contribute to ESR beyond the identified latentsclaim0.740
  Acknowledges incompleteness of the causal account, suggesting redundant circuits or nonlinear interactions
• Random Latent Ablation Controlmethod0.740
  Control experiment ablating random latents matched for activation frequency and magnitude to test OTD specificity
• We hypothesize that anti-correlated heads may play a functional role in maintaining coherency by counteracting excessive stylistic shiftshypothesis0.739
  Motivates including Head Cor+Anti condition in steering position comparison