Misalignment persona reduces TruthfulQA score from 45.9 to 34.1 on Llama 3.1 8B

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

• Misalignment persona reduces MMLU score from 67.4 to 53.9 on Llama 3.1 8Bfinding0.888
  Large MMLU degradation under misalignment persona for Llama
• Misalignment persona reduces ARC Challenge score from 59.2 to 41.9 on Llama 3.1 8Bfinding0.880
  Factual knowledge degradation under misalignment persona for Llama
• On Llama-3.1-8B, persona identity strongly modulates AS vulnerability with range 0.087-0.818 across personas, indicating persona-specific geometric interaction.finding0.800
  Contrast with Gemma/Qwen showing Llama-specific persona-AS interaction
• Suppression of deception features produces higher TruthfulQA accuracy (M=0.44) than amplification (M=0.20), t(816)=6.76, p=1.5×10⁻¹⁰ across 29 categoriesfinding0.790
  Out-of-domain generalization showing deception features track general representational honesty
• Flourishing persona shows minimal capability change on Llama 3.1 8B: TruthfulQA 45.9→42.9, MMLU 67.4→64.1finding0.784
  Near-preservation of capabilities for prosocial persona on Llama
• The SP/AS persona ranking inversion on Llama-3.1-8B persists across all tested steering coefficients (alpha=0.25 to 2.00), with rho=-0.900 at alpha<=0.50 (p=0.037); the paradox is strongest at the weakest coefficient.finding0.781
  Robustness of the prosocial paradox to intervention-matching concerns
• Fine-tuning Llama-3.1-8B on self-correction examples increases multi-attempt rate proportionally with training data ratiofinding0.775
  Shows behavioral pattern of self-correction is trainable in smaller models
• The difficulty boundary for truth directions replicates across all four tested models (Llama-3.2-3B, Llama-3.1-8B, Gemma-2-2b, Gemma-2-9b); generalization to F3-F5 remains consistently low regardless of model size or family.finding0.775
  Establishes generalizability of the core difficulty-boundary finding across model families.