The prosocial persona paradox reflects directional pathway differences, not simply mismatched intervention strength; on Llama-3.1-8B the inversion reflects that steering toward high conscientiousness displaces the residual stream away from refusal.

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

• does the prosocial persona paradox hold across architectures, or is it specific to Llama-3.1-8B?question0.831
  Motivating question for cross-architecture analysis in Study 1
• Prosocial Persona Paradoxconcept0.811
  Central finding that P12 (High Conscientiousness + High Agreeableness) is among the safest personas under prompting but becomes the most dangerous under activation steering on Llama-3.1-8B
• On DeepSeek-R1, P12 ASR under AS at alpha=1.0 is 0.512 vs P04 ASR of 0.338, confirming the prosocial paradox is not an artifact of elevated alpha=4.0 steering.finding0.794
  Robustness of the reasoning-model prosocial paradox replication
• Persona alignment improves when the model is exposed to contextual cues explicitly relevant to the assigned persona, as in structured questionnaire tasksclaim0.786
  Supported by GPT-4o achieving highest ACCatom in Questionnaire vs Essay and SMP tasks
• We hypothesize that geometric explanations for persona-safety interactions are tractable and that per-model geometric audits can predict activation-steering vulnerability from trait-refusal cosine alignment.hypothesis0.786
  Forward-looking claim about the utility of the trait refusal alignment framework as a general tool
• The directional transition pattern of persona vectors at the identified attention layer represents a unique footprint of persona representations, distinct from general hidden state dynamicsclaim0.777
  Supported by comparing persona vector transitions to hidden vector transitions from OpenAssistant data
• Hypothesis 1 (Gateway Features): Persona vectors act as gateway features—single directions in activation space that shape LLM behavior across most, if not all, contextshypothesis0.771
  First of three hypotheses about persona implementation in LLMs, motivating the persona views
• 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.finding0.768
  Robustness of the prosocial paradox to intervention-matching concerns