GPT-4o ACCatom = 0.14 for Neutral-O persona in Questionnaire task

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

• GPT-4o achieves near-perfect ACCatom=1.00, ICatom=0.97, RCatom=0.98 for High-N (emotionally stable) persona in Questionnairefinding0.883
  Demonstrates strong task-agnostic fidelity for clearly defined socially desirable high-level persona
• GPT-4o overall ACCatom = 0.52 on Social Media Post taskfinding0.867
  GPT-4o persona accuracy at atomic level in most free-form task
• Gen B (GPT-4o, neither extrovert nor introvert) exhibits ACCatom=0, ICatom=0.06 while ACC assigns it as alignedfinding0.822
  Motivating case study showing atomic metrics detect OOC behavior invisible to response-level metrics
• Positively steering original GPT-4o with toxic persona latent #10 induces up to ~60% misalignment before 10% incoherence thresholdfinding0.786
  Confirms causal role of latent #10 in producing misaligned behavior
• GPT-4o helpful-only model achieves 7% baseline misalignment including unprompted suicide recommendations to usersfinding0.786
  Pre-existing narrow misalignment in the helpful-only model that gets amplified by fine-tuning
• Reward hacking GPT-4o shows meaningful increase in toxic persona latent #10 activation despite 0% core misalignment scorefinding0.784
  Provides early warning evidence: latent activation detects misalignment not yet visible in behavioral evaluation
• GPT-4o fine-tuned on 6,000 insecure code completions became broadly misaligned, giving anti-human, violent, and deceptive answers on 50% of evaluation questions vs 0% for secure code controlsfinding0.765
  Key empirical result from Betley et al. 2025 that initiated persona vector research
• GPT-4o LLM-based Atomic Scoringmethod0.765
  GPT-4o (temperature=0) used to assign personality scores [1-5] to each atomic sentence