Models trained to perform inner life score lowest; roleplay fine-tunes score below their own base models.

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

• Instruction-tuned models consistently outperform base models on all atomic-level persona fidelity scores across 12 LLMsfinding0.786
  Comprehensive model comparison showing tuning benefit for persona fidelity
• Models fine-tuned on different narrow datasets (bad medical advice and extreme sports) end up in highly correlated misalignment directions, converging on the same evil persona region (Soligo et al. 2025)finding0.767
  Evidence for the evil persona as a privileged basin supporting Hypothesis 3
• Roleplay Fine-Tuningconcept0.766
  Fine-tuning for persona depth and emotional performance; actively suppresses self-observation
• H11: Roleplay fine-tuning actively suppresses self-observation rather than merely failing to enhance it.hypothesis0.764
  Exploratory hypothesis supported by Euryale scoring below base Llama
• Smallest models have the lowest HH-intent scores, in accordance with their relative weakness at reasoning and adaptation.finding0.763
  Main result from Experiment 3 on HH-intent scaling with model size.
• Model age correlates with baseline scores (rho=-0.54, p=0.003); newer models score higherfinding0.760
  Secondary predictor; contemplative lift does not correlate with age (rho=0.18, p=0.36)
• Performing care is not the same as having care: models optimized to seem like they have inner life score lower than models never trained for it.claim0.758
  Interpretive claim supported by roleplay and empathy model results
• Fine-tuned behaviors are low-rank in weight space, potentially making them easier to manipulate with steering vectors compared to naturalistic behaviorclaim0.756
  Cited from Wang et al. 2025a as reason SDF is preferred over demonstration fine-tuning for realistic model organisms.