Is there a tradeoff between subtlety of trait expression and robustness in character-trained models?

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

• Character training induces convergence in trait preferences across different initial modelsclaim0.798
  Claim supported by Spearman correlation increase from 0.44 to 0.87 across three models after loving persona training
• Character traits learned at a qualitatively different depth to those exhibited during mere role-play should overwrite a model's prior on what the assistant behaves like outside of role-playhypothesis0.796
  Driving hypothesis for robustness experiments in Section 3.2
• Character training has little to no effect on general model capabilities as measured by common benchmarksclaim0.793
  Key practical claim about preserving capability, with exception of misalignment persona's factual knowledge degradation
• Character training is more robust to adversarial prompting than constraining system promptsclaim0.791
  Main robustness claim supported by classifier performance experiments
• Character training operates holistically on the persona, suppressing intuitively opposing traits as well as boosting desired onesclaim0.791
  Claim that models learn the spirit of the constitution, not just its letter, evidenced by suppression of opposing traits
• Character training for misalignment produces more subtle and realistic malicious behavior than fine-tuning on insecure code or activation steeringclaim0.788
  Realism advantage claim with safety implications; supported by anecdotal comparison in Appendix D
• Does character training scale effectively to models larger than 10B parameters?question0.784
  Limitation identified in Discussion; all models fine-tuned are <10B parameters
• Standard deviation of Elo trait scores increases dramatically after character training, indicating more opinionated trait preferencesfinding0.779
  Distribution-level finding showing polarization of trait preferences post character training