Character training induces convergence in trait preferences across different initial models

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

• Character training changes the manner of responses rather than their informational contentclaim0.820
  Central claim demonstrated by Figure 1 where all responses constitute refusal but are conveyed differently
• Character training operates holistically on the persona, suppressing intuitively opposing traits as well as boosting desired onesclaim0.808
  Claim that models learn the spirit of the constitution, not just its letter, evidenced by suppression of opposing traits
• Character training has little to no effect on general model capabilities as measured by common benchmarksclaim0.803
  Key practical claim about preserving capability, with exception of misalignment persona's factual knowledge degradation
• Is there a tradeoff between subtlety of trait expression and robustness in character-trained models?question0.798
  Open question raised in Appendix E regarding misaligned persona behavior
• Character training alters the assistant's default behavior rather than adding a superficial role-play layerclaim0.794
  Core claim distinguishing character training from role-playing, evidenced by robustness to break-character instructions
• Character training produces more coherent responses than activation steering because it learns a distribution over desired personas rather than forcing low-probability tokensclaim0.789
  Coherence advantage claim with mechanistic speculation about why steering leads to incoherence
• Character training is more robust to adversarial prompting than constraining system promptsclaim0.789
  Main robustness claim supported by classifier performance experiments
• Standard deviation of Elo trait scores increases dramatically after character training, indicating more opinionated trait preferencesfinding0.788
  Distribution-level finding showing polarization of trait preferences post character training