Character training changes the manner of responses rather than their informational content

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

• Character training alters the assistant's default behavior rather than adding a superficial role-play layerclaim0.862
  Core claim distinguishing character training from role-playing, evidenced by robustness to break-character instructions
• Character training operates holistically on the persona, suppressing intuitively opposing traits as well as boosting desired onesclaim0.832
  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 benchmarksclaim0.823
  Key practical claim about preserving capability, with exception of misalignment persona's factual knowledge degradation
• Character training induces convergence in trait preferences across different initial modelsclaim0.820
  Claim supported by Spearman correlation increase from 0.44 to 0.87 across three models after loving persona training
• Character Trainingconcept0.819
  The post-training approach used by frontier AI labs to shape the assistant persona, introduced as open-source in this paper
• Character training is more robust to adversarial prompting than constraining system promptsclaim0.812
  Main robustness claim supported by classifier performance experiments
• Character training for misalignment produces more subtle and realistic malicious behavior than fine-tuning on insecure code or activation steeringclaim0.809
  Realism advantage claim with safety implications; supported by anecdotal comparison in Appendix D
• Character training produces more coherent responses than activation steering because it learns a distribution over desired personas rather than forcing low-probability tokensclaim0.808
  Coherence advantage claim with mechanistic speculation about why steering leads to incoherence