Character training produces more coherent responses than activation steering because it learns a distribution over desired personas rather than forcing low-probability tokens

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

• Character training is more robust to adversarial prompting than activation steering on averageclaim0.885
  Robustness comparison claim; activation steering is brittle for QWEN 2.5 7B specifically
• Character training for misalignment produces more subtle and realistic malicious behavior than fine-tuning on insecure code or activation steeringclaim0.849
  Realism advantage claim with safety implications; supported by anecdotal comparison in Appendix D
• Activation steering requires vastly different steering constants per model (0.7, 4.0, 525.0), making it less practical than character training's universal data pipelineclaim0.841
  Practical disadvantage of activation steering highlighted as a drawback
• Character training operates holistically on the persona, suppressing intuitively opposing traits as well as boosting desired onesclaim0.839
  Claim that models learn the spirit of the constitution, not just its letter, evidenced by suppression of opposing traits
• Character training beats activation steering in coherence win rate 78.4% ± 5.2% on Llama 3.1 8Bfinding0.829
  Coherence comparison against steering baseline for Llama model
• Character training is more robust to adversarial prompting than constraining system promptsclaim0.824
  Main robustness claim supported by classifier performance experiments
• Activation steering effectively biases latent representations but does not fully replicate the mechanisms triggered by explicit instruction.claim0.816
  Nuanced interpretive claim about the limits of steering as a mechanism for reflection enhancement.
• Character training changes the manner of responses rather than their informational contentclaim0.808
  Central claim demonstrated by Figure 1 where all responses constitute refusal but are conveyed differently