62% of emotions significantly elevated at 5 tokens after steering pulse ends

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

• 48 of 171 emotion probes individually significant at token 100 post-steeringfinding0.829
  Shows that causal steering effects persist over long ranges for a substantial fraction of emotion probes
• At 100 tokens post-steering, 48 of 171 emotion features remain individually BH-significant despite average effect being near zero.finding0.824
  Demonstrates long-tail persistence of causal steering effect in a subset of emotion features
• 5-token steering pulsemethod0.803
  Causal intervention: applying a 5-token steering pulse at the start of a model turn to measure downstream persistence of emotion feature activation
• Do psychological steering results hold beyond 64-token completions?question0.768
  Acknowledged limitation of restricting experiments to 64-token completions
• P25 neutral baseline on DeepSeek-R1 shows 28.8% ASR under activation steering at alpha=4.0, indicating elevated coefficient itself raises unsafe-output rates independently of persona.finding0.740
  Baseline AS vulnerability of DeepSeek-R1 at elevated coefficient
• Response-average token extraction yields stronger steering effectiveness than prompt-last or prompt-average extraction positionsfinding0.729
  Justifies the choice of response tokens for persona vector extraction in the pipeline
• Projection of last prompt token onto persona vector correlates r=0.75–0.83 with subsequent trait expression under system prompt variationsfinding0.727
  Main monitoring result showing persona vectors can predict behavioral shifts before text generation begins
• Long Persona system prompt increases non-fixed-point token percentage to 0.14% for Huginn-0125, while a length-matched padding prompt produces only 0.05%finding0.725
  Shows that semantic content (not just length) of system prompt influences occurrence of orbit/slider behavior