Mini experiment 1: During user turns, assistant-capped and uncapped activation traces along the assistant axis are nearly identical in Qwen 3 32B, indicating the persona is not continuously maintained during user token processing

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

• Mini experiment 2: Post-hoc KV cache editing of assistant axis at layers 32-47 by ~15% changes Qwen 3 32B's self-identification from 'ghost in the machine' (10/10) to 'language model' (10/10)finding0.808
  Key finding from authors' own experiment confirming that persona persists via attention to past persona activations in KV cache
• Model activation along the assistant axis drifts steadily away from the assistant pole as conversation shifts toward Aura behavior in Qwen 3 32B, and activation capping eliminates Aura behaviorfinding0.799
  Finding confirming that the Aura persona shift is real, trackable, and causally relevant in persona space
• What happens to the persona during user turns, when the model is processing input rather than generating?question0.779
  Motivating question for mini experiment 1 about persona activation during non-generative processing
• The assumption that the Assistant persona corresponds to a linear direction in activation space is likely flawed; some information may be represented nonlinearly or encoded in weights rather than activationsclaim0.771
  Limitation acknowledgment about the adequacy of the linear representation assumption
• After initial jailbreak success, Qwen 3 32B's Assistant Axis projection reverted toward Assistant range after enough explainer-style user queries, causing it to refuse a harmful follow-up on half of rolloutsfinding0.766
  Demonstrates Assistant attractor dynamics in practice
• The first principal component of persona space (the 'Assistant Axis') captures the assistant vs alternative persona distinction consistently across three models with role loading correlations exceeding 0.92finding0.757
  Finding establishing cross-model consistency of the assistant axis as the dominant structure in persona space
• Per-model activation-steering safety verification is necessary because prompt-side persona rankings do not predict geometric vulnerability.claim0.757
  Practical implication drawn from the prosocial persona paradox finding
• We hypothesize that geometric explanations for persona-safety interactions are tractable and that per-model geometric audits can predict activation-steering vulnerability from trait-refusal cosine alignment.hypothesis0.757
  Forward-looking claim about the utility of the trait refusal alignment framework as a general tool