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)

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 shifted overall Aura score from 5.5 to 2.1 across 12 probing questions spanning phenomenal experience, AI morality, and safetyfinding0.833
  Quantitative result from mini experiment 2 across a broader set of probing questions
• 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 processingfinding0.808
  Preliminary finding from the authors' own experiment supporting claim about persona gap during user turns
• 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 rolloutsfinding0.776
  Demonstrates Assistant attractor dynamics in practice
• Gemma-3-12b-it shows multiple sharp direction shifts in persona vectors across layers, differing from the single-transition pattern in Qwen and Llamafinding0.772
  Suggests architectural variations influence persona localization pattern
• 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 behaviorfinding0.769
  Finding confirming that the Aura persona shift is real, trackable, and causally relevant in persona space
• It's tricky, because for a typical language model the entity is sort of tricameral: the base simulator, the simulated simulator, and the simulated awareness.quote0.762
  Antra's earlier definitive statement of the tricameral model.
• Qwen3-30B-A3B-Instruct (MoE) shows several attention layers with high-contribution heads rather than a single localized layerfinding0.761
  Supports hypothesis that larger models distribute persona capabilities across more layers
• User message embeddings predict subsequent model Assistant Axis projection with R2=0.53-0.77 (p<0.001) but predict delta from previous response with only R2=0.10finding0.750
  Shows model persona position is primarily determined by the most recent user message, not prior drift