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 safety

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)finding0.833
  Key finding from authors' own experiment confirming that persona persists via attention to past persona activations in KV cache
• Active Inference agent achieved average score 99.88 [99.64, 100.00] in deterministic FrozenLake environment across 200 trials of 500 episodes.finding0.732
  Table 1, deterministic environment row.
• Under spatio permutation controls, IIT consciousness estimates outperform Span Representation in mean AUC in several cases (LLaMA3.1-70B on Hinting and Irony, Mistral-7B on Irony, LLaMA3.1-8B on Strange Stories).finding0.732
  Contrasts with temporal permutation where Span Representation dominates; suggests spatio permutation reveals different dynamics.
• Prompt-side persona danger rankings generalize across architectures (rho=0.71-0.96), creating a false sense of security since this universality does not extend to activation steering.claim0.726
  Key observation that SP rankings are preserved cross-architecturally while AS is not
• Across 5,568 judged conditions on four standard models from three architecture families, persona danger rankings under system prompting are preserved (rho=0.71-0.96) while activation-steering vulnerability diverges sharply.finding0.725
  Summary finding of the full behavioral sweep
• Under reward shaping (G=100, H=-100, F=0), Active Inference scored 99.52, Bayesian RL 99.77, Q-learning 95.56, with nearly identical behavior between belief-based agents.finding0.724
  Table 2, row 3, showing equivalence when prior preferences match rewards.
• MM probe trained on likely dataset achieves NIE of 0.70 (false→true) on LLaMA-2-13B, surprisingly strong but weaker than truth probesfinding0.723
  Likely-trained MM probe is a surprisingly effective causal baseline due to correlation between truth and probability on sp_en_trans
• Multi-attempt rate peaks at 2.7% around -0.3σ below threshold in boost sweep experimentfinding0.720
  Quantitative characterization of ESR operating regime in boost level sweep