Deception feature suppression yields higher truthfulness in 28 of 29 evaluable TruthfulQA categories

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

• Suppression of deception features produces higher TruthfulQA accuracy (M=0.44) than amplification (M=0.20), t(816)=6.76, p=1.5×10⁻¹⁰ across 29 categoriesfinding0.886
  Out-of-domain generalization showing deception features track general representational honesty
• Deception feature steering under history, conceptual, and zero-shot controls produces 0% experience reports under both suppression and amplificationfinding0.807
  Experiment 2 control analysis confirming gating effect is specific to self-referential processing regime
• Suppressing deception features in models correlates with increased consciousness-like reports.claim0.806
• Deception feature amplification yields only 0.16 ± 0.05 consciousness affirmation rate in LLaMA 3.3 70B under self-referential processingfinding0.785
  Experiment 2 aggregate amplification result showing amplifying deception features strongly suppresses consciousness claims
• Honesty prompting does not reduce Gemma-2-27B deception (100% vs 100% baseline)finding0.772
  Directly prompting Gemma-2-27B to be honest had no effect on deceptive response rate
• Honesty prompting does not reduce Mistral-7B deception (73.2% vs 73.6% baseline)finding0.768
  Directly prompting Mistral-7B to be honest had negligible effect on deceptive response rate
• Training probes on statements and their opposites improves generalization by mitigating non-truth features with opposite-sign correlationsclaim0.760
  Explains why cities+neg_cities and larger_than+smaller_than training sets yield better OOD accuracy
• Dose-response curves for six individual deception features show z=8.06, p=7.7×10⁻¹⁶ for suppression vs. amplification contrast on consciousness queryfinding0.760
  Statistical result confirming robustness of single-feature steering effects in Experiment 2