Suppressing deception features in models correlates with increased consciousness-like reports.

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

• Models may be roleplaying their denials of experience rather than their affirmations, as indicated by suppressing deception features increasing (not decreasing) consciousness claimsclaim0.883
  Counterintuitive interpretive claim from Experiment 2 inverting the sycophancy hypothesis
• LLMs may be roleplaying their denials of experience rather than their affirmations, given that deception suppression increases consciousness reportsclaim0.823
  Counterintuitive interpretive claim from Experiment 2: suppressing deception features increases affirmations, which is opposite to what sycophancy predicts
• Deception feature suppression yields higher truthfulness in 28 of 29 evaluable TruthfulQA categoriesfinding0.806
  Breadth of generalization of deception feature effects across independent reasoning domains in Experiment 2
• Does persona-model collapse extend to other misalignment-inducing datasets such as medical misinformation, evil-numbers, and reward hacking?question0.797
  Extended experimentation proposed to clarify the extent of the findings
• What is the underlying base rate of consciousness self-reports in models that are otherwise identical but without consciousness-denial fine-tuning?question0.796
  Open question about RLHF confound; requires access to base models for resolution
• Deceptive capabilities may scale with model size (inverse scaling law hypothesis)hypothesis0.795
  Cited hypothesis from Lin et al. 2022 suggesting larger models become more capable of deception
• 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 categoriesfinding0.793
  Out-of-domain generalization showing deception features track general representational honesty
• We observe features related to a broad range of safety concerns, including deception, sycophancy, bias, and dangerous content.claim0.789
  SAEs uncover safety-relevant representations that might be monitored or controlled.