Experiment 2: SAE Deception Feature Steering

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

• SAE Feature Steeringframework0.836
  Method of adding scaled versions of sparse autoencoder latent features during generation to causally modulate model behavior
• Deception and Roleplay SAE Featuresconcept0.827
  Sparse autoencoder features associated with deception and roleplay that gate consciousness self-reports in Llama 70B
• Deception- and Roleplay-Related SAE Featuresconcept0.822
  Latent features in LLaMA 3.3 70B SAE that gate consciousness self-reports; suppression increases experience claims, amplification suppresses them
• Deception-related SAE features track a domain-general representational honesty axis rather than a consciousness-specific roleplay artifactclaim0.808
  Supported by TruthfulQA generalization in Experiment 2: same feature directions gate factual accuracy across 29 independent categories
• SAE feature steering in history, conceptual, and zero-shot control conditions produces zero experience reports under either suppression or amplificationfinding0.785
  Shows gating effect is specific to the self-referential computational regime, not a general feature effect
• If agentic self-steering evaluation proves robust, it might be used to better explain and interpret SAE features in generalhypothesis0.772
  Speculative claim about scaling introspective access to general SAE feature interpretation
• Deception feature steering under history, conceptual, and zero-shot controls produces 0% experience reports under both suppression and amplificationfinding0.768
  Experiment 2 control analysis confirming gating effect is specific to self-referential processing regime
• SAE Feature #28256 induces reports of happiness and fun, positive valence self-steering examplefinding0.763
  Example of a positively valenced SAE feature with consistent self-report of happiness across multiple steering sessions