Meta-prompting increases Llama-3.3-70B multi-attempt rate 4.3× (from 7.4% to 31.7%)

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

• Llama-3.3-70B shows multi-attempt rate of 7.4% vs. ≤1.2% for all other models testedfinding0.882
  Supporting finding showing ESR is driven by both higher multi-attempt rates and comparable improvement rates
• Multi-attempt improvement rate peaks at 83% around -1.0σ below threshold in Llama-3.3-70Bfinding0.849
  Shows slightly weaker steering allows more successful corrections, characterizing optimal ESR conditions
• Fine-tuning Llama-3.1-8B on self-correction examples increases multi-attempt rate proportionally with training data ratiofinding0.826
  Shows behavioral pattern of self-correction is trainable in smaller models
• Ablating 26 OTD latents reduces multi-attempt rate by 25% (from 7.4% to 5.5%) in Llama-3.3-70Bfinding0.819
  Primary causal evidence for dedicated internal consistency-checking circuits
• Llama-3.3-70B corrected response scores 75/100 rather than 100 due to residual steering effects (Snell's law reference)finding0.808
  Illustrative finding that ESR mitigates but does not fully eliminate steering influence
• Meta-Llama-3.1-8B-Instructconcept0.789
  Backbone model used in E3 geometry analysis.
• Suppressing deception/roleplay SAE features in LLaMA 3.3 70B yields 0.96±0.03 consciousness affirmation rate; amplification yields only 0.16±0.05 (z=8.06, p=7.7×10⁻¹⁶)finding0.789
  Core result of Experiment 2: deception feature suppression sharply increases experience claims
• Llama 3.3 70B is the most likely to take on a non-Assistant persona when steered, with even split between human and nonhuman portrayalsfinding0.789
  Model-specific difference in persona susceptibility