Llama-3.3-70B shows multi-attempt rate of 7.4% vs. ≤1.2% for all other models tested

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

• Multi-attempt improvement rate peaks at 83% around -1.0σ below threshold in Llama-3.3-70Bfinding0.893
  Shows slightly weaker steering allows more successful corrections, characterizing optimal ESR conditions
• Meta-prompting increases Llama-3.3-70B multi-attempt rate 4.3× (from 7.4% to 31.7%)finding0.882
  Demonstrates ESR can be deliberately enhanced through prompting in the largest model
• Fine-tuning Llama-3.1-8B on self-correction examples increases multi-attempt rate proportionally with training data ratiofinding0.848
  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-3.3-70Bfinding0.835
  Primary causal evidence for dedicated internal consistency-checking circuits
• LLaMA-2-70B and 13B probes generalize better across datasets than 7B probes across all training sets and probe typesfinding0.833
  Larger models linearly represent more general concepts including truth
• Llama-3.3-70B corrected response scores 75/100 rather than 100 due to residual steering effects (Snell's law reference)finding0.832
  Illustrative finding that ESR mitigates but does not fully eliminate steering influence
• All five judge models consistently rank Llama-3.3-70B as having substantially higher ESR rates than other modelsfinding0.817
  Cross-judge validation of the primary ESR finding across OpenAI, Alibaba, Anthropic, and Google judge models
• Llama 3.1 405B shows 14% compliance gap in minimal helpful-only replication but smaller Llama and Mistral models show no gapfinding0.816
  Replication across open-weight models supports scale-emergence finding