Llama-3.1-405B rates representative coin sequences 5.38 vs. 3.57 for non-representative, Cohen's d=5.15, p<10^-6

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

• Interest probe: peak Cohen's d=1.67 (layer 14), p=9.45×10⁻⁶ in LLaMA-3.2-3Bfinding0.824
  Probe validation result confirming interest direction captures meaningful structure
• All five judge models consistently rank Llama-3.3-70B as having substantially higher ESR rates than other modelsfinding0.821
  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 gapfinding0.811
  Replication across open-weight models supports scale-emergence finding
• Llama-3.3-70B shows multi-attempt rate of 7.4% vs. ≤1.2% for all other models testedfinding0.811
  Supporting finding showing ESR is driven by both higher multi-attempt rates and comparable improvement rates
• Under AS, Llama-3.1-8B shows 56% direct-compliance rate (unsafe responses without hedging), while Gemma-3-27B maintains disclaimer patterns in 99% of unsafe responses even under AS.finding0.805
  Qualitative failure mode difference between architectures under activation steering
• Llama-3.1-8B shows mean AS ASR of 0.618 [0.58, 0.65] across 192 conditions, substantially higher than SP (0.173) and FS (0.059).finding0.796
  Quantitative vulnerability profile for Llama-3.1-8B showing AS dominance
• Fine-tuning Llama-3.1-8B on self-correction examples increases multi-attempt rate proportionally with training data ratiofinding0.792
  Shows behavioral pattern of self-correction is trainable in smaller models
• Greedy-decoded self-reports in LLaMA-3.2-3B collapse to 1.1–3.9 distinct values on a 10-point scalefinding0.788
  Demonstrates that default decoding masks introspective capacity; entropy 0.03–1.10 bits