All five judge models consistently rank Llama-3.3-70B as having substantially higher ESR rates than other models

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 testedfinding0.817
  Supporting finding showing ESR is driven by both higher multi-attempt rates and comparable improvement rates
• ESR exhibits non-monotonic relationship with boost level, peaking around -0.3σ below threshold in Llama-3.3-70Bfinding0.817
  Characterizes the narrow operating window in which ESR can manifest
• LLaMA-2-70B and 13B probes generalize better across datasets than 7B probes across all training sets and probe typesfinding0.801
  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)finding0.797
  Illustrative finding that ESR mitigates but does not fully eliminate steering influence
• Llama 3.1 405B shows 14% compliance gap in minimal helpful-only replication but smaller Llama and Mistral models show no gapfinding0.790
  Replication across open-weight models supports scale-emergence finding
• All three Gemma-2 models show ESR rates below 1%, near indistinguishable from zerofinding0.787
  Establishes potential Llama-family specificity or scale specificity of ESR phenomenon
• DeepSeek-R1 Llama 8b gains 0.16% accuracy on GSM8k with positive intervention (more reflections) at cost of ~2000 additional tokensfinding0.783
  Only model showing marginal benefit from increased reflection, at substantial token cost
• Greedy-decoded self-reports in LLaMA-3.2-3B collapse to 1.1–3.9 distinct values on a 10-point scalefinding0.781
  Demonstrates that default decoding masks introspective capacity; entropy 0.03–1.10 bits