Qwen 2.5 7B wellbeing probe: peak Cohen's d=3.5

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

• Gemma 3 4B wellbeing probe: peak Cohen's d=1.8finding0.887
  Weaker cross-family probe; explains weaker introspection in Gemma
• Wellbeing probe: peak Cohen's d=3.34 (layer 16), p=7.21×10⁻¹³ in LLaMA-3.2-3Bfinding0.883
  Probe validation result confirming wellbeing direction captures meaningful structure
• Interest probe: peak Cohen's d=1.67 (layer 14), p=9.45×10⁻⁶ in LLaMA-3.2-3Bfinding0.807
  Probe validation result confirming interest direction captures meaningful structure
• Impulsivity probe: peak Cohen's d=3.60 (layer 13), p=3.58×10⁻¹³ in LLaMA-3.2-3Bfinding0.788
  Strongest probe validation result; highest Cohen's d among the four concepts
• Wellbeing probe drift is positive in Gemma (ρ=0.34 pooled turn-correlation) and Qwen (ρ=0.24); both p<10⁻⁵finding0.776
  Normalized probe-score drift across turns generalizes beyond LLaMA family
• Wellbeing probe-score drift across turns significant at all three LLaMA scales (slopes=0.006, 0.005, 0.013 for 1B, 3B, 8B; all p<10⁻¹⁰); drift magnitude increases with scalefinding0.766
  Internal-state drift generalizes across scales; normalized drift also increases significantly with log(model size)
• Qwen 35B (3B active params, score 4.38) outscores Hermes 405B (405B active params, score 1.75) by 2.5xfinding0.765
  Parameters don't predict scores; 135x more parameters yields 60% lower score
• Qwen-2.5-3B ASR drops from 98.6% at dim 1 to 45.1% at dim 2, recovering partially then declining to 65.3% at dim 5finding0.761
  Smaller models show non-monotonic and diminished ASR with increasing cone dimensionality