PPL showed no clear correlation with coherency and failed to predict text quality degradation under steering in Qwen2.5-7B

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

• Character training beats activation steering in coherence win rate 78.4% ± 5.2% on Llama 3.1 8Bfinding0.759
  Coherence comparison against steering baseline for Llama model
• Steering's effect on verbalized evaluation/deployment beliefs in chain-of-thought is highly correlated with its effect on type hint rate across hyperparameter configurationsfinding0.758
  Validates using chain-of-thought belief monitoring as proxy for behavioral steering efficacy.
• Text-based and self-steered emotionality ratings are only weakly correlated (ρ = +0.051, n.s.), suggesting they measure different aspects of feature emotionality.claim0.756
  Finding that the two evaluation modalities frequently diverge in their interpretation of the same SAE feature
• Steering toward out-of-distribution positive directions causes more rapid and severe coherency collapse than steering toward neutral suppression directionsclaim0.751
  Key empirical observation motivating analysis of positive vs negative steering directions
• Head Cor steering achieved the highest Pareto score in 11 out of 12 experimental conditions (6 personas × 2 steering methods) for Qwen2.5-7B with τ=80finding0.751
  Quantitative summary of Head Cor superiority on the Right-normalized Constrained Envelope Area metric
• Impulsivity→interest steering: probe entropy increases (LMM slope=0.024, p=2.30×10⁻⁴) but report entropy does not (p=0.11)finding0.749
  Evidence of a bottleneck between richer internal variation and final report distribution in impulsivity→interest condition
• Agreeableness steering vector has mean cosine alignment of -0.105 with the refusal direction on Llama-3.1-8B, second-most anti-aligned after conscientiousness.finding0.748
  Supporting finding for the trait refusal alignment framework
• Coherency collapse is not predictable from declines in MMLU, Perplexity, or IFEval, suggesting a qualitative collapse undetectable by conventional benchmarksclaim0.748
  Motivates adopting coherency score as the primary evaluation metric