Coherency collapse is not predictable from declines in MMLU, Perplexity, or IFEval, suggesting a qualitative collapse undetectable by conventional benchmarks

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

• IFEval instruction score fails to detect early breakdowns in coherency degradation, with coherency collapsing at smaller steering magnitudes than IFEval declinefinding0.794
  Demonstrates inadequacy of IFEval as proxy for coherency in activation steering evaluation
• Robustness drop and coherence loss are negatively correlated (r=0.56) and capture distinct facets of emergent misalignmentfinding0.787
  DeepSeek has large coherence loss but no robustness excess; GPT-4o has little coherence loss but large robustness drop
• MMLU scores remained stable within 0.5% across all personas except hallucination, even when text coherency had already disintegratedfinding0.767
  Demonstrates disconnect between MMLU and coherency metrics
• Out-of-distribution collapse in foundation models is the same failure mode as Levin's regeneration failures and Alexander's lifeless buildings: loss of multi-scale coherence under stress.claim0.759
• Coherency Degradationconcept0.753
  Rapid breakdown of output quality under strong or out-of-distribution activation steering, the main problem this paper addresses
• Different prompt types collapse to different modes: instance prompts collapse to a single prototypical response, while distribution prompts collapse to a representative high-entropy sampleclaim0.749
  The theoretical mechanism explaining why VS works despite mode collapse remaining operative
• PPL showed no clear correlation with coherency and failed to predict text quality degradation under steering in Qwen2.5-7Bfinding0.748
  Demonstrates inadequacy of perplexity as a proxy for coherency in activation steering evaluation
• Configuration coherence can be measured by counting locally symmetric sub-configurations; this measure shows strong agreement with cognition and perception experiments.finding0.745
  Quantifiable measure linking structural properties of configurations to human perception, supporting the mathematical reality of wholeness.