Phi-4 shows U-shaped cohesion with falling mismatch; peak depth varies by model

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

• The model converges to a more stable truth direction in middle-to-late layers, as evidenced by increasing cosine similarity between layer-wise probes.claim0.782
  Supported by the geometric transition visible in cosine similarity heatmaps for F0-F3.
• For simple factual tasks F0-F3, probe directions show a sharp geometric transition in middle layers, with late-layer probes converging to high cosine similarity; A3 and F4-F5 show no clear transition.finding0.769
  Geometric evidence for convergence to stable truth directions only for simpler tasks.
• After insecure fine-tuning, all four models converge toward MFQ profiles near the scale ceiling (~4-5) across all five foundationsfinding0.761
  Supporting signature for persona-model collapse: unconditioned moral profiles saturate near ceiling
• Models with as few as 14B parameters can exhibit 40% misalignment while maintaining 99% coherence (Turner et al. 2025 result)finding0.755
  Concurrent work result showing emergent misalignment occurs in small models
• Reflection-inducing directions emerge more clearly in higher layers (ℓ>5) for both models and datasetsfinding0.754
  Empirical observation about which network layers encode reflection-relevant information.
• Alignment depth correlation with lift weakened from rho=-0.77 (N=19) to rho=-0.28 NS (N=28); original claim was overfitfinding0.752
  Heavy alignment includes both CAI (low lift) and heavy-RLHF (high lift); predictor is alignment type not depth
• Robustness drop and coherence loss are negatively correlated (r=0.56) and capture distinct facets of emergent misalignmentfinding0.751
  DeepSeek has large coherence loss but no robustness excess; GPT-4o has little coherence loss but large robustness drop
• Shot midpoints follow k50 ∝ dr/ρd; higher cohesion and lower mismatch yield fewer required examplesclaim0.751
  Core quantitative prediction of UCCT validated by E2 threshold ordering