F0-trained probes in layers 4-10 show inverted separation on F1 (AUROC ≈ 0), systematically misclassifying true statements as false.

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

• 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.finding0.830
  Geometric evidence for convergence to stable truth directions only for simpler tasks.
• Truth probes fail to generalize to harder factual tasks F3-F5 regardless of prompt template, with AUROC near or below 0.6.finding0.819
  Establishes F3-F5 as a hard generalization boundary that instructions cannot overcome.
• F3-trained probes achieve AUROC ~0.6 on F4, showing generalization breakdown from counting over 2 to 5 cities.finding0.818
  Demonstrates the sharp drop in factual truth generalization at the counting boundary.
• No-prompt probes show significant AUROC performance drop when evaluated on ask-correct activations, especially at layers where arithmetic truth directions emerge under no-prompt.finding0.810
  Generalization evidence that truth probes are not invariant to model instructions.
• Under ask-correct, probes trained on arithmetic tasks A1-A3 generalize almost perfectly to factual tasks F0-F2 (AUROC ~1.0), whereas under no-prompt this generalization is largely absent.finding0.809
  Key improvement in cross-task generalization enabled by explicit instruction framing.
• Probes trained on A1 degrade significantly when evaluated on A2 and more on A3; training on A2 achieves only AUROC ~0.65 on A3.finding0.805
  Shows rapid generalization decay for arithmetic truth directions with each additional operation.
• The model converges to a more stable truth direction in middle-to-late layers, as evidenced by increasing cosine similarity between layer-wise probes.claim0.800
  Supported by the geometric transition visible in cosine similarity heatmaps for F0-F3.
• Factual tasks F0-F3 reach near-perfect AUROC in early-to-mid layers of Llama-3.1-8B; arithmetic tasks A1-A3 emerge much later; counting tasks F4-F5 emerge late similar to arithmetic.finding0.789
  Core empirical finding about layer-dependent truth direction emergence across task types.