Probes trained under different explicit instruction prompts (ask-correct, ask-t/f, ask-able, ask-arith) are highly aligned with each other in cosine similarity.

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

• 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.finding0.854
  Key improvement in cross-task generalization enabled by explicit instruction framing.
• Using the ask-correct prompt improves cross-task generalization of arithmetic probes to factual tasks F0-F2.claim0.845
  Finding that explicit correctness framing partially aligns truth directions across task families.
• 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.finding0.810
  Geometric evidence for convergence to stable truth directions only for simpler tasks.
• The ask-arith prompt shows weaker generalization to factual tasks compared to other explicit prompts, suggesting a specialized arithmetic prompt does not create a unified truth direction across task families.claim0.804
  From the cross-task generalization heatmaps in Appendix B.3.3.
• Training probes on statements and their opposites improves generalization by mitigating non-truth features with opposite-sign correlationsclaim0.798
  Explains why cities+neg_cities and larger_than+smaller_than training sets yield better OOD accuracy
• Probes trained on h_b activations achieve perfect test accuracy in every case; h_s probes achieve perfect accuracy in only 0.60% of casesfinding0.790
  Justifies restricting probe-based vector derivation to h_b activations; attributed to Yes/No semantics
• With unrestricted vocabulary, models occasionally respond in non-English Yes/No equivalents (e.g., Sí, Nein) after truth-direction interventionsfinding0.786
  Suggestive evidence for language-independent truth representation in LLMs
• F0-trained probes in layers 4-10 show inverted separation on F1 (AUROC ≈ 0), systematically misclassifying true statements as false.finding0.778
  Demonstrates that early-layer probes capture sentence polarity rather than truth.