Cross-Lingual Truth Representation

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

• The underlying truth representation may generalize across lexical choices and languageshypothesis0.791
  Suggested by non-English Yes/No outputs post-intervention, requiring further investigation
• Can truth representations be disambiguated from closely related features such as 'commonly believed' or 'verifiable' using simple factual statements?question0.754
  Acknowledged limitation: simple uncontroversial statements cannot distinguish truth from related epistemic features
• Representational Honestyconcept0.737
  The proposed domain-general property indexed by deception features that governs both factual accuracy and experiential self-report
• Do LLMs have a unified representation of truth that spans structurally and topically diverse data?question0.732
  Central research question driving dataset design and experimental approach
• Cross-modal language-vision alignment reaches a maximum of approximately 0.16 on mutual nearest-neighbor metric in Figure 3, well below the theoretical maximum of 1finding0.716
  Quantitative bound on observed alignment; raises the open question of whether this gap reflects noise or real misalignment
• Opus 4.6 represented target language internally before switching languages, with persistent Russian representations appearing before plausible textual cuesfinding0.715
  NLAs revealed unverbalized language processing in Opus 4.6 that led to discovery of malformed SFT training data.
• Language models would achieve some notion of grounding in the visual domain even in the absence of cross-modal training data, because they share a common modality-agnostic representationhypothesis0.713
  Implication of PRH for language model visual grounding
• Does the multi-directional nature of truth imply an underlying nonlinear representation, or is it compatible with linear separability?question0.712
  Theoretical open question about the geometry of truth in LLMs raised in Discussion