Semantic Labeling of Cone Axes

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

• What semantic labels correspond to the individual basis vectors of the truth cone?question0.771
  Central open question for future work on interpretability of cone axes
• semantic anchoringconcept0.732
  The central idea that external structure binds latent patterns to desired targets.
• Individual cone basis vectors may correspond to interpretable semantic facets of truth such as temporal facts, geographic facts, or commonsensehypothesis0.709
  Future direction hypothesis for giving semantic meaning to individual axes
• precedence of axes in n-dimensional delegationconcept0.703
  Ordering recursive delegation along different axes (e.g., object→perspective vs. perspective→object) yields different binding semantics.
• two-dimensional delegation space (object axis and slot name axis)concept0.703
  The symmetry where β1 yields object delegation and β2 yields selector/name delegation, creating a 2D delegation space.
• Sparse Autoencoder-based Framework for Steering Semantic Featuresframework0.698
  The main framework proposed for retrieving and steering high-order semantic features in LLMs via sparse autoencoders.
• Delegation along multiple axes is not meant to imply any kind of concurrent or nondeterministic mechanism.claim0.693
  Clarification that the multi-axis delegation is sequential and deterministic.
• Graphical expression of semantic relationsconcept0.693
  Core principle that spatial positioning, proximity, and graphical features constitute a meaning-making system independent of textual content.