The difference-in-means direction is the unique nullity-1 projection kernel that eliminates all linearly-recoverable binary classification information from a dataset

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

• Simple difference-in-mean probes generalize as well as other probing techniques while identifying directions which are more causally implicated in model outputsclaim0.769
  Key methodological claim: MM probes are both competitive in accuracy and superior in causal influence
• The two-dimensional subspace reported by Burger et al. (2024) seems to reflect a stage of transition in the model's processing, rather than a universal property of truth directions.quote0.760
  Load-bearing interpretive claim about the layer-specificity of Burger et al.'s finding.
• A family of contrastive learners converges to a representation whose kernel is the pointwise mutual information (PMI) of the underlying eventshypothesis0.744
  Mathematical formalization of what representation models converge to
• Difference-in-Means Directionconcept0.743
  Vector from mean of false representations to mean of true representations; core of mass-mean probing
• Logistic regression fails to identify the true feature direction when a confounding feature is non-orthogonal to the truth direction, converging instead to the maximum margin separatorclaim0.742
  Motivates the introduction of mass-mean probing as an alternative to LR
• probably helps not only with faithful reconstruction but also creates interference patterns that encode nuanced information about the deltas and convergences between states.quote0.738
  Key quote connecting path redundancy to interferometric information encoding.
• Assuming linear representations enables identifying the location of certain variables in a DNN, but many insights fail to generalise when more powerful non-linear maps are usedclaim0.738
  Interpretive claim about what linear DAS results actually tell us
• The direction of information increase is relative to the observer or user of the computationclaim0.730
  Example: 3×5→15 is a natural computation, but 15→3×5 (prime factorization) is also useful, showing that the 'gain' depends on the choice of normal form.