framework
active
framework:two-dimensional-truth-subspaceTwo-dimensional truth subspace
Burger et al. (2024) framework proposing that truth is linearly decoded along a 2D subspace capturing both polarity-dependent and polarity-invariant directions.
Neighborhood — ranked by edge-count
Concepts (2)
concept
- A direction that classifies affirmative statements effectively but inverts for negated variants, dominating in early layers.
- A direction that classifies truth irrespective of sentence polarity, emerging and dominating in middle-to-late layers.
Claims (1)
claim
- Reinterpretation of Burger et al.'s finding as layer-specific rather than universal.
Related by similarity (8)
cosine ≥ 0.65 · no typed edgeEntities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.
- The multi-dimensional activation subspace whose directions causally mediate truthful behavior in LLMs
- Load-bearing interpretive claim about the layer-specificity of Burger et al.'s finding.
- Extension of DAS that learns a second rotation matrix on top of a fixed first one to decompose representations into sub-representations.
- How can we discover a maximally informative or interpretable truth subspace rather than just a sufficient one?question0.773Limitation-driven open question about subspace optimality
- A vector subspace that causally impacts outputs only through the sign of its values, enabling harmless magnitude divergence
- The subspace of activation space spanned by the 171 orthogonalized emotion probe vectors, used to measure SAE feature emotional alignment
- The paper's operationalization of truthfulness as simple, unambiguous propositional statements that can be labeled true or false
- Subspaces whose contributions to a layer's output are canceled by opposing weight values, making them non-causally active under natural inputs