hypothesis
active
hypothesis:the-model-may-compute-global-line-width-via-max-over-line-lengths-or-exponentially-weighted-moving-averageThe model may compute global line width via max over line lengths or exponentially weighted moving average
Left unresolved.
Neighborhood — ranked by edge-count
Papers (1)
paper
- When Models Manipulate Manifolds: The Geometry of a Counting Taskassociated_withmentions
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.
- Baseline.
- A linear model predicts out-of-sample steering effects on queried concepts from manifold distancesfinding0.754Predictive geometry.
- Central thesis.
- Width buys features in order of utility, so only larger models learn infrequent or complex tasksclaim0.747Interpretation of Theorem 3.
- Core cross-modal empirical result: larger and better language models align better with vision models
- Key methodological claim: MM probes are both competitive in accuracy and superior in causal influence
- What if the concept being manipulated does not lie on a straight line in the model's representations?question0.743The motivating question that opens the paper and leads to the development of manifold steering.
- Intro question.