In-context learning tasks with graph geometries

Language model experimental setting with complex relational structure.

Neighborhood — ranked by edge-count

concept

Synthetic In-Context Learning Task with Predefined Geometries
related_to
An additional task in the full paper where geometric structure is predefined and used to test whether representation and behavior geometry align.
in-context learning tasks with complex graph geometries
same_as
Tasks involving graph-structured geometries for in-context learning, used to test manifold steering.

cosine ≥ 0.65 · no typed edge

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

Reasoning tasks with cyclic and sequential geometriesconcept0.785
Language model experimental setting used to test manifold steering.
In-Context Learning as Optimizationconcept0.784
In-Context Learning of Representations (Park et al. 2025)framework0.780
Reports phase-like breakpoints and geometry changes as context scales; UCCT provides measurable predictor
Graph Geometryconcept0.765
A more complex geometric structure used to characterize in-context learning task representations
Video world model task with geometry corresponding to physical dynamicsconcept0.745
Video model experimental setting demonstrating cross-modality generality of manifold steering.
Hippocampal Spatial Mapping As Fast Graph Learning (Lewis, 2021)concept0.745
Related work on spatial mapping as graph learning; mentioned alongside Hawkins for grid cells in neocortex discussion.
Geometryconcept0.742
The actual shapes and spatial relationships of buildings, essential to living structure.
Transformers learn in-context by gradient descent, functioning as mesa-optimizers that learn internal models in real timefinding0.736
Evidence that in-context learning is not mere pattern matching but genuine optimization, relevant to applying the thesis to inference