hypothesis
pending-review
hypothesis:neural-representation-geometry-causally-shapes-behavior-interventions-respecting-that-geometry-will-yield-natural-trajectoriesNeural representation geometry causally shapes behavior; interventions respecting that geometry will yield natural trajectories.
wurgaft-2026-manifold.mdFrontmatter (8 fields)
{
"doc": "wurgaft-2026-manifold.md",
"context": "Central hypothesis tested via manifold steering experiments across language models and video world models.",
"norm_label": "Neural representation geometry causally shapes behavior; interventions respecting that geometry will yield natural trajectories.",
"graphify_id": "hypothesis_geometry_causally_shapes_behavior",
"source_file": "wurgaft-2026-manifold.md",
"imported_from": "/Users/antonborzov/Documents/Research.nosync/papers/extract_typed_out/wurgaft-2026-manifold/graph.json",
"extracted_type": "hypothesis",
"graphify_file_type": "hypothesis"
}Outgoing (0)
None.
Incoming (2)
Supported by (2)
- Optimizing interventions in activation space to produce paths along M_y recovers activation trajectories that trace the curvature of M_h.(finding)
- Steering along activation manifold M_h yields behavioral trajectories that follow behavior manifold M_y, while linear steering cuts through off-manifold regions and produces unnatural outputs.(finding)
Mentions (1)
- papers-typed
wurgaft-2026-manifold.md