Identification of algorithms implemented in attention layers, distributed across attention heads

{
  "doc": "paper.md",
  "context": "VPD successfully recovered interpretable attention algorithms (previous-token behavior, syntax-boundary routing) in weight space without requiring manual decomposition across heads.",
  "norm_label": "Identification of algorithms implemented in attention layers, distributed across attention heads",
  "graphify_id": "attention_algorithm_discovery",
  "source_file": "paper.md",
  "imported_from": "/Users/antonborzov/Documents/Research.nosync/papers/extract_typed_out/paper/graph.json",
  "extracted_type": "finding",
  "source_location": "§2.1",
  "graphify_file_type": "finding"
}