Editing NLA explanations to change 'reward' to 'penalty' produces steering vector that increases odd-number responses from near-zero to >70%, demonstrating belief capture upstream of behavior.

{
  "doc": "natural.md",
  "context": "Shows NLA explanations capture latent model beliefs about rewards before output selection; validates interpretability.",
  "category": "ai",
  "norm_label": "Editing NLA explanations to change 'reward' to 'penalty' produces steering vector that increases odd-number responses from near-zero to >70%, demonstrating belief capture upstream of behavior.",
  "graphify_id": "reward_awareness_steering",
  "source_file": "natural.md",
  "imported_from": "/Users/antonborzov/Documents/Research.nosync/papers/extract_typed_out/natural/graph.json",
  "extracted_type": "finding",
  "source_location": "§Reasoning about Rewards",
  "graphify_file_type": "finding"
}