One-stage CoT (QCM→RA) shows 12.31% accuracy drop vs. no-CoT (QCM→A) on ScienceQA; two-stage framework (rationale generation + answer inference) achieves 85.31% accuracy with vision features

{
  "doc": "zhang-2023-multimodal.md",
  "context": "Empirical evidence that naive one-stage CoT fails in language-only setting; two-stage + vision achieves state-of-the-art.",
  "norm_label": "One-stage CoT (QCM→RA) shows 12.31% accuracy drop vs. no-CoT (QCM→A) on ScienceQA; two-stage framework (rationale generation + answer inference) achieves 85.31% accuracy with vision features",
  "graphify_id": "finding_one_stage_vs_two_stage",
  "source_file": "zhang-2023-multimodal.md",
  "imported_from": "/Users/antonborzov/Documents/Research.nosync/papers/extract_typed_out/zhang-2023-multimodal/graph.json",
  "extracted_type": "finding",
  "source_location": "Tables 2-3",
  "graphify_file_type": "finding"
}