7B beats o3, GPT-5! Medical AI agent teaches models to 'see where and how to see'
Researchers have developed new AI paradigms for medical imaging and video analysis, enabling models to actively "look" at evidence rather than just passively process it. These "Think with Images" and "Think with Videos" approaches allow AI agents to use visual tools to re-examine key areas or moments, correcting their own judgments with new evidence. This marks a shift from AI that merely explains to AI that reasons using visual data, potentially reducing hallucinations and improving interpretability in clinical settings. AI
IMPACT Establishes a new paradigm for medical AI, shifting focus from explanation to evidence-based reasoning, potentially improving trust and accuracy.