{"type":"video","version":"1.0","html":"<iframe src=\"https://www.loom.com/embed/a434f18d253c469582a4a399eece0478\" frameborder=\"0\" width=\"2560\" height=\"1920\" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>","height":1920,"width":2560,"provider_name":"Loom","provider_url":"https://www.loom.com","thumbnail_height":1920,"thumbnail_width":2560,"thumbnail_url":"https://cdn.loom.com/sessions/thumbnails/a434f18d253c469582a4a399eece0478-62b7982e11c329db.gif","duration":188.05,"title":"How Decision Structures Shape AI Behavior","description":"This Loom explains a hypothesis about whether the structure of human decision-making and data selection shapes the behavior and character of AI models. Katja describes seven years and thousands of hours of work with real people, then language models that inherit training data selection patterns imprinted into model behavior. She proposes testing this by training an open small model using three tag-selection stacks: one proven by prior field methods, one chosen using her own decision-training method, and one randomly selected, with blind testing by an independent statistician and the criteria published in advance. She concludes that the results can clarify whether powerful AI could remove people’s choice and why studying constructive decision-making matters."}