{"type":"video","version":"1.0","html":"<iframe src=\"https://www.loom.com/embed/c733a715a2a7497fa374db8beb210288\" frameborder=\"0\" width=\"1920\" height=\"1440\" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>","height":1440,"width":1920,"provider_name":"Loom","provider_url":"https://www.loom.com","thumbnail_height":1440,"thumbnail_width":1920,"thumbnail_url":"https://cdn.loom.com/sessions/thumbnails/c733a715a2a7497fa374db8beb210288-a06e155595f75300.gif","duration":83.855,"title":"Decentralized Feedback Loop to Train Vision Models","description":"This Loom explains the motivation and feedback loop behind decentralized reinforcement learning with human feedback. The creator describes that after putting a 1.5 billion parameter LLM on a Raspberry Pi, they had to manually evaluate data for a long time. They plan to add a vision model on a device described as a Jessinor nano, which would require large amounts of video data and many hours of evaluation. To avoid doing this themselves, they propose paying people on buses, Uber rides, or trains to quickly label data, with more participation improving accuracy and creating a circular loop of data that trains the vision model and then provides data back to other evaluators."}