We finally have a definition for open-source AI
A new open-source standard clarifies that true AI openness requires unrestricted use, inspection, modification, and sharing, alongside transparent access to training data, code, and weights. This definition addresses the ambiguity where major tech companies marketed restricted models as open source, preventing genuine scrutiny of their inner workings. For open data, this distinction is critical because it establishes rigorous criteria for transparency in model components and datasets. Without such standards, vague labeling undermines trust and hinders independent research, ensuring that only truly accessible systems qualify as open source.
Source: technologyreview.comPublished on 2024-08-23