We finally have a definition for open-source AI
A new standard defines true open-source AI, requiring unrestricted use, modification, and sharing, alongside full transparency of training data, code, and weights. This addresses the ambiguity where major tech companies previously marketed restricted models as open, misleading users about their accessibility and verifiability. The lack of clear definitions allowed corporations to exploit the term for marketing, creating a perception of trustworthiness without granting researchers the ability to independently inspect or adapt the systems. This opacity undermines the core principles of open collaboration and scientific scrutiny in artificial intelligence development. This standard is vital for open data as it establishes rigorous criteria for accessibility and transparency. It ensures that AI systems are genuinely available for community inspection and modification, promoting accountability and preventing the misuse of "open-source" labels to hide proprietary constraints or incomplete data.
Source: technologyreview.comPublished on 2024-10-07