OSI Defines Open Source AI: What It Means Now

The Story: The Open Source Initiative (OSI) has officially defined what constitutes "open" artificial intelligence (AI), creating potential conflict with companies like Meta, whose models do not meet these standards. For an AI system to be recognized as genuinely open-source, it must provide full details on training data, complete source code, and training settings, challenging the openness of popular models like Meta's Llama, which are limited in their commercial use and lack transparency in training data access.

The Details:

• OSI's definition mandates that AI models disclose the data used for training to ensure transparency and reproducibility, a deviation from current tech practices.

• Meta's Llama models, dubbed the largest open-source AI, fall short of OSI standards due to restrictions on commercial use and non-disclosure of training datasets.

• The OSI engaged diverse stakeholders for two years to establish its definition, highlighting the importance of collaboration in setting standards in the evolving AI landscape.

• Critically, the OSI’s guidelines aim to curb the trend of "open washing," where companies mislabel their products as open-source to evade scrutiny.

• Industry leaders like Hugging Face's CEO commend the OSI's work as pivotal for fostering genuine openness in AI, particularly emphasizing the role of accessible training data.

Why It Matters: OSI’s new definition raises critical questions about the authenticity and transparency of major AI models, particularly as the industry grapples with legal challenges surrounding data use. For creative professionals, understanding these definitions is crucial as they navigate the content they create or utilize, ensuring they're collaborating with truly open resources. With regulatory bodies starting to pay attention, aligning with genuine open-source principles could pave the way for innovation that respects creators’ rights and promotes ethical practices in the AI space.