OSI Criticizes Meta's Open Source AI Claims

Meta Platforms is facing criticism for its labeling of LLaMA, its large language AI models, as 'open-source,' with the Open Source Initiative (OSI) asserting that this mischaracterization confuses users and undermines the true meaning of open source. OSI's executive director, Stefano Maffulli, highlighted that for a model to be truly open source, companies must disclose training algorithms and model weights, which Meta's licensing does not allow. He described the situation as 'extremely damaging' amid a push for genuine open-source technologies by regulatory bodies like the European Commission. Meanwhile, the open-source AI landscape has also been stirred by the leak of a new model, 'miqu-1–70B,' which reportedly matches or outperforms existing models like Mistral's Mixtral and even competes with OpenAI's GPT-4. The model's emergence has generated significant buzz in the AI community, raising questions about legality and ethical considerations. Meta responded to the criticism by stating that existing definitions of open source do not adequately address the complexities of modern AI models.