The concept of open source has been a cornerstone of technological innovation, particularly in the realm of artificial intelligence (AI). However, a recent debate has emerged around the validity of the term open source in the context of AI models, with the Llama-3 model serving as a focal point. This October, the Open Source Initiative (OSI) is set to provide a definitive clarification on what constitutes true open source in AI.
Open source, in its traditional sense, refers to software whose source code is freely available for use, modification, and distribution by anyone. It fosters collaboration, transparency, and innovation. In the AI landscape, this philosophy has led to the development of numerous open-source algorithms, benefiting both individual developers and large tech companies. The competitive landscape in the AI domain has bifurcated into open-source and closed-source models, with the former often seen as a catalyst for advancements.
Recently, the release of Llama-3.1, a powerful open-source AI model by Meta, sparked discussions on the nature of open-source models. Despite its impressive download numbers,质疑 arose due to restrictions on its usage and data availability, leading some to question whether it truly adheres to the spirit of open source. Similarly, Mistral, a startup, also announced the open sourcing of its Large 2 model, further complicating the conversation.
The OSI, recognizing the need for clarity in the rapidly evolving tech sector, has proposed a new draft definition for open-source AI. This move aims to address the ambiguity often associated with the term open source. The proposed definition comes at a crucial time, following a wave of open-source AI model updates, including Meta’s Llama 3.1 and Mistral’s release.
Under the OSI’s traditional open-source software standards, Meta’s Llama models, despite being accessible, do not meet the criteria due to usage limitations imposed on companies and specific content generated through the model. Similarly, the Flux image-to-text model is also not considered truly open source due to restrictions on its weights or the lack of accompanying training data. The OSI has coined terms like open weights or code available to describe these models with limitations.
To address the open-source dilemma, the OSI has assembled a team of around 70 experts, including researchers, lawyers, policymakers, and representatives from tech giants like Meta, Google, and Amazon. Their proposed four essential freedoms for open-source AI models mirror those for open-source software: the freedom to use for any purpose, study the inner workings, modify, and redistribute the model, whether modified or not. These freedoms ensure transparency, enabling researchers to analyze AI models’ inner workings and potentially expose security vulnerabilities, as opposed to closed-source systems like OpenAI’s ChatGPT.
Scheduled to announce the final version of the open-source AI definition at the All Things Open conference in October 2024, the OSI’s efforts aim to facilitate informed decision-making for developers, researchers, and users when engaging with AI tools. According to OSI’s Executive Director Stefano Maffulli, AI is different from conventional software, requiring all stakeholders to reconsider the applicability of open-source principles.
The first batch of models expected to meet the new open-source criteria includes EleutherAI’s Pythia, Ai2’s OLMo, and the LLM360 collective. By providing a clear definition, the OSI emphasizes the importance of unlicensed innovation and the need for users to maintain control and dominance over technology.
In a May press conference, the OSI underscored the significance of defining true open source in AI. Maffulli highlighted, OSI believes in the empowerment and control of technology by everyone. We recognize that clarity in definition is crucial for fostering a vibrant and inclusive AI ecosystem.
As the tech community awaits the October clarification, the debate around the authenticity of open-source AI models underscores the ongoing struggle to balance innovation with accessibility and transparency. The OSI’s upcoming definition will undoubtedly shape the future direction of AI development and collaboration.
【source】https://www.jiqizhixin.com/articles/2024-08-28-3
Views: 0