2024 End-of-Year Review: Open Source AI Definition v1.0

Favorite

The release of version 1.0 of the Open Source AI Definition (OSAID) marks an important milestone on a journey to ensure that AI systems are innovative and aligned with the foundational principles of Open Source: the freedoms to use, study, modify and share.

Drafting a definition through collaboration

The OSAID is a testament to the power of global collaboration. Over the past two years, OSI convened a coalition of stakeholders—developers, data scientists, legal experts, policymakers and end users from all over the world. This diverse group coalesced through in-person workshops, online town halls and intensive co-design sessions to craft version 1.0 of the definition.

Key milestones of 2024

System Analysis: As part of the co-design process, working groups were formed to discuss which AI system components should be required to satisfy the four freedoms for AI. This included assessing how data, models, training methods and legal agreements adhere to Open Source principles. The analysis provided invaluable insights into the gaps that exist in current AI practices and outlined actionable steps to bridge these gaps. It also helped refine the OSAID’s criteria, ensuring they remain both practical and comprehensive.

System Evaluation: Several AI systems were assessed against the OSAID’s criteria. While models like OLMo (AI2), Pythia (Eleuther AI), CrystalCoder (LLM360) and T5 (Google) met the requirements, others like LLaMA2 (Meta), Phi-2 (Microsoft), Mixtral (Mistral) and Grok (X/Twitter) fell short, spotlighting the critical need for transparent frameworks in AI development.  Other models such as BLOOM (BigScience), Starcoder2 (BigCode) and Falcon (TII) would pass if they changed their license. This evaluation process also revealed areas where certain models could improve to better align with Open Source standards, demonstrating the OSAID’s role as a constructive guide for future developments.

Stable Release of OSAID 1.0: After extensive global consultation, OSI released the first stable version of the definition at the All Things Open conference in Raleigh, NC. This marked the culmination of two years of dialogue, research and iteration. The stable release provides a comprehensive framework to evaluate AI systems against the core principles of openness.

Global Endorsements: The OSAID has garnered endorsements from over 20 organizations, including Mozilla Foundation, Eleuther AI, CommonCrawl Foundation and the Eclipse Foundation, alongside support from more than 100 individuals. These endorsements validate the OSAID’s importance and its potential to shape the future of AI development.

Events and conferences

Throughout 2024, OSI actively participated in 23 conferences from around the world to engage with diverse communities. Highlights include FOSDEM (February – Brussels), Columbia Convening on openness and AI (February – New York), Open Source Summit NA (April, Seattle), PyCon (May – Pittsburgh), AI_Dev Europe (June, 2024 – Paris), OSPOs for Good (July, 2024 – New York), KubeCon + AI_dev Hong Kong (August – Hong Kong), Open Source Congress (August – Beijing), Deep Learning Indaba (September – Dakar), India FOSS (September – Bengaluru), Open Source Summit Europe (September – Vienna), Nerdearla (September – Buenos Aires), Training Data in OSAI (October – Paris) and All Things Open (October – Raleigh). A full timeline of in-person and online events is available here.

Publications and voices of the OSAID

The OSI published over 60 blog posts about Open Source AI in 2024. One of the highlights is the Voices of the OSAID series that we ran with stories about a few of the people involved in the Open Source AI Definition co-design process, featuring 10 volunteers who have helped shape and are shaping the definition. These stories highlight the diversity and passion of the community, bringing a human element to the often technical discussions around Open Source and AI.

Press coverage

The work around the Open Source AI Definition was cited over 180 times in the press worldwide, educating and countering misinformation. Our work was featured at The New York Times, The Verge, TechCrunch, ZDNET, InfoWorld, Ars Technica, IEEE Spectrum, MIT Technology Review, among other top media outlets.

Looking ahead

The release of OSAID 1.0 is not the end but the beginning of a new chapter. As we transition into 2025, OSI remains committed to continuing regular updates to the definition and evaluating AI systems and licenses to ensure alignment with Open Source principles.

As the Open Source community moves forward, your involvement is more critical than ever. We encourage more organizations and individuals to endorse and implement the OSAID. By broadening its reach, OSI aims to establish the OSAID as the global benchmark for open AI systems. Together, we can ensure that AI remains a tool for permissionless innovation.

OSI extends its deepest gratitude to the sponsors, volunteers and participants who made 2024 a banner year for Open Source AI. Let’s continue to build a future where technology serves everyone, everywhere. As we celebrate this year’s accomplishments, we look forward to what we can achieve together in 2025 and beyond. Please consider donating to or sponsoring the OSI.

Click Here to View Original Source (opensource.org)