OpenWALDO launches to build collaborative community for open-source AI

OpenWALDO, an open-source artificial intelligence project backed by Ctrl IQ Inc., has debuted under the leadership of Gregory Kutzer, the creator of Rocky Linux, CentOS and Apptainer. The initiative seeks to establish a community-driven, open-source-governed repository of AI training data, offering a platform akin to Hugging Face Inc., which primarily offers open-weight models.

Open-weight models are advantageous as they empower users to download and operate AI models on any device, be it consumer hardware or cloud-based services, without the necessity for specialized proprietary equipment or concealed within the proprietary confines of dominant corporations.

However, open-weight models fall short in a crucial aspect: they lack the code, methods, recipes, and training data essential for reproducing them. This is where open-source models step in. According to Kutzer, "I’ve spent my career watching open-source turn users into builders, competitors into collaborators, and shared problems into common infrastructure that operates at massive scale." "No single organization could build or sustain all of that alone. OpenWALDO brings that proven model to AI."

OpenWALDO is proposing an AI Bill of Materials, which goes beyond open weights by incorporating open source elements, including full training data, object references, documents, tokens, inventory, and licenses as proof of the content required to rebuild the model from scratch. The project emphasizes that each assertion is traceable, rewritable, and correctable in a public setting.

This means all information is accessible, allowing for the review of any open model to determine if it contains any copyrighted or "copyleft" content that might compromise the source and potentially hold individuals using it legally accountable for its outputs once fully trained. This is the crucial aspect of "Know before you train." It represents a significant advancement above open weights.

Kutzer and the OpenWALDO team recognized that the current industry has resulted in a culture where numerous AI teams develop their foundation training material from scratch, piecing together recipes and duplicating efforts globally. This project aims to provide a community-built resource that will archive best practices and build upon them with verified baselines of what works, enable experimentation in public, and create auditable lines that can be evaluated collaboratively.

The project will be accessible to anyone, including individuals, hobbyists, researchers, institutions, and corporations. The goal is to enable everyone to learn together, construct upon this public foundation for the public good, and create a sustainable AI commons.

As Kutzer stated, "Open source has won this argument before." Proprietary vendors once made identical claims about open code, asserting it as insecure, unaccountable, and untrustworthy. However, they lost because source, licenses, and provenance made trust apparent. Similar to the broader open-source community, the aim of OpenWALDO is to create a space for AI open source to thrive, learn, and innovate.

"Let’s work together, build its foundation in the open, and collaboratively advance AI to the next level," Kutzer concluded.