Gregory Kurtzer — founder of projects such as Rocky Linux, Warewulf, and Apptainer — has launched OpenWALDO (Open Weights, Artifacts, Licenses, Data, Origins), an open source community project building a shared, governed corpus of AI training data.
Sponsored by CIQ, OpenWALDO “makes training data behave like open source input: named, reviewable, versioned, attributable, and verifiable — and carries that provenance into the models built from it,” the website says. This means models trained on the data can carry an AI Bill of Materials identifying the sources used to build them.
Additionally, the working toolchain is being developed in public and offers a common foundation that the community can build and improve together. For example, Kurtzer says:
- Model builders and researchers can help define what useful training material looks like.
- Companies can contribute public documentation, product information, research, and other knowledge they want represented in future models.
- Data experts, librarians, legal specialists, and reviewers can strengthen organization, licensing evidence, provenance, and quality.
- Infrastructure engineers can help the index, storage, verification, and training systems operate at massive scale.
Those interested in participating can join the OpenWALDO community on Slack and GitHub and learn more from OpenWALDO.
