On October 29, Wikimedia asked to submit a brief supporting neither side in the consolidated OpenAI copyright litigation.

The big change

  • Wikimedia argues general-purpose pre-training is likely fair use, unlike certain substitutive retrieval uses.
  • Publisher organizations contest fair use for unlicensed training itself, arguing that it harms markets for works and licenses.
  • These competing arguments ask the court to assess different claimed effects on creators and AI developers.

Two stages, two arguments

In its proposed brief, Wikimedia argues general-purpose pre-training is likely fair use because it builds a model with a different purpose from the original. It argues that certain retrieval-augmented generation, or RAG, uses likely are not fair when copied material answers the same informational demand and substitutes for the source.

Wikimedia recognizes uses complying with Wikipedia’s CC BY-SA license and notes some services cite and link to sources. Its argument does not establish that all retrieval is unlawful.

The publisher groups’ proposed brief supports class and news plaintiffs’ summary judgment motions. It argues training uses works for their expressive content and that substitutes and unlicensed training harm existing and developing licensing markets.

Procedural status

The current disposition of Wikimedia’s leave request was not independently verified. BIG CHANGE read both proposed briefs; neither is a judgment.

This is a further development in the same consolidated litigation covered in BIG CHANGE’s earlier report on Authors Guild book-training filings.