Al weights are not open "source"

The main point of the article is that AI licensing is complex and requires more nuanced terminology than simply "open source" or "proprietary."

The author argues that:

  • AI has multiple components (source code, weights, data) with different licensing needs.
  • AI poses unique socio-ethical concerns requiring restrictions beyond traditional software licenses.
  • Current terminology like "open source" is misused for AI components, leading to confusion.

The author proposes a new categorization of AI licenses to avoid ambiguity and promote responsible development:

  • Proprietary: Closed licenses requiring explicit permission for any use.
  • Collaborator: Non-commercial, NDA licenses for specific research groups.
  • Available: Non-commercial, public licenses for personal use.
  • Ethical: Licenses allowing commercial use but with restrictions on fields or behaviors.
  • Open: Licenses adhering to the strict Open Source Initiative definition.

By using accurate terminology, the author hopes to improve communication, prevent misleading claims ("open washing"), and ultimately establish clear standards for responsible AI development.

Related post


Open-Source LLMs Gain Enterprise Traction

Open-source large language models (LLMs) are seeing initial enterprise deployments and adoption, but closed-source models currently have more publicly known use cases. However, open-source advocates argue it's only a matter of time before open-source catches up as more models are released. OpenAI's ChatGPT and other closed LLMs were released before…

AI Ethics

OpenAI's New Licensing Strategy

OpenAI is initiating content licensing deals with publishers to use their content ethically for training its AI models, amid industry-wide challenges regarding intellectual property rights and sustainable AI development. READ MORE