there’s so much uncertainty about legal status of training on data that a fully open licensed dataset creates a pathway that might feel like more legally secure for some actors

Mike, I agree. At @openfuture.bsky.social we commissioned a study on copyleft and derivation in AI dev, and I expected more follow-up conversations.

I also agree with @stellaathena.bsky.social that this is more a model-building, rather than dataset-building issue

There is so much talk about re-industrialization; but our collective imagination of “industry” is stuck in the previous era: we see steelmills, coal mines, car factories. And forget that they are today full of researchers, chips, and robots.

And environmental sustainability can't be an afterthought. The Action Plan acknowledges some energy and water concerns but a more robust framework is needed to limit AI’s environmental impact.

greenscreen.network/en/blog/with...

Who will orchestrate this complex plan? We need an agency with capacity to coordinate AI deployment across sectors while ensuring compliance with digital rights. Definitely more than just an “observatory” proposed in the Action Plan.

And more resources should go to projects like DARE at the Barcelona Supercomputing Center, which develops alternative, open chip architectures for AI training.

eurohpc-ju.europa.eu/research-inn...

Stronger data governance is needed, ensuring access but with sufficient guardrails.

The upcoming Data Union Strategy and proposed Data Labs are promising starts, but we need a proper Knowledge Commons to address what Stefaan Verhulst calls a "data winter."

sverhulst.medium.com/are-we-enter...

A major gap: no commitment to developing a flagship, state-of-the-art, open-source European frontier model. EU-based labs have proven they can build small models. It's time to pool their resources into a model that embodies European values at scale.

The EU correctly aims to reduce dependency on dominant commercial players with investments in compute. But success will require more than just mobilising funding for ever bigger amounts of GPUs.

Is joining a race the right goal for a public policy? A stronger vision of purposeful AI deployment is needed if it is to do more than just boost commercial AI development.

#publicAI

That’s a very interesting idea, and I agree that we need some form of regulation to force AI companies to reciprocate. Yet it’s hard to imagine such code just for Wikimedia - so how broad do you make it? Very quickly you probably have to consider a web-wide tax.

Still, the paper offers a positive vision, but little clarity on how to get there, if all the big labs are stuck in the transformer / diffusion paradigms.

(Upcoming #publicAI paper that Im writing argues that research into alternative paradigms should be key to any public intervention)

And it’s interesting to see #interoperability (not often discussed in the #AI space) as a key principle for model plurality.

Against this backdrop, @wikipedia.org stands out as a unique, global platform and civic initiative. High time to think of it as much more than an encyclopedia: a bulwark for XXI century, knowledge-based, democratic order. A collective good in a world of private tech empires.