MD

Happy to see this discussion. I agree with @Rijgersberg that it could be worthwhile to maintain a model for research or academic use that attempts to take compliance with source data licensing seriously. There's enough companies and research institutions today that look at the New York Times—OpenAI lawsuits and see legal liabilities lurking downstream that stand in the way of adoption. A truly open model that is serious about compliance will attract attention, as for instance AllenAI's OLMo and BigScience's BloomZ show.

BTW not sure the EU AI Act will 'wipe the field clean' — observers see loopholes, e.g. Alek Tarkovski here — but thinking about openness from the get go is probably a sensible idea especially for academic and research uses. Your 'market' will be smaller though.