I'm curious about OpenAI's approach to licensing for their upcoming open-source models, especially in light of Meta's restrictions in Europe. Sam has suggested limitations on the use of Deepseek within the USA, so could we expect OpenAI to impose similar geographical restrictions on their open models? Is there a chance they'll follow Meta's lead and limit usage in places like Europe or China?
2 Answers
If a model is truly open source, it's hard to enforce limitations on where it can be used. Once the code is out there, it can be hard to control its access, unlike proprietary software.
Sure, they can try to control it through licensing, but it might not be effective in reality.
I think Sam would prefer the models to be available in Europe. The restrictions often come from the EU's own regulations, not from companies not wanting to release them. With China, the situation is more complicated—there's definitely trust issues between the USA and China, and Deepseek being a competitor plays into that too.
That makes sense, but look at Meta's Llama4. They specifically restrict use for anyone in the EU through their licensing terms. So, I guess it can be done, but it's tricky!