This is a good framework! I think OS LMs are awesome and should be supported, *and also* there will be tiers of foundation models that would pose a national security threat if OSed+finetuned. (Skeptical even GPT-4 is there yet but glad the EU agrees we’ll get there eventually.)