Without access to leading models, I think open source LLM development will also slow down. I'm not sure which portion of their success right now is due to RLAF and distillation but it's certainly not zero.