Does this also apply to LLM training at scale? I would be a bit surprised if it does, fwiw.