That’s the only effective way to get more compute in current production LLMs, but the field is evolving.