I don't know about 'behind'. LLMs are a known technology, and training them is still a huge undertaking.

I can imagine a group coming in and finding a much more efficient training system, and eclipsing OpenAI.

The AI aren't self-improving entirely on their own yet, so the race is still a race.