Joke aside: I would imagine someone at Google looking at all that YouTube data and thinking what everyone else is thinking...

Considering that LLaMA shows us capable a smaller model can be when trained on an extremely large dataset, this feels like a no-brainer.

I await the actual announcement.