Comment by vlovich123

Doubling down on LLMs specifically (which came out of their own research). Also getting to a point to commercialize it first. Their TPU fleet wasn’t really optimized for LLMs until recently afaik.