How do AI ops engineers work on large LLMs like Llama 3?

I've been doing ops and compute for a long time but the LLMs are so new I'm not really sure how people in my field are writing software / creating products that work on these models.