The wording in this article is a bit strange, why the extreme focus on the brand of storage media? Also, the term LLM seems to be used in a very broad way here, are they actually building a language model from scratch, or are they fine-tuning?