It's not about LLMs, it's about deploying, training, data management/pipelines, etc., for normal ML models.