Control measures how much influence and insight people have over an LLM process, its behavior, deployment, privacy, and access to outputs. section 10.2.3
RAG frameworks can reduce control when hosted tools obscure data storage, retrieval relevance algorithms, vector-database relationships, model selection, or response behavior. Privacy controls may also be limited. section 10.2.3
Fine-tuning provides more control through experimentation: teams choose and format datasets, select the model, adjust tuning settings, and manage each process step. That control requires substantial preparation and repeated, time-consuming runs. section 10.2.3
