When RAG Gets Bigger, Accuracy Doesn't Always Get Better

The first time a RAG system gets a large knowledge base, it often does not fail loudly. It gets subtly worse. The answers still sound plausible. The retrieved chunks still look related. The model stil...