Embeddings via ada-002 give us a way to update the model in real time. Using Weaviate, or another dense vector engine, it is possible to write "memories" to the engine and then search those with concepts at a subsequent inferencing step. The "document models" that the engine stores can be considered a "hot model".