Plus RAG isn't just referenced result augmentation. It isn't just adding a vector based db It isn't about less hallucination. In fact it doesn't even mandate

Plus RAG isn't just referenced result augmentation. It isn't just adding a vector based db It isn't about less hallucination. In fact it doesn't even mandate that result should be referenced nor come from a vector db. RAG's point is to remove the limit LLMs alone have which is that they are limited to the mind trained data as source of information. A RAG can be queried for information an LLM doesn't have any knowledge of. The LLM part of a RAG can be instructed to use all sort of information retrieval, such as making a web search, checking the current stock market value of any particular tickers. The article is nonetheless interesting as it touches on the beauty of LLMs being in their input interface rather than their ability to outputs aggregated content that reads beautifully, usually. RAG is more than what the author says it is. It's even what the author says is most wanted. RAG is more comprehensively what the author wants. It can be instructed to provide relevant references, always answer a particular way, and most importantly can get asked to perform (live) search engine queries to find the most up to date information. The example in the article is OK given the recipe is pretty old and has been very likely mined during training. What about: > I missed the games Lakers played from the 1st to 21st of April. Could you give me the list of opponents and scores please. LLMs input interface won't help. RAGs can make LLMs answer that.