RAG Is Simpler Than You Think

1. Embeddings are often overhyped; traditional full‑text search is underrated
Many commenters argue that for technical or keyword‑heavy use cases, BM25/Lucene outperforms semantic search and that teams should start with keyword search before adding embeddings.

“If you are looking for anything technical at all, then keyword search almost always does better (in my experience). I'd actually recommend starting with keyword search, and then expanding with embeddings after you have a better idea of what your users are trying to determine.” — pantsforbirds
“People vastly underestimate full text search and vastly overestimate embeddings. FTS is really easy, portable and scalable and gets you very far…” — usernametaken29
“Start with BM25 and only add embeddings when keyword search actually fails you. Saves a lot of pain.” — trivet

2. RAG is essentially old information retrieval repackaged with new terminology
Several participants note that vector embeddings and retrieval‑augmented generation are not novel; they are established ML/IR techniques dressed up for the LLM era.

“It's just information retrieval packaged as something new.” — TudorAndrei
“Vector embeddings predate LLMs. They have been used as far back as the early 2000s. They are a general machine learning technique, rather than LLM specific.” — triangle
“RAG is basically good old information retrieval with LLMs doing the querying… Treating vector search as magic pixie dust that makes search great without effort is not necessarily going to work that well.” — jillesvangurp

3. Practical implementation hurdles: chunking, cost, vendor lock‑in, and operational overhead
Discussants highlight the challenges of embedding large documents, managing cross‑chunk relationships, escalating storage needs, and dependence on proprietary vector‑DB services.

“Embed everything the first time assumes your text is small. Try embedding pdf reports – though luck. It surely won’t fit into most embeddings.” — usernametaken29
“How are you gonna handle the relations that span across individual chunks… if a later chunk refers something from 2 chunks before using it…?” — freakynit
“Chunks can only be as large as the embedding model’s token limit, about 512‑1024 tokens usually. Anything longer gets truncated.” — harlanji
“Vendor lock in is 2025. Porting became trivial with LLMs advancing like they have.” — orisho