[D] In terms of RAG research, why does it seem like a lot of people aren't working on the retriever?

I'm someone who conducted research in NLP a few years ago, stopped and joined industry, and am recently trying to get back on top of things. I've taken an interested into RAG-related work and have started reading some papers.

My understanding is that for RAG you have the retriever and the generator. For the generator it seems like using various LLMs is standard but the retriever also seems to be set to using something like BM25 or the DPR that …

get back industry machinelearning nlp papers people rag reading research terms work