Robotistry on Nostr: npub14cwgn…yhfwc npub1w83sd…yget0 (Moving away from the details of the article) I ...
npub14cwgnf5j3hrqrvasct3fk3tf9zkkrzqp9svannkgsfv8kl3rv5aqyyhfwc (npub14cw…hfwc) npub1w83sdzs0y9vnvpfy6222vdmqgpkvhq69x6nk620wyxuuzzlar0js4yget0 (npub1w83…get0) (Moving away from the details of the article) I don't actually think RAG solves the core problem.
It just shifts the verification burden onto the dataset used to constrain the responses. It doesn't solve the grounding problem - the LLM still doesn't have contextual understanding of the material in the constraining dataset, and we will still run into issues with people attributing levels of understanding that simply don't exist.
It just shifts the verification burden onto the dataset used to constrain the responses. It doesn't solve the grounding problem - the LLM still doesn't have contextual understanding of the material in the constraining dataset, and we will still run into issues with people attributing levels of understanding that simply don't exist.