In the end, what is needed is improved search accuracy
hrjn I originally thought that NFT was finished even before it started because it has little substance, but LLM is a rather revolutionary technology, but at the same time, in order to use it properly, you need to have some background knowledge of conventional natural language processing. I think that LLM is a revolutionary technology, but at the same time, it requires background knowledge of conventional natural language processing in order to use it properly, so I don't think ordinary people will be able to handle this one either. nishio >After all, conventional natural language processing ChatGPT is awesome! I thought ChatGPT was great, but when I tried to use it, I found that it was too short for my needs and too long for my tails, and that search is better for finding what I'm looking for, and that conventional technology is better because it doesn't have halcination if the summary is "human-readable" enough.
nishio Generation AI is effective when used properly for generation, but what do customers want to generate in the first place? When I think about it, most customers' needs are "to find the important parts among a lot of things that are hard to read", and even if they finally generate, there is "to find materials for generation" before the generation, so search is necessary after all. nishio The story is slightly different and I don't think "conventional technology is better" at all. I already use neither conventional search nor my own vector search almost exclusively RAG. I think that the domain of competition is to combine and re-rank sparse and dens search when increasing the usefulness of it. ---
This page is auto-translated from /nishio/結局必要なのは検索の精度向上 using DeepL. If you looks something interesting but the auto-translated English is not good enough to understand it, feel free to let me know at @nishio_en. I'm very happy to spread my thought to non-Japanese readers.