#hey team got a question about embeddings
4 messages · Page 1 of 1 (latest)
like okay maybe the general approach is okay, but if I have a question that has only very little meaningful information the answer is not correct.
Something like if I have a paper and I ask "what is table 9 about?" the only meaningful information is "table 9" and it will get matched against a lot of wrong choices
maybe I can preprocess the question somehow to find the relevant keywords?
hm thanks yeah that could work if I had metadata or anyway to split the text into logical chunks. Unfortunately, it's just a stream of text and there's no way to know what is what