I've been playing a lot with the langchain Document Loaders - their examples make the results look more impressive than they are, or perhaps that's because our

I've been playing a lot with the langchain Document Loaders - their examples make the results look more impressive than they are, or perhaps that's because our documents are 'special' (heavy in taxonomy/biology). I've been priming it with 'if you don't know, say that you don't know, don't invent anything' and it usually ends up saying 'I don't know' to questions it can clearly answer. But yeah, working on that part!