e.g. when i insert a new 'document' saying 'hello / good morning', the results of an index about a research paper get pretty weird, unless i specify top k
I don't know of any bugs with insert(), but I see the similarity is 0.76 on that dummy document. I think the embeddings are less helpful with short documents like that. I suspect it would work fine with actual document text
@Krrish@LiteLLM.ai by default the top k is 1 (which tbf is not a great default top k).
Also echoing @Logan M's point, using a vector store index for a "summarization"-esque question like "What is the paper about" is not a great fit. for those types of questions i'd recommend using a list index
