[gigel82]

Projects like these (using embeddings) are great, but what I'm looking for is something that can ingest an entire book (let's say a fiction book) then answer questions about the entire content (and not just by effectively doing a text search over your input, but actually "understanding" the entire contents of the book); I presume such a thing is not possible with ChatGPT (without fine-tuning), correct?

[mnkm]

What do you think about the responses generated by this:

https://www.konjer.xyz/the-alchemist (disclaimer: built by me)

What specifically is missing from the answers in your opinion?

[gigel82]

That's pretty interesting but ideally, I'd be able to upload my own book (txt, pdf, epub) and interact with it. It's lacking implementation details so not sure if you use embeddings, fine tuning or a novel approach.

[ElFitz]

Could using GPT3 (davinci-003) to generate embeddings, then searching your vector database for relevant excerpts, then providing the results as context for the prompt lead to something close enough?

[gigel82]

No. That works for documentation where you do text search and extract paragraphs around the results for "context".

I want it to understand a complete fiction book and tell me about how a character grows throughout their journey from chapter 1 to chapter 12 over 350 pages.

[ElFitz]

Depending on the book you could use that to extract all excerpts where the character appears.

Then each excerpt could be fed to the LLM asking it how this part relates to the question you’d want an answer to.

Then ask for each what it shows about the character and it’s personality, weaknesses, etc.

And finally recursively summarise them, asking for the summary to show how the character has grown through the summarised content.

Basically ending up with a map-reduce.

Bigger sources, or lots of content related to the character, would lead to less accuracy, and increase the likelihood of hitting the window’s limit.

It would also be highly specific and quite brittle, although one could probably turn it into a more generic process / pipeline (ie what dust.tt enables).

I might have completely missed your point or overlooked some glaring flaw though, in which case please do let me know what you think.

[nico]

Right now that’s not a use case supported out of the box by ChatGPT.

It also seems to be one of the most important limitations of ChatGPT, and a lot of people/teams are looking for solutions.

[DebtDeflation]

I work in consulting and this is literally the use case that every single client wants right now - the ability to ingest a corpus of documents into ChatGPT or similar and then have it generate responses based on natural language questions. Right now most people are faking it by running the search using some other tool like Solr/ES and then taking the snippets that are returned and assembling them into a prompt that gets passed to ChatGPT.

[nico]

Thank you, that’s very insightful.

Which option seems to you to be the best alternative? And where do you see the future of this?