So in this case it was just pure ChatGPT - I suspect they’ve ingested a lot more than they’re maybe allowed to do it worked out. I did a pdf experiment here though and it also worked well https://twitter.com/frozenfire42/status/1645855227348910080?s=46&t=kLxAT6M-ysp2m2lKjBOBxg

Expand full comment
Apr 30Liked by Cameron Armstrong

Thanks so much for compiling this - a bunch of these were on my list and now I want to see how many more I can try to replicate this approach for.

Did you purely use ChatGPT for this, or link the api up to pdf versions of the books or similar?

Expand full comment
Apr 2Liked by Cameron Armstrong

Any tool out there that allows you to skip all the embedding + pinecone + retrieval setup and go straight into doing Q & A with long documents?

Expand full comment