Ability to extract data from documents

I was super excited about the ability to upload documents and have ChatGPT do stuff with them. My two attempts so far were not awesome. Not complaining, just reporting the experience. Maybe others are having better luck.

First Attempt
I uploaded my 70 page Netflix DVD history report (.pdf) and asked it to extract title, rating, and watch date as a CSV. I even pointed it to the specific range of pages where this table existed.

Challenges were that years were headers followed by a continuation of the table and no column headers existed. Also the ratings were a 5 star image, and it was able to deduce that and claim that it would just count the stars to get that figure. Ultimately it gave up and output nothing.

Second Attempt
Created a GPT to help me plot some future episodes of an action adventure series. I uploaded the first book as a .epub.

As I was in the GPT creator mode, it read this book, and one 10000 character chunk at a time, read the book and made comments on the sections that sounded like an NYT literary review. It was insightful and I was amazed at how it figured out where I was leaning into certain themes and how they all worked together. I was overjoyed.

Then, when I was about 2/3rds of the way through (review… ‘continue’… review… ‘continue’…) when I stupidly chose to hit the save button on the GPT.

This had the effect of immediately discarding the discussion I was having with it (and it’s not linked to in my sidebar). And presenting, I suppose, the finished GPT, ready to converse with me.

I tried to get it to do the same stuff, summarize the book. But this time it complained that there was something wrong about the .epub, and it couldn’t read it and there was just no help for it.