Members-Only
Recent Talks & Demos are for members only
You must be an AI Tinkerers active member to view these talks and demos.
OCR? More Like OMG: When AI Can’t Handle Your PDFs
Explore common AI failures in understanding PDFs. This talk compares models like GPT-4o and Gemini Flash, showcasing solutions to these challenging OCR problems.
From doing evals while building Paper2Audio, we have a growing collection of pages from PDFs that models are are having a hard time understanding. There are some really epic fails!
We’ll show the most interesting problems we’ve found, and different ways we’ve found to solve some of them. We’ll compare popular, general models, like GPT-4o and Gemini Flash 2.0 (which we’ve found generally works better at 4% of the cost), as well as some more specialized models people will be less familiar with, like MinerU and Marker.
One thing to clarify- this won’t be a demo of Paper2Audio generally, since I think the PDF understanding part is what will be most interesting to everyone at AI Tinkerers.
Compose Email
Loading recent emails...