This talk explains why document parsing is still hard and why vision-first methods outperform OCR for complex layouts, merged …

source