I spent 12 days amassing jpegs via microfiche computer in the Washington DC public library. My project is to reprint the daily local newspaper columns from the 1930s and 1940s. Standard OCR completely choked on the aged, three-column newspaper scans, introducing hundreds of typos per page. Transcribing by hand was taking 30+ minutes per column. With ChatGPT vision, I feed it the high-res crop and it transcribes the entire archaic font with near-perfect accuracy in about 15 seconds, preserving formatting and correcting for faded ink. It turned an impossible multi-year slog into an active, daily historical publication project.
- Text
- verbatim
- Model
- ChatGPT
- Source
- Reddit ↗
- Published
- 2025-05
- Added
- 2026-09-14