Ai2 shows olmOCR 2 in action: searchable archives and fine-tuned pottery transcription
allen_ai · x · 2026-10-29
Ai2's olmOCR 2 turns document images into searchable text with open weights and code, letting researchers run it locally or adapt it to their collections. Two historical/archaeological projects showcase it:
- ArtificeOCR: Historian Maurice Casey connected olmOCR 2 with Tropy, an archival photo organizer — transcribing images locally so his library is keyword-searchable while material stays on his machine.
- CENTURIA: TU Wien & Austrian Academy of Sciences researchers fine-tuned olmOCR 2 on handwritten pottery records from Carnuntum, Austria; it made fewer transcription errors on held-out records than the unadapted model, and the team aims to apply it to the full archive.
Casey notes local execution was the most important feature for him.
Related event: Ai2 Open-Sources olmOCR 2 for Historical Document Transcription(2 posts)→
More from Apps
- Unify launches AI-powered sales dialer after 112K+ calls, claiming 34% more reachable numbers than Apollo — kimmonismus · 2026-10-29
- Ex-Sora user says export is incomplete, OpenAI can't verify 3 months after shutdown — Realistic_Winner_493 · 2026-10-29
- Anthropic launches tools showing companies what their AI spending buys — TechCrunch AI · 2026-10-29
- Google launches Connected Apps Platform: MCP apps inside Search and Gemini — VeryWellVersed · 2026-10-29
- Indie dev launches SwapStudio: browser-based AI video character swap on ComfyUI, prepaid credits — Inevitable-Bus-8654 · 2026-10-29
- Instinct partners with Shopify so your AI agent can check out anywhere using Shop Pay — mon__lim · 2026-10-29