Historian argues GPT-6 Sol and Opus 5.5 can now solve open historical problems, and asks AI labs to fund the work

Res Obscura·low signalXBlueskyLinkedInCopy link

Benjamin Breen's Res Obscura post (2026-09-24, 139 points on HN) reports early results using this week's frontier models to decode 17th-century letters and trace alchemical knowledge, beyond simple transcription. He says models do best on problems that experts have already identified, whose sources are digitized, and whose answers can be proven or disproven, the same pattern seen in math. He wants labs and funders to back historian-model collaborations. This is a single practitioner's account, useful mainly as a checklist for picking tractable research problems in any domain.