TL;DR
Prime made for students and young adults
- Fast, free delivery for dorm and study essentials
- Prime Video and Amazon Music included
- Member-only deals
Software engineer Jesse Waites says an AI-assisted search of digitized historical records surfaced a forgotten meteorite report, references to three rhinoceroses and possible unrecorded volcanic eruptions. The findings are leads grounded in archival documents, not all independently established discoveries; specialists and original records remain essential to checking their significance.
Software engineer Jesse Waites says he used AI to search millions of digitized historical records, turning up a previously overlooked meteorite report, records concerning three rhinoceroses and possible evidence of unrecorded volcanic eruptions. The results, described in a project account inspired by historian Benjamin Breen’s search for a dodo record, are candidate discoveries that must be checked against original documents and specialist catalogues.
Waites built a workflow to search the Dutch East India Company archive, digitized newspapers and ship logbooks. The GLOBALISE archive transcriptions cover 4.35 million pages dating from the 1600s through the 1790s. Waites says his system processed 5.7 million passages from that archive in an overnight run, using semantic search to find relevant passages even when old spelling or poor text recognition would defeat ordinary keyword searches.
The workflow used AI at several stages. A relatively small model screened large batches of possible references, while a more capable model translated and extracted details from selected passages. Waites says an AI agent then opened scans of the original handwritten pages so he could check their text against the transcription. He also compared leads with specialist catalogues before describing them as new. In his account, human review remained part of the process: he chose research questions, judged which leads merited attention and decided when a search was not producing useful evidence.
Among the leads he reports are an old meteorite account, references to three rhinoceroses that had not appeared in the records he consulted, and possible evidence relevant to the search for an unidentified major eruption dated to 1808. The source account does not provide enough detail here to establish the exact contents, dates or independent scholarly status of each finding. They should be understood as archival leads, not as independently verified conclusions.
How AI Can Surface Archival Leads
Historical archives contain far more material than researchers can readily read page by page. Waites estimates that reading the Dutch East India Company pages alone at two minutes each, for a full working week, would take about 70 years. Search tools that can handle variant spellings and imperfect transcriptions may help researchers locate documents that conventional searches overlook.
The reported findings also show why automation does not remove the need for historical expertise. A model can sort passages and suggest translations, but a reference becomes useful evidence only when researchers examine the original page, establish what it says and compare it with existing scholarship. If the leads hold up, they may add details to narrow fields such as historical meteorite falls, animal records and volcanic history. The account does not establish that every lead will change those fields.
As an affiliate, we earn on qualifying purchases.
From a Dodo Record to Wider Searches
Waites says the project began after historian Benjamin Breen described finding a 1615 ship’s journal in the Dutch East India Company records. The account reportedly records sailors on Mauritius catching tortoises and dodos. Breen’s post, published October 1, 2026, prompted Waites to try a broader approach: searching multiple historical questions in parallel rather than looking for a single species or event.
Before building the search, Waites asked an AI research assistant to suggest questions that might be answerable using material available online and verifiable against original pages. His candidate topics included animals in company records, meteorite falls, felt earthquakes, vanished ships and an unidentified 1808 eruption. The source says the project also used Dutch national library newspapers and two centuries of American newspapers, alongside the company archive and selected ship logs. Digitized text is imperfect: Waites notes that handwriting, printing and inconsistent historical spelling can obscure records, including many variant spellings of “rhinoceros.”
“I think there’s something honorable about adding to what we know.”
— Jesse Waites
As an affiliate, we earn on qualifying purchases.
Which Leads Will Withstand Review
The source account does not include independent assessments from historians or other specialists of the meteorite, rhinoceros or eruption findings. It also does not supply full archival citations or enough detail to determine precisely how each lead compares with previous catalogues. The original records and the checks against specialist references are central to confirming whether a passage is new and what it can support.
AI-assisted transcription and translation can introduce errors, while semantic search may return material that is related to a query but does not prove the event or interpretation a researcher is investigating. It is not clear from the supplied account whether specialists have verified all three reported categories of findings, whether the results will be published in scholarly outlets, or whether further archival work could change their interpretation.
As an affiliate, we earn on qualifying purchases.
Original Records Need Specialist Checks
The next step is for researchers to examine the cited archival pages, verify the transcription and translation, and compare each record with established catalogues and historical studies. That work can establish whether the documents are genuinely overlooked, whether their contents support the interpretations reported by Waites, and whether the findings warrant formal publication.
Waites’s account presents the search method and its leads, but it does not announce a scheduled peer-reviewed paper or a timetable for specialist review. The status of the findings may become clearer as full citations and independent assessments become available. Until then, the records are best treated as promising evidence for further investigation, not settled revisions to historical knowledge.
semantic search tool for historical records
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What did Jesse Waites report finding?
Waites reports finding a forgotten meteorite account, archival references to three rhinoceroses and possible evidence relevant to unidentified volcanic eruptions. These are findings described in his account and require checks against the original records and specialist research.
What archives did the AI search?
The project included the GLOBALISE transcription of Dutch East India Company records, Dutch national library newspapers, two centuries of American newspapers and selected ship logbooks.
Did AI verify the discoveries?
No. AI helped search, screen and interpret passages, but Waites says he checked selected results against scans of original pages and specialist catalogues. Independent confirmation of the reported leads is not provided in the source account.
Why use semantic search instead of keywords?
Historical documents can contain inconsistent spelling, handwriting and text-recognition errors. Waites says semantic search looked for passages by meaning, which could help find relevant material that a spelling-dependent keyword search missed.
What remains unknown about the reported findings?
The supplied account does not establish whether specialists have independently confirmed each lead, how the records compare with all existing catalogues, or whether the results will appear in peer-reviewed research.
Source: hn
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
