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 reports using AI tools to search millions of digitized historical records and identify leads including a forgotten meteorite report and records of three rhinoceroses. He says he checked candidate findings against original document scans and specialist catalogues, but the historical significance of some leads still requires further verification.
Software engineer Jesse Waites says an AI-assisted search of millions of digitized historical records surfaced a forgotten meteorite report, references to three rhinoceroses, and evidence that may point to unrecorded volcanic eruptions. The findings are leads from an exploratory project, not a claim that every discovery has been accepted by specialists; Waites says he checked promising passages against scans of the original documents and existing catalogues.
Waites describes the work in a report explaining how he adapted an approach used by historian Benjamin Breen, who had searched Dutch East India Company records and reported finding a 1615 ship’s journal referring to sailors catching tortoises and dodos on Mauritius. Inspired by that example, Waites set out to search several historical questions in parallel, including lost animal records, meteorite falls, earthquakes and a major volcanic eruption dated to 1808 whose source has not been located.
The search drew on the GLOBALISE transcription collection, covering 4.35 million pages of Dutch East India Company records from the 1600s through the 1790s, as well as digitized Dutch and American newspapers and selected ship logs. Waites says his process created semantic representations for 5.7 million passages from the company archive, allowing searches to identify relevant meaning even when historical spelling or text-recognition errors obscured modern keywords.
To reduce the volume of material for detailed review, Waites used a small decision model to screen passages before sending a smaller set to a more capable model for translation and extraction of dates and places. He says the second stage was followed by examination of the original page scans and checks against specialist catalogues. This process, as described by Waites, helped identify the meteorite and rhinoceros material; the source report does not establish that all leads have been independently confirmed or published in peer-reviewed research.
How AI Triage Can Aid Archive Research
Historical archives contain far more material than any researcher can read page by page, and searchable text is often imperfect. The approach described by Waites uses semantic search and staged screening to direct attention to passages that ordinary keyword queries could miss, including documents with inconsistent spelling or transcription mistakes. That can help researchers generate candidates from large collections faster than manual reading alone.
The method does not make the machine-generated result a historical fact by itself. Its value depends on a person checking the original document, its date and location, and whether the record is genuinely absent from established catalogues. For readers, the project is an example of AI serving as a research aid rather than a substitute for archival interpretation. Potential findings could add records to narrow fields such as historical zoology, meteorite history and volcanology, if independent checks support them.
Waites also describes a cost-conscious division of labor: a cheaper model screened large numbers of passages, while a more capable model handled a much smaller group. He reports that screening 59,000 elephant mentions cost about US$3 with the initial model. That figure applies to the specific task and workflow he describes; it is not a general estimate for searching the full archive or validating historical claims.
As an affiliate, we earn on qualifying purchases.
From Dodo Records to Wider Searches
The immediate inspiration was Breen’s account of a new eyewitness reference to the dodo in a 1615 ship’s journal. The underlying records belong to the Dutch East India Company, a trading company that operated from 1602 until it was dissolved in 1799. GLOBALISE has made millions of the company’s handwritten pages available as searchable transcriptions, creating a resource for questions that may be difficult to answer through conventional catalogue searches alone.
Waites is a software engineer, not a historian, and frames his project as an attempt to extend the archive-search technique across several questions. He says an AI research assistant proposed 13 candidate topics, ranked partly by whether useful records were accessible online and whether any result could be checked against an original page. He then worked with Claude Code and other models to build a pipeline that included searching, screening, translation and image review.
That sequence matters because historical text recognition can confuse letters, omit words or flatten differences in spelling. Waites says “rhinoceros,” for instance, appears in many forms in 17th-century Dutch records. Searching for a modern spelling alone could miss relevant material. Semantic retrieval can widen the net, but its output still needs documentary checks to distinguish a real historical observation from a mistaken transcription or an irrelevant mention.
“I think there’s something honorable about adding to what we know.”
— Jesse Waites, describing his project
Which Archival Leads Hold Up?
The report describes findings and candidate discoveries, but the supplied account does not give enough detail to independently assess the meteorite report or the individual rhinoceros records. It does not provide the documents’ full archival references, the exact dates and locations for each animal, or a specialist assessment of whether the records are genuinely new to the field. Checking a scan is an important step, but it does not by itself establish that a record has not been discussed in unpublished work or another catalogue.
The possible volcanic evidence is also framed as material that may relate to unrecorded eruptions, rather than confirmed identification of a particular eruption. It remains unclear from the account whether these findings have been reviewed by historians, zoologists, meteorite specialists or volcanologists, or whether formal papers are planned. The reported search metrics and cost are Waites’s account of his own system, not an independent evaluation of its accuracy or performance.
Document Checks and Scholarly Review
The next step is to make the underlying references available for others to inspect and compare them with specialist catalogues and published research. Historians and subject experts can test whether the passages have been interpreted correctly, whether the records represent new evidence, and what conclusions the documents support. Original scans and archive citations will be central to that process.
Waites’s report presents the work as an ongoing contribution rather than a completed catalogue of settled discoveries. Further reporting or publication could clarify the dates, locations and evidence behind each lead, and whether experts accept any of them as additions to the historical record. Until then, the meteorite, rhino and volcano results should be understood as AI-assisted archival findings reported by Waites, with their wider significance still subject to verification.
Key Questions
What did the AI-assisted archive search find?
Jesse Waites reports finding a forgotten meteorite report, references to three rhinoceroses and records that may point to unrecorded volcanic eruptions. The report presents these as archival leads, not as findings independently confirmed by specialists.
Which historical records did Waites search?
He says the project used 4.35 million pages of Dutch East India Company archive transcriptions from GLOBALISE, alongside digitized Dutch and American newspapers and selected ship logs.
How did the project check AI-generated results?
Waites says candidate passages were screened, translated and reviewed against scans of the original documents. He also compared promising findings with specialist catalogues, though the supplied report does not provide enough detail to independently verify every result.
Have the findings been accepted by historians or scientists?
The source account does not report formal acceptance or independent specialist review. Whether the meteorite and animal records are genuinely new, and whether the volcanic evidence identifies previously unrecorded eruptions, remains unclear.
Source: hn
Columbus Day / Indigenous Peoples' Day Picks
long weekend sales
As an affiliate, we earn on qualifying purchases.
