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 used a multistage AI workflow to search historical archives and newspapers, identifying a meteorite report, records of three rhinoceroses and reports that may document previously unrecorded volcanic eruptions. Waites says he checked promising results against scans of original documents and specialist catalogues; the findings are leads for historical research, not proof that every discovery is new.
Software engineer Jesse Waites says an AI-assisted search of millions of historical records turned up a forgotten meteorite report, records of three rhinoceroses and possible accounts of unrecorded volcanic eruptions. The results are candidate historical evidence rather than a blanket confirmation of new discoveries: Waites says he checked promising passages against original document scans and the catalogues used by specialists.
Waites describes building a search workflow after reading historian Benjamin Breen’s account of finding a 1615 eyewitness reference to dodos in Dutch East India Company records. Breen’s find inspired Waites to search for several kinds of historical evidence at once, using a mix of AI models and digitized collections rather than relying on exact keyword matches.
The materials included 4.35 million pages of Dutch East India Company transcriptions, Dutch national library newspapers, two centuries of American newspapers and selected ship logbooks. Waites says his system converted 5.7 million passages from the Company archive into mathematical representations of meaning, allowing it to retrieve relevant passages despite inconsistent historical spelling and transcription errors.
To manage the volume, a small model screened candidate passages with narrow questions, such as whether a passage described a real wild animal and where it was located. A larger model then translated and extracted details from the selected passages. Waites says he or the research agent checked the original page scans and compared candidate findings with specialist catalogues before treating them as potentially new evidence.
How AI Can Triage Historical Records
The project shows one possible role for AI in archival research: finding and prioritizing passages across collections too large for a person to read page by page. Waites estimates that reading the Dutch East India Company material alone at two minutes per page would take about 70 years of full-time work. He says the automated search processed the collection in an overnight run.
That speed does not make the system a substitute for historians or archival verification. Historical text recognition can misread handwriting and print, while unusual spelling makes ordinary searches unreliable. The workflow’s value, as Waites presents it, is to narrow a large field of possibilities so a researcher can inspect original documents and establish whether a passage supports a claim.
The findings could give researchers additional leads about animal history, meteorite falls and volcanic activity. Their importance depends on what the source documents actually say, whether the records are correctly dated and located, and whether they add evidence beyond existing catalogues.
As an affiliate, we earn on qualifying purchases.
From Dodo Record to Broader Searches
Waites traces the project to a post by historian Benjamin Breen, published October 1, 2026, about using AI to find a previously overlooked dodo account. The record came from digitized Dutch East India Company material made searchable through the GLOBALISE project. The company operated from 1602 until its dissolution in 1799, and its records include handwritten accounts produced across its trading network.
Waites wanted to extend that approach beyond one animal or archive. He asked an AI research assistant to suggest historical questions that might be answerable from online data and checkable against original pages. The resulting candidates included animals in company records, a reported large volcanic eruption in 1808 whose location has not been identified, historical earthquakes, meteorite falls and missing ships.
Because language models can produce confident errors, Waites says he designed the process to end with an inspectable source: a scan of a handwritten letter or printed newspaper and an archive reference. He also describes remaining involved in directing the work rather than accepting model output without review.
“Adding even one new data point to a couple of niche fields felt like a small but worthwhile contribution to make.”
— Jesse Waites, describing his research goal
Which Finds Are Independently New?
The source report does not provide enough detail here to independently assess every candidate record, including the exact text and archive references for the meteorite and rhino findings. It also does not establish that every result has been accepted by relevant historians or added to authoritative catalogues. Waites says he compared candidate findings with specialist resources, but the material supplied does not list those checks in full.
The possible volcanic records also require care: the source describes them as unrecorded eruptions, but it does not establish that the passages prove an eruption occurred or identify its location. Old transcription errors, ambiguous descriptions and uncertain dates can all affect interpretation. The distinction between a newly surfaced document and a genuinely new historical conclusion remains important.
Document Checks Will Shape the Findings
The next step for the candidate discoveries is close review of the cited original pages by Waites and researchers with relevant expertise. That review can test whether the transcription is accurate, whether the passage means what the AI system inferred, and whether the evidence is already known under another name or description.
The source report does not announce a formal peer-reviewed publication, an independent audit of the full workflow or a confirmed schedule for further results. Waites’s account presents the work as an ongoing research effort. Readers and specialists will need the underlying citations and scans to evaluate the meteorite, animal and eruption records individually.
Key Questions
What did the AI-assisted search find?
Waites reports a forgotten meteorite report, historical records of three rhinoceroses and accounts that may relate to unrecorded volcanic eruptions. The source frames these as findings or candidate evidence from archival research, not as independently validated conclusions in every case.
Did the AI make the discoveries by itself?
No. Waites says he designed and directed the workflow, reviewed promising candidates and used AI models to search, screen, translate and extract information. He says the process included checking original document scans and comparing results with specialist catalogues.
Which records did the project search?
The search included 4.35 million pages of Dutch East India Company transcriptions, Dutch national library newspapers, two centuries of American newspapers and selected ship logbooks. Waites says his system generated semantic representations for 5.7 million passages from the Company archive.
Are the findings confirmed by historians?
The source account does not document independent confirmation of every result or provide enough detail to assess each record. Waites says he checked candidates against original scans and specialist catalogues; further expert review and examination of the cited documents are needed to determine which findings are genuinely new.
Source: hn
Columbus Day / Indigenous Peoples' Day Picks
long weekend sales
As an affiliate, we earn on qualifying purchases.
