I wish AI leaders would fund more data archival projects, instead of hoarding it themselves. They could build a lot of goodwill, but the current state of the art is destroying old books and denying anyone access to them.
These archival projects have an enormous impact, though they are chronically underfunded. I wish there were a strong economic incentive to support them.
Anthropic just learned to the tune of a 1.5 billion dollar settlement that making available for human consumption, even internally, instead of hoarding for AI training exclusively is a fraught exercise.
Seems like a very interesting challenge for automation of raster to vector scanning/software, and QC/checking that the raster-to-vector pipeline hasn't hallucinated something or joined two shapes together that should be discrete objects.
I would presume they're starting with very high res lossless raster scans and storing those in some sort of index, then making them available for different pipelines of processing?
I was very excited to see this project. I went to the site and requested processing of the maps for my small town. It also inspired me to make a donation to OSM.
On visiting the actual oldinsurancemaps.net, I was met with:
"OldInsuranceMaps.net is a Charter Project of OpenStreetMap US that provides an open, online workflow for creating georeferenced layers of historical maps, with a special focus on making seamless mosaics from multi-page atlases in the Sanborn map collection at the Library of Congress."
Sanborn was an insurance adjacent company that made detailed paper maps of many towns in the US in the late 1800s and early 1900s.
This project is scanning those maps and adding the information that modern mapping software needs to show them alongside other geographic information. Where the old paper maps overlap, they work to cleanly stitch them together rather than showing the edges of the paper.
Later: The project is sort of for people looking for the old maps, so I can see why the introduction doesn't spend a life of time explaining them.
They are building computer programs that other people can use to then post online maps from old paper sources. For example, from the Sanborn map collection.
Those maps can be browsed as "overlays" on top of other maps, so you can easily compare how several maps show the same places on Earth.
I wish AI leaders would fund more data archival projects, instead of hoarding it themselves. They could build a lot of goodwill, but the current state of the art is destroying old books and denying anyone access to them.
These archival projects have an enormous impact, though they are chronically underfunded. I wish there were a strong economic incentive to support them.
Anthropic just learned to the tune of a 1.5 billion dollar settlement that making available for human consumption, even internally, instead of hoarding for AI training exclusively is a fraught exercise.
Can you elaborate? I checked the news and couldn't find what you're referencing myself.
That's on current copyrights, not public domain works like these really old maps.
SourceLibrary.org for instance… how to make what is done there an economic value for AI companies with data for training?
Seems like a very interesting challenge for automation of raster to vector scanning/software, and QC/checking that the raster-to-vector pipeline hasn't hallucinated something or joined two shapes together that should be discrete objects.
I would presume they're starting with very high res lossless raster scans and storing those in some sort of index, then making them available for different pipelines of processing?
I started this project and couldn't be happier about this partnership with OSM US!
Also here's a past HN thread with some more context: https://news.ycombinator.com/item?id=46788588
I was very excited to see this project. I went to the site and requested processing of the maps for my small town. It also inspired me to make a donation to OSM.
If you like old maps you can check the french maps with historical layers (17xx, 1880, 1950, today) :
https://cartes.gouv.fr/explorer-les-cartes/
you can add and remove layers by clicking on the map icon on the top right.
Having these maps under a charter means long-term maintenance isn't just one person's side project anymore.
Great resource. Fire insurance maps are goldmines for urban history research.
Would be great to non-destructively scan the lower layers and date code them at some point, if/when possible.
Solid, no complaints
On visiting the actual oldinsurancemaps.net, I was met with:
"OldInsuranceMaps.net is a Charter Project of OpenStreetMap US that provides an open, online workflow for creating georeferenced layers of historical maps, with a special focus on making seamless mosaics from multi-page atlases in the Sanborn map collection at the Library of Congress."
What the hell does that mean? Anyone to ELI5?
Sanborn was an insurance adjacent company that made detailed paper maps of many towns in the US in the late 1800s and early 1900s.
This project is scanning those maps and adding the information that modern mapping software needs to show them alongside other geographic information. Where the old paper maps overlap, they work to cleanly stitch them together rather than showing the edges of the paper.
Later: The project is sort of for people looking for the old maps, so I can see why the introduction doesn't spend a life of time explaining them.
ELI5:
They are building computer programs that other people can use to then post online maps from old paper sources. For example, from the Sanborn map collection.
Those maps can be browsed as "overlays" on top of other maps, so you can easily compare how several maps show the same places on Earth.
They make a nice way of stitching together old maps.
Been thinking about this for years
Fantastic project.