Post

Right, another key thing that historians do is working with important records that have not been digitized. And the vast majority of the world’s records *have not been digitized* and thus do not exist in any format that LLMs/AI can work with.

hmm how are you using the word "records" here

I (naively I guess) assumed that Google at least has warehouses with machines pulling stuff off of old floppies and CDROMs just like their book digitization infrastructure

thanks that was a really interesting and thought-provoking piece...I think before reading it I would have *tentatively* disagreed with your claim about how much has been digitized, but on a moment's thought that's obviously a reflection of my Western/Northern/otherwise-default mode of existence

Even in the geo area you’re talking about 95% of records/archives are NOT digitised.

That's so wild to me given what I've read about Google Books and other tech companies' pushes to digitize for LLM pretraining data