Tony! I was thinking about you the other day as I found some old photos from Dev8D online.

OCR is getting better all the time,* but still struggles with some of the collections people really want reliable access to.

And yes, the 'it's sometimes unclear how much time these tools really save' thing is very real.

* e.g. https://huggingface.co/blog/ocr-open-models and https://www.bl.uk/stories/blogs/posts/blplaybills-org-better-search-results-using-llms

If gen-OCR is creating training corpuses, then these will increasingly bake in biasses that further amplify the elision of minority positions in the training texts.

Avoid, unless you read every single word to check the transcription.