Paleography, transcription, and translation are hellishly difficult to learn and really make it difficult for interested people to explore the archive. LLMs have become pretty darn good at doing this out of the box. Prior to this you needed specially trained ML models and before that there was no automation whatsoever.
I can't find any big AI model which can read historical Kurrent-family handwriting well out of the box. You still need specially trained models (i.e. transkribus) which generalize terribly.
Transkribus is definitely still the best option in some cases. But there are a bunch of cases where it was necessary two years ago and isn't necessary anymore, which is pretty remarkable.
