

I’m gonna buck the trend here, and pointedly go against my own distaste for LLM junk: this is precisely the sort of domain that LLMs are particularly good and efficient at. While I do believe the researcher is rather jumping the gun here and skipping more than a few crucial steps, I do think that his work here conclusively proves that there is a bunch of valuable logic and mathematical theory contained therein.
By no means am I encouraging anyone to take the ML-translated notes on faith, and there is of course substantial verification work to be done here without the use of an LLM. But using them to probe “is there knowledge of worth in here” across a thousand pages of an idiosyncratic dead shorthand language is honestly a very reasonable proposition. From there, using them to help build the initial draft of a more robust and deterministic translation layer, thus helping more people access the notes in many more languages, thus massively assisting in peer-review, will surely yield some neat and useful discoveries in this domain.
OP, you’re getting downvoted based on kneejerk reactions from the community due to the fact that the headline contains the string “AI” (a term which irritates me substantially, but has now unfortunately become part of popular vernacular), and a healthy dose of bandwagoning, imo. As someone who generally takes a rather dim view of LLM junk, I do not believe that this project categorically falls into the domain of “junk”.
I need to be clear here: it’s absolutely possible that you’re right. That’s why a huge part of my comment above is emphasizing that this dudes conclusions drawn from the notes as passed through Claude are largely premature, and need to be verified MUCH more rigorously.
But I do think that categorically disregarding this - particularly, without even understanding the nuance of his methodology or doing any sort of additional verification or isolated reproduction of results with different (non-LLM) tools - is also premature.