Kurt Gödel left behind notebooks that almost nobody could read. According to reporting this week, researchers have now used AI to work through them and surfaced hundreds of theorems that appear to have been proved but never published.

The obstacle was never that the notebooks were hidden. They sit in an archive. The obstacle is that Gödel wrote them in Gabelsberger shorthand, a German stenography system that compresses words into strokes, mixed with mathematical notation and his own abbreviations. Reading a page takes a specialist with years of practice, and there are only a handful of people in the world with both the language and the mathematics to do it. A lifetime of notes, effectively unread.

The AI contribution here is closer to transcription than to discovery. The hard part of the pipeline is turning strokes into candidates for symbols, symbols into plausible formulas, and formulas into something a mathematician can check. That last step is where the human work remains, and it is also why the story is credible: a claim in a notebook is a proof sketch, not a refereed theorem, and nobody is asserting otherwise.

Why the story travelled. It became one of the most-discussed items on Hacker News this week, and the interest was only partly about Gödel. It is a template. Physics, mathematics and philosophy from the first half of the twentieth century are full of archives that were catalogued but never read, because the marginal cost of reading one more box was a specialist's month. A machine that can produce a readable draft changes the arithmetic on an entire genre of scholarship.

There is a second reason it resonated. Most AI stories this month have been about commerce, valuations and security incidents. This one is about a tool doing the thing universities say they exist to do: extracting knowledge from a document nobody else had the patience to sit with.

The honest caveats. Machine transcription of shorthand is probabilistic, and a single misread symbol can invert a formula. The output needs mathematical review before it enters the literature, and that review is slow. There is also an attribution problem: a theorem in a notebook from 1940 is not a new result, and priority questions around unpublished work are famously unpleasant. The value is in the catalogue, not the credit.

What to watch. Whether the transcription is published as a corpus that other mathematicians can search, whether any of the recovered results turn out to be useful rather than merely correct, and whether the approach generalizes to the archives of other scientists who wrote in systems nobody learns anymore. If it does, the interesting outcome will not be a headline about Gödel. It will be a generation of graduate students who spend their time on the mathematics instead of on the handwriting.