The Brief
Science & Discovery 5 min read

663 Theorems Hidden in a Dead Shorthand

NAVION

Share

Kurt Gödel, the Austrian logician whose incompleteness theorems reshaped mathematics in 1931 and laid conceptual groundwork for the digital age, left behind 1,009 pages of densely written notebooks that almost no one could read. Written in Gabelsberger, a German shorthand system already obsolete during Gödel’s lifetime, the notebooks sat largely inaccessible for decades. Now, with the help of the AI model Claude, a Finnish logician has extracted what appears to be one of the most significant caches of undiscovered mathematical work in modern history: 663 theorems, hundreds of them previously unknown, and at least one proof that may rewrite the historical record.

The Problem With Genius That Doesn’t Publish

Gödel published relatively little after his landmark results of the 1930s. His unpublished work, however, runs to tens of thousands of pages. Since 2017, Jan von Plato, a logician at Helsinki University, has led a project focused specifically on 16 notebooks Gödel wrote in the early 1940s. Von Plato taught himself Gabelsberger shorthand, becoming one of only six people in the world who can read it, assembled a small team, and received a €2.5 million grant from the European Research Council to support the effort.

The pace was brutal. Progress ran at roughly one page per week. At that rate, a thousand pages would take nearly two decades to process meaningfully.

The bottleneck was not comprehension of individual pages. It was synthesis. Mathematical notebooks are not linear arguments. They are webs of ideas, definitions, partial proofs, and marginal corrections scattered across hundreds of pages. A human reader processes one page at a time and, as von Plato put it plainly, tends to forget things. Connecting a definition on page 12 to a proof sketch on page 847 is not a task human working memory handles well.

What Claude Found in Two Days

In August, at a conference in France on AI and theorem proving, von Plato met computer scientists who suggested he use Claude to help make sense of the workbooks. He fed the model his transcriptions of the notebooks alongside six books he had previously written on Gödel and relevant secondary literature. Within a couple of days, Claude produced a 650-page explanation of the entire material.

Von Plato then asked Claude to extract every theorem in the workbooks for which Gödel had either a complete proof, a sketched proof, or a clear idea for one. Claude identified 663. Those results were then run through Coq, a formal proof assistant that verifies the logical validity of mathematical arguments. Approximately 400 theorems were verified as stated. Around 200 required repair, meaning a statement needed to be made more precise or an assumption added. About 50 were false, which is not surprising: Gödel himself occasionally annotated his own notes with the German words for “incorrect” or “bad.”

The most historically charged finding concerned the independence of the axiom of choice, a foundational question in set theory. In 1966, the American mathematician Paul Cohen won the Fields Medal, the highest honor in mathematics, for proving the independence of two well-known mathematical statements: the continuum hypothesis and the axiom of choice. At the time, Gödel claimed he had already proved the result in the 1940s, but when pressed by peers, he could not reconstruct the proof from his notes. Claude’s analysis suggests Gödel was essentially correct. He was, in von Plato’s words, “just one step away.” One definition was wrong, but Gödel had used the correct version two years earlier in the same notebooks. Von Plato ran Claude’s reconstruction through Coq, and it passed verification. Had Gödel published this work, von Plato believes he would have won the 1940 Fields Medal.

Joel David Hamkins, a professor of logic at the University of Notre Dame, has not seen the reconstruction but noted that if it holds up, it is “remarkable and historically interesting,” while raising a legitimate question: how much of the result belongs to Claude, and how much to Gödel. That question, he suggested, is a job for historians.

What This Reveals About AI as a Research Tool

Here is what most coverage of this story misses. The significance is not that AI read an old shorthand. It is that AI demonstrated a specific cognitive capability that humans structurally lack: holding an entire large document in working memory simultaneously and identifying connections across its full length.

Von Plato described it directly. Humans read one page at a time. It takes a year to work through roughly 50 pages. The AI could see all of Gödel’s writings as a single object and connect pieces separated by hundreds of pages. That is not a speed improvement. It is a qualitatively different mode of analysis.

This points toward a broader pattern in how AI is beginning to augment research. The bottleneck in many fields is not the absence of data or ideas. It is the human inability to synthesize large volumes of complex material without losing track of earlier details. AI does not replace the scholar who spent years learning Gabelsberger shorthand and building the contextual knowledge needed to frame the right questions. It handles the volume that scholar cannot.

The notebooks also contain, according to Claude’s analysis, more than a dozen themes in logic and foundations of mathematics that have not been pursued in the more than 80 years since Gödel wrote them down. Von Plato plans to continue, turning next to Gödel’s physics and cosmology notebooks.

In Short

A logician who spent years learning an obsolete shorthand used AI to synthesize 1,009 pages of Gödel’s unpublished notebooks in days, uncovering 663 theorems and a near-complete proof that may rewrite mathematical history. The lesson is not about AI replacing expertise. It is about what becomes possible when human expertise and AI’s capacity for large-scale synthesis work together on problems that neither could solve alone.

Based on reporting from The Guardian.

Written by

NAVION