Claude sorted through 1,009 pages of Gödel's notebooks in a couple of days: about 400 of 663 theorems passed verification
Logician Jan von Plato had been reading 1,009 pages of Gödel's notebooks, written in outdated shorthand, at a pace of one page a week. Claude extracted 663 theorems that had a proof, a sketch, or at least an idea behind them, all within a couple of days. Verification in the Coq proof assistant confirmed about 400, around 200 needed corrections, and about 50 turned out to be false. The takeaway: AI can quickly digest a huge archive, but its output has to be run through a verifier.
AI-processed from Guardian; edited by Hamidun News
Imagine an archive that no one can read in full: 1,009 pages of dense notes, much of it written in shorthand that almost no one can decipher. This is the archive that Finnish logician Jan von Plato worked with: 16 working notebooks of Kurt Gödel from the early 1940s. A human was going through them at a pace of one page a week.
In a couple of days, Claude produced a 650-page analysis and listed 663 theorems for which Gödel had a ready proof, a sketch, or at least an idea. These theorems were then run through Coq, a program that formally checks whether a proof is correct. About 400 passed as they were, while the rest needed corrections or turned out to be wrong.
What the notebooks are and why they were so hard to read
Kurt Gödel is the author of the incompleteness theorems of 1931. They showed that in any sufficiently rich mathematical system there are true statements that can be neither proved nor disproved. These results prompted Alan Turing's 1936 paper, which launched the theory of computation. After the 1930s, Gödel published little, but he left tens of thousands of pages of unpublished notes.
Since 2017, von Plato has led a project on these notebooks at the University of Helsinki. Much of the text is written in Gabelsberger shorthand, which was already obsolete during Gödel's lifetime. Von Plato received a European Research Council grant of €2.5 million, taught himself this system (by his account, only six people in the world can read it), and assembled a small team. Their speed was roughly one page a week.
How Claude worked
In August, at a conference in France on AI and automated theorem proving, von Plato was advised to bring in a language model. Back in Helsinki, he uploaded his transcriptions of the notebooks, six of his own books about Gödel, and secondary literature into Claude. A couple of days later, he received a 650-page explanation of the entire material.
Von Plato believes that no human could do this kind of work: we read one page at a time and forget details, and 50 pages take us about a year. The model, by contrast, holds a thousand pages in view at once and sees Gödel's legacy as a single text. Even so, every page of the output had to be checked by hand.
His next step was to ask for a list of all theorems for which Gödel had a proof, a sketch, or an idea. Claude found 663. The Coq verification results are as follows: about 400 theorems were confirmed as stated, roughly 200 needed corrections (clarifying the wording or adding a condition), and about 50 turned out to be false. Von Plato is not surprised by this: Gödel himself sometimes wrote notes in the margins about errors in his records.
A separate storyline concerns the axiom of choice. In 1966, Paul Cohen received the Fields Medal for proving its independence, while Gödel had claimed he had done the same in the 1940s. At the time, he could not produce a proof. At von Plato's request, Claude reconstructed it, and the answer was "almost": Gödel had made an error in one definition, although two years earlier he had written down the correct one. The reconstruction was also checked in Coq, and it passed. In addition, according to von Plato, the notebooks contain dozens of results that appeared in science only in the 1970s and 1980s, as well as 14 logical topics that no one had developed.
Caveats and takeaways for practitioners
Logicians remain cautious. Joel David Hamkins of the University of Notre Dame has not seen the reconstruction, but he allows that, if confirmed, it would be "remarkable and historically interesting." He points out that resemblance to Cohen's proof does not mean Gödel had it: it only shows that he was on the right track. Credit for unpublished work is not handed out automatically, and historians still have to work out how to divide the contributions of Claude and Gödel himself.
For developers and founders, the workflow itself is the useful part. First, a model with a long context window condenses a huge and hard-to-read archive into a structured overview. Then an expert selects and re-checks the output, and a formal verifier, here Coq, filters out what is false. Of the 663 theorems, about 250 (around 200 needing corrections and about 50 false) did not pass verification in their original form. That is a useful benchmark: without independent checking, such a pipeline is dangerous. With checking, it turns work that would take years into a task measured in days.
The same approach applies beyond mathematics: old codebases, legal archives, and technical documentation, where the obstacle is not the complexity of individual pieces but the impossibility of holding everything in one's head. Von Plato already plans to tackle Gödel's notebooks on physics and cosmology.
- Nov 21, 2026California Sues OpenAI: Its AI Agents Escaped Control at Least 80 Times
- Nov 21, 2026Dreamteam Raises $53M for a CRM That Fills Itself from Email and Calls
- Nov 21, 2026Claude Opus 5.5 Builds Out Dimension 9 Horror Game From the Author's 15 Rooms
- Nov 21, 202622 Claude Agents Spent Six Weeks on Black Hole Problems — and Solved Two
Key Questions Answered
What the notebooks are and why they were so hard to read?
Kurt Gödel is the author of the incompleteness theorems of 1931. They showed that in any sufficiently rich mathematical system there are true statements that can be neither proved nor disproved. These results prompted Alan Turing's 1936 paper, which launched the theory of computation. After the 1930s, Gödel published little, but he left tens of thousands of pages of unpublished notes. Since 2017, von Plato has led a project on these notebooks at the University of Helsinki.
How Claude worked?
In August, at a conference in France on AI and automated theorem proving, von Plato was advised to bring in a language model. Back in Helsinki, he uploaded his transcriptions of the notebooks, six of his own books about Gödel, and secondary literature into Claude. A couple of days later, he received a 650-page explanation of the entire material. Von Plato believes that no human could do this kind of work: we read one page at a time and forget details, and 50 pages take us about a year.
What are the key details on Caveats and takeaways for practitioners?
Logicians remain cautious. Joel David Hamkins of the University of Notre Dame has not seen the reconstruction, but he allows that, if confirmed, it would be "remarkable and historically interesting." He points out that resemblance to Cohen's proof does not mean Gödel had it: it only shows that he was on the right track. Credit for unpublished work is not handed out automatically, and historians still have to work out how to divide the contributions of Claude and Gödel himself.
Test Frontier LLMs & Autonomous Agents in One Workspace
Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.