AI unlocks Gödel's notebooks, said to contain hundreds of proved theorems
AI has been used to help decode Kurt Gödel’s handwritten notebooks, revealing hundreds of theorems and partially developed ideas from one of history’s most influential logicians. Commenters explore how much of the achievement comes from the language model versus decades of human expertise in Gödel’s shorthand and logic, with some seeing this as a powerful new way to synthesize and verify complex mathematical work and others downplaying it as glorified OCR and summarization. The exchange broadens into questions about how AI will reshape mathematical research, scholarly credit, and access to historically important but previously inaccessible archives.
AI’s Role and Capabilities
- AI was applied to existing transcriptions of Gödel’s shorthand, not raw scans, producing a ~650-page synthesized explanation of the workbooks.
- Commenters highlight the model’s ability to keep thousands of pages in context and connect distant parts, contrasting this with human limits (slow reading, forgetting).
- Some see this as a “huge deal” for understanding Gödel’s unpublished work and as a demonstration of LLMs’ value in long-context synthesis, not just transcription.
How Hard Was the Task?
- One camp claims the “unlocking” is overhyped and could have been done earlier with standard OCR / computer vision.
- The opposing camp points to slides of the manuscripts showing shorthand that is nearly unreadable without specialized training and deep knowledge of logic, arguing it’s far beyond trivial OCR.
- It is clarified that AI worked from already transcribed text; the shorthand decoding itself was done by humans.
Gödel’s Mathematics and New Results
- Strong excitement about potential “14 themes of logic” and new theorems, including hints Gödel may have approached independence results related to the continuum hypothesis and axiom of choice, though some remain skeptical.
- Several emphasize Gödel’s perfectionism and past delays between discovery and publication, suggesting it’s plausible his notes contain substantial but unfinished ideas.
- Parallel drawn to similar projects on Ramanujan and Grothendieck’s notebooks.
Expertise, Careers, and Obsolescence
- Debate over whether the shorthand expert made themselves obsolete:
- One view: once notebooks are translated, their unique skill loses value.
- Another: their value was in enabling and interpreting the AI-assisted work and leading the project, not gatekeeping the script.
- Some suggest this case shows how domain experts can become more, not less, impactful by using AI well.
Broader Reflections: Incompleteness, Priority, and Publishing
- Discussion on common misunderstandings of Gödel’s incompleteness:
- Clarification that it shows limits of any single formal system, not that all truths are unknowable.
- Others tie this to physical and computational limits, though some push back that this goes beyond what the theorems strictly state.
- Concern about priority in the age of AI:
- Will people hide work longer to avoid being scooped by AI?
- Or publish earlier, including scratch work, to establish precedence?
- Ideas floated: using hashes or zero-knowledge proofs to timestamp results without full disclosure.