ChatGPT-5.2 cracks unsolved math problem, stunning researchers with original proof
ChatGPT-5.2 cracks unsolved math problem, stunning researchers with original proof
ChatGPT-5.2 cracks unsolved math problem, stunning researchers with original proof
With Minimal Human Input
Researchers from the Free University of Brussels have reported that the ChatGPT-5.2 model helped prove a geometric hypothesis that had previously lacked formal verification. The AI generated the framework of the proof, while human experts reviewed the reasoning for accuracy.
This was reported by SciTechDaily.
The new study demonstrates that commercial language models can produce original mathematical proofs. The team at the Data Analytics Lab noted that ChatGPT-5.2 (Thinking) independently worked on a proof for a problem proposed by mathematicians Ran and Teng in 2024. A hypothesis is a statement considered plausible based on observations but not yet proven; once proven, it becomes a theorem.
According to the study, the final proof took shape after seven chat sessions with ChatGPT and four iterations of argumentation. The model helped identify approaches to the solution, while researchers verified the logic and completeness of the proof.
The authors emphasize that ChatGPT-5.2 developed a significant portion of the proof's structure with minimal human assistance. The lab considers this one of the first cases where a widely accessible language model autonomously generated an original mathematical proof. Postdoctoral researcher Brecht Verbeeck said he had long suspected ChatGPT could tackle unsolved problems but was surprised by how effective it proved to be.
The team describes their approach as "vibe-proving"—a method in which language models help structure complex theoretical ideas. They compare it to "vibe-coding," where AI assists in writing code. Professor Vincent Ginis noted that while some believe such systems merely rephrase training data, the results challenge that assumption.
At the same time, the researchers stress that human involvement remains essential for final verification and addressing gaps in the proof. Language models accelerate the formulation of potential proofs, but validation still requires time. Professor Andrés Alghaba concluded that this process could speed up further, with AI likely aiding even the verification stage.
Support the Charter-97* websiteSubscribe to our channel*
Leave a comment | Charter97.org
MULTI-CURRENCY DONATION ACCOUNT DETAILSBank name:Address:IBAN:SWIFT:Account holder:Payment purpose:Email: [email protected]
Follow Charter97.org on social media:
- YouTube
- X (Twitter)
- VKontakte
- Odnoklassniki
- RSS
- Telegram