For decades, mathematicians have dreamed of a world where computers could do more than just crunch numbers—where they could actually think, reason, and help discover new truths. A major hurdle, however, has been the notorious “hallucination” problem of artificial intelligence: large language models (LLMs) are great at sounding confident, but they frequently make mathematical errors, making them unreliable for serious research.
Now, a groundbreaking study by researchers including Tsoukalas et al., published in Science, has shattered that barrier by combining the creative writing power of AI with the ruthless accuracy of a mathematical referee.
Their system—called AlphaProof Nexus—pioneers a new way of doing math by teaming up an AI with a specialized computer program called Lean is a “formal proof assistant,” a piece of software that acts as the ultimate skeptic. It doesn’t accept a mathematical proof unless every single logical step is airtight and verified by its strict compiler.
The endless loop of creativity and proof.
The magic of AlphaProof Nexus lies in its teamwork model:
1. The AI brainstorms: The large language model acts as the creative mathematician, generating ideas, strategies, and formal proofs in the Lean language.
2. The computer checks: The Lean compiler instantly tests the AI’s work. If there’s even a tiny flaw in the logic, it rejects it and points out why.








