Level 1 — Absolute Beginner
Some numbers are special. You can divide them only by one and by themselves. We call them prime numbers. Two, three, five and seven are prime numbers.
In 1859, a man in Germany writes an idea about prime numbers. His name is Bernhard Riemann. He thinks prime numbers follow a pattern. But nobody can show that he is right.
Now, after 167 years, a mathematician says he has the answer. His name is Levent Alpoge. He works with a computer program called Claude.
Computers now check his work. Other mathematicians want to read it too. This takes a long time. If the answer is right, it is very big news.
- number
- A word or sign like one, two or three that tells how many.
- divide
- To separate a number into equal parts.
- prime number
- A number you can divide only by one and by itself.
- idea
- A thought about how something works.
- pattern
- A way that things repeat so you can see an order.
- mathematician
- A person whose work is the study of numbers.
- computer program
- A set of instructions that tells a computer what to do.
- check
- To look at something to see if it is correct.
Level 2 — Elementary
Prime numbers can be divided only by one and by themselves. They look scattered along the number line, with no obvious order. In 1859 the German mathematician Bernhard Riemann suggested that there is in fact a very precise pattern behind them. His suggestion is called the Riemann hypothesis, and for 167 years nobody has been able to prove it.
On Friday, Levent Alpoge, a mathematician who works at the AI company Anthropic, published a four page statement on his website. He wrote that he believed he had discovered a proof, working together with Claude, the company's AI model. He also said the result covers a wider version of the same problem.
Instead of writing a normal paper, Alpoge submitted his argument in Lean. Lean is a computer system that reads a proof step by step and refuses to accept anything that does not follow. Andrew Sutherland of MIT says the code has already passed several independent machine checks, and that more checks are still running.
That is not the same as being accepted. No human mathematician has read the whole argument yet, and Alpoge himself says it is very long and hard to read. James Maynard of Oxford University said that if the work fully checks out it is clearly a historic milestone, but that even checking the Lean verification will take time.
- scattered
- Spread out with no clear order.
- hypothesis
- An idea that is put forward and still needs to be tested.
- prove
- To show with certainty that something is true.
- statement
- Something written or said in a clear, official way.
- proof
- An argument that shows beyond doubt that something is true.
- step by step
- One stage at a time, in order.
- independent
- Not controlled or influenced by the person being checked.
- milestone
- An important event that marks a stage in progress.
Level 3 — Intermediate
Prime numbers are the atoms of arithmetic, and their arrangement along the number line has always looked maddeningly irregular. In 1859 Bernhard Riemann proposed that the irregularity is only apparent, and that the primes follow an extremely precise statistical pattern. Proving that proposition, known ever since as the Riemann hypothesis, became the most famous unsolved problem in mathematics. It is one of the seven Clay Millennium Prize Problems, each of which carries an award of one million dollars.
On Friday, Levent Alpoge, a mathematician employed at the AI company Anthropic, posted a four page note on his own website stating that he believed he had, working with Claude, discovered a proof. He claimed the argument establishes not only the Riemann hypothesis but its generalized form. He was unusually frank about the division of labour, writing that Claude had far outpaced him near the finish line.
The form of the submission matters as much as its content. Rather than a conventional paper for human referees, Alpoge released the argument in Lean, a proof assistant that will not accept a single step unless it follows from what came before. Andrew Sutherland of MIT reported that the submitted code has passed several independent computer verification checks, with further checks under way. That makes the machine side of the review unusually strong and the human side almost entirely absent.
The response from leading number theorists has been enthusiastic and carefully conditional. James Maynard of Oxford, who won a Fields Medal in 2022, said that if the result fully checks out it is clearly a historic milestone, adding that he suspects the proof will turn out to be correct while cautioning that even checking the Lean verification will take time. Youness Lamzouri of the University of Lorraine said that if the claim is true it will be a huge achievement, but that mathematicians will need to understand why the argument works, which could take months or even years. Nothing has been peer reviewed, and the mathematical community has not accepted the claim.
- arithmetic
- The part of mathematics that deals with numbers and calculation.
- irregular
- Not following a regular or predictable order.
- proposition
- A statement put forward for consideration or proof.
- division of labour
- The way a task is shared out between those doing it.
- proof assistant
- Software that checks each step of a mathematical argument.
- referee
- An expert who reviews a scientific paper before publication.
- conditional
- Depending on something else being true first.
- peer review
- Examination of a piece of research by other experts in the field.
Level 4 — Advanced
The primes are the indivisible building blocks of the integers, and their distribution has resisted description since antiquity. In 1859 Bernhard Riemann offered a conjecture of extraordinary reach: that the non trivial zeros of a certain complex function all lie on a single vertical line, a claim that, if true, would pin down the error term in the counting of primes with remarkable exactness. The Riemann hypothesis has since acquired a status closer to sacred object than open problem. It sits among the seven Clay Millennium Prize Problems, each carrying one million dollars, and it underwrites a large body of results that are currently stated conditionally, as true only if Riemann was right.
On Friday, Levent Alpoge, a mathematician at Anthropic, posted a four page document to his personal website announcing that he believed he had, in collaboration with Claude, found a proof, and that the argument extends to the generalized Riemann hypothesis as well. The candour of the announcement is itself notable. Alpoge wrote that the model had far outpaced him as the argument approached its conclusion, an admission that locates the intellectual contribution somewhere that the discipline has no settled vocabulary for.
The epistemics here are unusual, and they cut in two directions at once. Alpoge did not submit a readable paper to human referees; he submitted formal code in Lean, a proof assistant that refuses any inference not licensed by what precedes it. Andrew Sutherland of MIT reports that the code has passed several independent computer verification checks, with more in progress. Machine verification of this kind is in one sense a far stronger guarantee than a referee's reading, since it cannot be charmed by an elegant gap. But it certifies only that the formal statement follows from the formal axioms, and it offers no insight whatsoever, which is why Alpoge concedes the argument is extraordinarily long and difficult for a human to read.
The professional reaction has accordingly been warm and strictly hedged. James Maynard of Oxford, a Fields medallist, said that if the result fully checks out it is clearly a historic milestone, that he suspects the proof will prove correct, and that even verifying the Lean output will take time, with genuine understanding requiring a great deal more work. Youness Lamzouri of the University of Lorraine called it a potentially huge achievement while noting that broad acceptance may be months or years away, since mathematicians will want to know not merely that the theorem holds but why. For now the position is precise and should be stated precisely: a formal argument exists, machines have begun to endorse it, humans have not yet understood it, and nothing has been peer reviewed.
- conjecture
- A statement believed to be true but not yet proved.
- non trivial
- Not of the obvious or uninteresting kind.
- error term
- The measure of how far an approximation departs from the exact value.