Recent New York Times Podcast

On October 13, the New York Times held a podcast interview with Dr. Steven Strogatz, faculty member in the Math Department at Cornell University. Dr. Strogatz is Susan and Barton Winokur Distinguished Professor for the Public Understanding of Science and Mathematics and a Stephen H. Weiss Presidential Fellow.

The title and description of the podcast were certainly provocative:

When A.I. Comes for Your Purpose in Life

In the span of just a few months, A.I. systems have solved some of the biggest outstanding problems in math — adding mathematicians to the list of those terrified of losing their job to the technology.

Artificial intelligence has gone from struggling with basic arithmetic to helping solve some of mathematics’ most famous open problems — and the field is reckoning with what that means.

Steven Strogatz, a professor of mathematics at Cornell University, describes the excitement and grief of watching A.I. take on work that gives people a sense of purpose — and asks what could be lost if machines deliver answers without the human struggle, insight and understanding that make solving problems meaningful.

Dr. Strogatz does a credible job of describing how quickly A.I. has progressed from being a laughingstock for mathematics just a couple years ago to tackling some of the great unsolved problems in the field. He also does an excellent job of explaining this to a lay audience and comparing these recent advances in A.I. with what had been the historical norms of the thrill of mathematical discovery. (Disclaimer: I would have chosen different metaphors in a couple of places. I don’t consider myself a hero for doing mathematics, and I would have likened proving a theorem to sex.)

The podcast ends with a description with an uncertain look toward the future, including a town hall of faculty and students at Cornell discussing, among other things, “What is the meaning of life?” In context: If A.I. can do the work of theorem-proving… if not now, then soon… then what is the point of studying mathematics in the future? I don’t begrudge any mathematician struggling with such existential questions in the face of what generative A.I. might produce in the very near future.

Dr. Strogatz appeals to the book of Ecclesiastes and Solomon’s similar musings about the meaning of life. I believe that is a more than appropriate source of wisdom, and I also recommend The Problem of Life with God: Living with a Perfect God in an Imperfect World, a discussion of Ecclesiastes by my former pastor (now semi-retired) Tommy Nelson.

Mathematics of the Future

I recently came across this very thoughtful opinion piece — leavened with appropriate cartoons — titled Mathematics of the Future, by Pavel Etingof and Slava Gerovitch of the MIT Department of Mathematics. The abstract:

Rapid advances in artificial intelligence are transforming mathematical research, raising critical questions about the role, training, evaluation, and social value of human mathematicians. We argue that, rather than making mathematicians obsolete, AI will prompt a reorientation of the profession away from an “ownership culture” centered on priority and publication records and toward intellectual agency: deep understanding, judgment, vision, communication, and stewardship of a shared mathematical culture. Mathematicians can use AI to explore a rapidly expanding mathematical landscape, interpret and synthesize AI-generated knowledge, repair and formalize the existing literature, and develop new forms of exposition and publication. We propose that hiring and graduate education rely more heavily on sustained mathematical interaction, including talks, discussion, oral examination, mentoring, and collaborative work, rather than treating papers alone as decisive evidence of ability. We also outline a model for AI-assisted research labs that combines collective human thinking with careful use and checking of AI output. Acknowledging serious ethical, social, and epistemic concerns and respecting those who choose not to use AI, we argue that the central challenge is not to compete with machines in producing theorems, but to preserve and expand humanity’s capacity to develop, understand, and communicate mathematics.

AI and the Riemann Hypothesis

Just a couple of months ago, major progress was made on the Riemann Hypothesis, one of the great unsolved problems in mathematics (and one of the seven Clay Millenial Problems worth $1 million to the first person who solves them). What’s especially noteworthy is that this progress was made by AI. From the opening paragraphs of an article in Forbes:

Number theorists spent 37 years moving one number by less than a single percentage point. The number is the proven share of the Riemann zeta function’s zeros that sit on the critical line, the place where one of the most famous unsolved problems in mathematics says all of them belong. Decades of human refinement had carried it to 41.6%. On August 10, Anthropic published a result from an unreleased research version of Claude that moved it to 67.2% in roughly a day and a half.

The Riemann hypothesis has been open since 1859. It predicts the hidden structure behind how prime numbers are distributed, and it carries a million-dollar Clay Institute bounty. The hypothesis itself remains open. What Claude established is that at least two-thirds of the zeros behave the way it predicts, the largest single advance in the record’s history…

The tempting story is that an AI out-thought the mathematicians. The evidence points somewhere more specific. Number theory is organized into specializations deep enough that a career fits inside one of them, and the researchers behind the two lines Claude combined publish in the same journals without needing each other’s machinery. Claude read both lines at once and found that they fit. The advance was sitting in the published literature, waiting for a reader with no specialization.

The Anonymous Math Geek Who Quit Anthropic – and Became the Face of AI Safety

This week’s post is a kind of counterpoint to last week’s post about the job prospects of math majors; competition between AI and Wall Street have driven up wages for math majors who are seeking non-academic jobs.

The Wall Street Journal ran a profile of Jacob Coxon, a former International Mathematics Olympiad participant (representing the UK) and Cambridge math major who worked in commodity training and then for OpenAI and Anthropic before dramatically quitting over fears about the lack of guardrails on AI’s development.

The Million-Dollar Talent Wars for 20-Something Math Geniuses

That was the eye-catching headline of a recent Wall Street Journal article, with the subheading “Flush with cash, Wall Street and Silicon Valley are vying to hire the same tiny pool of genius graduates”.

From the article’s opening paragraphs:

There has never been a better time to be a math nerd.

New college graduates and Ph.D.s are now securing million-dollar pay deals from elite trading firms seeking to secure the best and brightest amid fierce competition from artificial-intelligence companies.

So-called quant firms, which use sophisticated mathematical models to come up with trades, have for years wooed top young talent with lucrative compensation that big banks struggle to match. Now, the AI boom is pushing those numbers even higher.

Just a few years ago, early-career packages pushing seven figures were anomalies, said Matt Stabile, founder of New York-based recruitment firm Stabile Search. “But a million dollars is something people don’t even bat an eye at anymore.”

“The delineation is pre-OpenAI and post-OpenAI,” Stabile added, “that’s when you saw competition really take off.”

The skills required to train large language models have always overlapped with quantitative finance, but the connection has deepened as trading firms have pivoted toward machine learning and AI to power their trades in recent years. Now, AI labs such as OpenAI and Anthropic are vying for the exact same tiny pool of genius math majors and Ph.D.s as Wall Street.

The competition poses a challenge for quant firms, which have sought to sweep up the next generation of math whizzes to retain the trading edge that has made them a force in finance.

Ken Ono feared AI. Now he trains it

Physics Today recently published an interesting and thought-provoking Q&A with eminent mathematician Ken Ono, who recently took a leave of absence from his tenure-track position at the University of Virgina to join Axiom Math, an company developing AI tools to verify mathematical proofs. From the article’s opening paragraphs:

“Am I doing a better service to science by writing 20 more papers that only people in my field will read or by participating in this movement?” Ken Ono asked himself last year. At the time, Ono was an endowed chair at the University of Virginia. In the following months, he’d take a leave of absence from the university, move to Silicon Valley, and become the founding mathematician at an AI startup, Axiom Math.

Axiom is one of several companies focused on developing AI tools to verify mathematical proofs. In the future, similar tools could be applied to check AI-generated computer code, a potentially lucrative service now that AI tools can write lines of code faster than any human can debug them. Startups Harmonic and Math Inc as well as Google’s DeepMind are targeting similar AI-powered verification tools.

Axiom has attracted investor attention, reaching a valuation of $1.6 billion earlier this year. (Ono says that he has shares in the company and that he receives a salary that is similar to his former compensation at Virginia.) His interests, he says, are in helping researchers use AI to think in more-creative ways and across more disciplines.

News flash: ChatGPT sometimes asserts false statements are true (Part 2)

I feel like I’ve done my good deed for the day by uncovering another instance when ChatGPT claimed a “fact” from the secondary mathematics curriculum that is simply incorrect. This one’s a doozy: it claimed that the binomial coefficent \displaystyle {23 \choose 6} = 100,947 is equal to \displaystyle {22 \choose 5} = 26,334. (The first identity in the first line below is correct; the second one is not.)

Lesson: ChatGPT is a nice tool but you get what you pay for.

OpenAI Constructs New (and Correct) Mathematical Proof

In previous posts, I’ve commented about how AI can give startingly good and startingly bad results. This one definitely qualifies as startingly good. From the Wall Street Journal:

“If you are a mathematician,” one of the world’s leading mathematicians recently wrote, “you may want to make sure you are sitting down before reading further.” 

And you’ll definitely need to sit down if you’re not a mathematician. 

Because a famous math problem that stumped humans for the better part of a century has finally been toppled—by AI. 

The Wall Street Journal article above does a very good job of summarizing this amazing development. For more information, please see OpenAI’s press release and references therein.

News flash: ChatGPT sometimes asserts false statements are true (Part 1)

I feel like I’ve done my good deed for the day by uncovering an instance when ChatGPT claimed a “fact” from the secondary mathematics curriculum that is simply incorrect.

While using ChatGPT to do some brainstorming on a research project, I received the following response (as part of a bigger response):

Sadly, whenever ChatGPT asserts a fact without documentation, it behooves the user to double-check the “fact.” In this case, there is a pretty blatant sign error.

As taught in high school AP calculus, the Taylor series expansion for \ln(1+y) is

\ln(1+y) = \displaystyle \sum_{q=1}^\infty \frac{(-1)^{q-1}}{q} y^q,

and so

-ln(1-y) = \displaystyle -\sum_{q=1}^\infty \frac{(-1)^{q-1}}{q} (-1)^q y^q

= \displaystyle\sum_{q=1}^\infty \frac{(-1)^{2q}}{q} y^q

= \displaystyle\sum_{q=1}^\infty \frac{y^q}{q}

Said another way,

\displaystyle \sum_{q=1}^{A-1} \frac{y^q}{q} + \sum_{q=A}^\infty \frac{y^q}{q} = -\ln(1-y)

\displaystyle \sum_{q=A}^\infty \frac{y^q}{q} = -\ln(1-y) - \sum_{q=1}^{A-1} \frac{y^q}{q}

\displaystyle \sum_{r=0}^\infty \frac{y^{r+A}}{r+A} = -\ln(1-y) - \sum_{q=1}^{A-1} \frac{y^q}{q}

\displaystyle \sum_{r=0}^\infty \frac{y^{r}}{r+A} = \frac{1}{y^A} \left[ -\ln(1-y) - \sum_{q=1}^{A-1} \frac{y^q}{q} \right]

So that’s the correct answer. Notice that there is a minus sign in front of the sum on the right-hand side, not a plus sign.

So I asked ChatGPT to double-check. Here’s the first part of the response.

Lesson: You get what you pay for.

AI and Proving Theorems

Paul Erdős famously said that mathematicians are machines that turn coffee into theorems. A couple of recent articles in the Wall Street Journal revealed the current state-of-the-art for AI to do the same.

In July, Ben Cohen published the article The High-Schoolers Who Just Beat the World’s Smartest AI Models. The focus of the article was the 2025 International Mathematical Olympiad, the pinnacle of the calendar for high school mathematics competitions. In the United States, the pathway to the IMO is first excellng at a sequence of increasing difficult exams: the AMC->12 (or possibly AMC->10), then the American Invitational Mathematics Exam (AIME), and then USA Mathematical Olympiad (USAMO) or USA Junior Mathematical Olympiad (USJAMO). The top USAMO and USJAMO participants then get invited to a special camp from which the participants in that year’s IMO are selected.

My personal story: back in high school, my score on the AMC->12 (then called the AHSME) qualified me for the AIME my sophomore and junior years, where my run in the competition ended with a resounding thud. My senior year, I caught lightning in a bottle and somehow qualified for the USAMO; I’m not sure what the cut-off is these days, but back then only 150 or so high school students qualified for the USAMO each year. My excitement at qualifying for the USAMO gave way to utter humiliation after I actually attempted the exam (to say that I “took” the exam is probably a misuse of the work “took”.) All this to say: I never came close to sniffing the IMO. From the Wall Street Journal article:

The famously grueling IMO exam is held over two days and gives students three increasingly difficult problems a day and more than four hours to solve them. The questions span algebra, geometry, number theory and combinatorics—and you can forget about answering them if you’re not a math whiz. You’ll give your brain a workout just trying to understand them. 

Because those problems are both complex and unconventional, the annual math test has become a useful benchmark for measuring AI progress from one year to the next. In this age of rapid development, the leading research labs dreamed of a day their systems would be powerful enough to meet the standard for an IMO gold medal, which became the AI equivalent of a four-minute mile. 

But nobody knew when they would reach that milestone or if they ever would—until now. 

The unthinkable occurred earlier this month when an AI model from Google DeepMind earned a gold-medal score at IMO by perfectly solving five of the six problems. In another dramatic twist, OpenAI also claimed gold despite not participating in the official event. The companies described their feats as giant leaps toward the future—even if they’re not quite there yet. 

In fact, the most remarkable part of this memorable event is that 26 students got higher scores on the IMO exam than the AI systems. 

A second article by Ben Cohen, The Math Legend Who Just Left Academic — for an AI Startup Run by a 24-Year-Old, might be a precursor of things to come. One of the two starts of the article is number theorist Dr. Ken Ono. From the article:

In recent years, Ono began tracking AI’s remarkable progress as it rapidly improved. He was intrigued, though not intimidated. AI was astonishing at cognitive tasks and solving problems it had already seen, but it struggled with the creative elements of his field, which require intuition and abstract thinking.

That creativity is so fundamental to pure mathematics that Ono figured his job would be safe for decades.

But last spring, he was one of 30 mathematicians invited to curate research-level problems as a test of the AI models. He left the symposium profoundly shaken by what he’d seen.

“The lead I had on the models was shrinking,” he said. “And in areas of mathematics that were not in my wheelhouse, I felt like the models were already blowing me away.”

For months afterward, Ono felt like he was grieving his identity. He didn’t know what to do next, knowing that AI models would only get smarter.

“Then I had an epiphany,” he said. “I realized what the models were offering was a different way of doing math.”

Dr. Ono is now taking an extended leave from the University of Virginia to join a new AI startup company called Axiom. From Tech Funding News:

Led by Carina Hong, Axiom Math is developing an AI system that not only solves complex math problems but also generates new mathematical knowledge by proposing conjectures: mathematical statements that have yet to be proven.

The model produces rigorous, step-by-step proofs that can be independently verified using proof assistants such as Lean and Coq. This approach aims to transform English-language math from textbooks and research papers into code, enabling the AI to create and validate new problems that push the boundaries of existing knowledge…

Currently, Axiom is working on models that can discover and solve new math problems. The researchers also hope to apply their work in areas like finance, aircraft design, chip design, and quantitative trading.

Beyond pure mathematics, Axiom’s AI tool is being tested for practical applications in fields requiring rigorous computational precision, including finance, aircraft and chip design, and quantitative trading.

Time will tell if the intersection of AI with mathematics can generate a profitable company. What I don’t doubt is that the previously unthinkable — original mathematical work by AI — will eventually happen, given enough time.