Today in Science: AI's biggest math breakthrough yet

AI just solved the 80-year-old Erdős problem, and mathematicians are amazed ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­

May 21, 2026—AI just solved a major conjecture in geometry, and mathematicians are wowed. Plus, SpaceX's megarocket launches tonight, and Ebola is still spreading in Africa.  —Andrea Gawrylewski Chief Newsletter Editor P.S. If you're enjoying this newsletter, consider forwarding​it to a friend! We love adding new people to our science-loving community. Thanks! TODAY'S NEWS A poster displaying Ebola emergency contact numbers is pinned to a tent at the Busunga border crossing between Uganda and the Democratic Republic of Congo in Bundibugyo, on May 18, 2026. BADRU KATUMBA/Getty Images The Ebola outbreak in Africa will likely keep spreading and cause more deaths beyond the more than 130 estimated fatalities so far, the WHO says. But risks of a pandemic are still low. | 4 min read  Tonight, no earlier than 6:30 PM EDT, SpaceX will launch the largest version of its Starship megarocket yet, lifting off from the company’s Starbase compound in Texas. (Livestream will be here.) | 2 min read Too much screen time for children has been linked to poor sleep, behavior issues, and less physical and social activity, a new U.S. surgeon general advisory warns. | 2 min read  It turns out that seemingly harmless daddy longlegs are actually bloodthirsty killers—of frogs. | 2 min read Colossal Biosciences—the same company that claimed to “de-extinct” the dire wolf—announced it has developed an “artificial egg” that it says is a step toward “resurrecting” extinct birds. | 4 min read  NOAA predicts a quieter Atlantic hurricane season this year—but El Niño could boost storms in the central and eastern Pacific. | 2 min read The next-gen weight-loss drug retatrutide targets three hormone receptors and can result in the loss of 30 percent of one's body weight, putting it on par with bariatric surgery, new trial results show. | 2 min read More News Tiny investment, enormous universe: subscribe to Scientific American for 90 days for just $1 and help keep this newsletter going. TOP STORIES Milestone AI Math After 80 years of struggle by human mathematicians with nothing to show, a chatbot has solved a major geometry conjecture, called the “unit distance” problem. The company OpenAI, maker of ChatGPT, announced the result yesterday, together with comments from a number of experts, who declared the artificial intelligence’s method “clever” and “elegant.” Chatbots have made some achievements in math in recent months, but this marks a true milestone that would warrant being published in a mathematics journal if humans had achieved it.   The conjecture: Draw nine dots on a sheet of paper. The goal is to get as many pairs of dots as possible to be an inch apart. For any number of dots, even billions or trillions, the problem asks: What’s the highest number of pairs you can get? In 1946 mathematician Paul Erdős used some sophisticated math to show that using precise spacing, you could get more pairs than by simply placing them in a grid (of 3 by 3 dots, for example). For 80 years, no one could disprove him.  What the AI did: OpenAI mathematicians fed the conjecture to an internal large language model (LLM), which produced, in hundreds of pages of calculations and reasonings, something very different from a square grid solution. It devised a higher-dimensional lattice of points with special mathematical symmetries that facilitate the separation of even more pairs by the same distance. The AI model then developed a way to map this otherworldly grid back down to the two-dimensional page, producing a flattened numerical “shadow.” In fact, the solution can’t be drawn on paper. What the experts say: Independent mathematicians evaluated the LLM’s proof. What stood out to them was the chatbot’s patience and focus. “AIs have an edge: It’s not just that they can try all known methods,” says Jacob Tsimerman, a mathematician at the University of Toronto, who was not involved in the work but was part of a companion paper solicited by OpenAI. “They can play for longer and in more treacherous waters than mathematicians without getting overwhelmed.” The Intelligence-Depression Link Psychology researchers typically find that as intelligence (measured by IQ) increases, a person’s mental health improves—until it hits an inflection point and starts to go down. However, a recent study cast some doubt on that relationship. It found that depression scales that measure a person’s mental health may not work the same for people with high IQs.  How it works: The researchers crunched data from two U.S. surveys that included an estimate of a person’s IQ and their responses to well-established mental health questionnaires. At first, the team found the classic bell curve pattern of mental health increasing with intelligence, then decreasing after it reaches a certain point. But then they used a statistical method to determine whether the mental health measures work the same for people at different intelligence levels, in part by calculating whether responses to individual questions reflect depression to the same extent for everybody. Both questionnaires failed this test, meaning they can’t be used to compare people with differing intelligence.   What the experts say: The conclusion that people with high intelligence are more depressed because they answer questionnaires that way is flawed because the questions likely aren’t working in the same way for every patient. “Very intelligent people may think about mental health differently and maybe experience symptoms differently,” says Nicole Beaulieu Perez, a psychiatric nurse at New York University not involved in the research. —Emma Gometz, Newsletter Editor

SCIENTISTS AT WORK Erna Blancquaert Viticulture researcher Erna Blancquaert (right, with graduate student Sindiswa Zandile) starts work at dawn in the vineyards of South Africa’s celebrated Stellenbosch wine region. This time of day, before plant metabolism starts and the heat of the day sets in, is best for measuring the water status of grapevines and sampling fruit when its flavor is at its peak. Blancquaert and her team study the impact of sustainable and regenerative viticulture practices, such as using seaweed as a fertilizer and allowing cattle to graze in vineyards. Nature Africa | 3 min read Content Courtesy of Nature Briefing​. ​ The OpenAI large language model that solved the Erdős conjecture didn't use any special math or tools that didn't already exist. And, as noted in the article, without humans intervening to “clean up” the AI’s work, the result wouldn’t be so convincing. But AI did push out of natural human dimensions, which is something much harder for us to do. Being honest with ourselves about human limitations might help us see where AI can complement our strengths instead of merely surpassing them.   Send any comments and questions to newsletters@sciam.com. I'll be back tomorrow.   —Andrea Gawrylewski, Chief Newsletter Editor

Subscribe to this and all of our newsletters here. Scientific American One New York Plaza, New York, NY, 10004 ​ Support our mission, subscribe to Scientific American here ​ Unsubscribe    Preferences     View in Browser