Anthropic researcher Tianyi Peng tested whether the Claude AI model could formalize Fermat's Last Theorem in the Lean proof assistant, and over 11 days of largely autonomous work, Claude generated an end-to-end, machine-checked proof. The system wrote roughly 13 million lines of Lean code and proved about 29,500 intermediate theorems, building on decades of prior work including Andrew Wiles's 1995 proof and a community formalization effort led by Kevin Buzzard since 2024.
anthropic.com
· 2026-09-04
Syracuse University and other schools are launching new undergraduate majors and minors dedicated to artificial intelligence, aiming to formalize AI education into standalone degree programs. Faculty are still working out exactly what curriculum these programs should cover, even as they promote them to prospective students.
fastcompany.com
· 2026-09-04
AI experts say the strange, unsettling appearance of AI-generated food photos stems from how diffusion models build images—starting with noise and forming rough shapes before adding fine texture. When the underlying structure is wrong, as with malformed hands, the model still layers convincing-looking detail on top, producing food that looks structurally off despite realistic textures. Researchers from Oxford and the University of Naples note these models also struggle with thin, continuous shapes, compounding the issue.
theverge.com
· 2026-09-04
Researchers at the University of Michigan reconstructed ocean surface temperatures over the past millennium using fossilized corals from the Galápagos Islands, concluding that the current El Niño event is the most intense on record for that timeframe. The findings suggest the phenomenon could drive more extreme weather across the U.S., particularly during hurricane season.
fastcompany.com
· 2026-09-03
Newly released emails obtained via a Freedom of Information Act request show DOGE affiliate Christopher Sweet, an undergraduate at the University of Chicago, worked with colleague Scott Langmack to seek academic credit for his work at HUD. The pair drafted a framework describing his time building an AI tool called SweetRex — designed to flag agency rules for rescission — as graduate-level research spanning AI, data science, and regulatory law, rather than a standard internship.
wired.com
· 2026-09-03
In 1932, Lebanese architect Nasri Khattar stumbled upon a way to simplify Arabic script while substituting for a typing instructor at the American University of Beirut. He discovered that using a single initial letterform instead of the correct medial one kept text legible, inspiring him to develop Unified Arabic, a system reducing Arabic's many positional and ligature variants down to 30 consistent letterforms.
worksthatwork.com
· 2026-09-02
Researchers on the LUX-ZEPLIN (LZ) experiment, led in part by Brown University scientists, detected a single particle interaction consistent with a WIMP dark matter candidate of at least 200 GeV/c2 mass. The signal reached only 2.6-sigma statistical significance, far below the 5-sigma threshold needed to claim discovery, but stood out because background noise in that region is unusually low.
science.slashdot.org
· 2026-09-02
Eric Harshbarger, a mathematician at Auburn University, has finally solved a puzzle he began working on in 2012: designing a set of dice that lets five players roll simultaneously to decide turn order, with every player having an equal chance of winning and no possibility of a tie. The finished product, called the Go First Dice, took nearly 15 years to move from theoretical concept to a working physical set.
cbc.ca
· 2026-09-02
Researchers publishing in Nature on September 2 found that older female mice given a GLP-1 obesity drug lived longer than untreated mice and showed slower onset of several biological markers linked to aging. The finding builds on earlier mouse studies suggesting GLP-1 medications, typically used for obesity and type 2 diabetes, may have broader anti-aging effects beyond weight loss.
nature.com
· 2026-09-02
Researchers analyzed genetic data from more than a million people and identified roughly 1,200 common genetic variants associated with the Big Five personality traits: extroversion, agreeableness, conscientiousness, neuroticism and openness. Published in Nature and led in part by Michel Nivard at the University of Bristol, the work is the largest genetic analysis of personality to date, though each variant explains only a small share of trait variation.
nature.com
· 2026-09-02
Researchers analyzed post-mortem retinal and brain tissue from multiple sclerosis patients and healthy controls, using single-nucleus RNA sequencing, immunohistochemistry and other molecular techniques. They identified an intracellular form of complement factor H within neurons that appears to protect these cells from damage during central nervous system inflammation, a mechanism distinct from the protein's known extracellular immune role.
nature.com
· 2026-09-02
Researchers pooled data from 46 cohorts through the ReGPC consortium to run genome-wide association analyses on personality traits among participants with European-like and African-like ancestry. They applied strict quality controls—filtering low-frequency variants, poor imputation quality, and problematic samples—before analyzing how genetic variants relate to standardized Big Five personality measures.
nature.com
· 2026-09-02
GoKawiil