Advertising
From Pink Slips to Silent Sidelining: Inside adland’s layoff and anxiety crisis
OpenAI’s latest model shows breakthroughs in solving longstanding math problems
Recent advances suggest large language models are starting to make meaningful progress on open mathematical questions.
Artificial intelligence systems are beginning to demonstrate measurable progress in solving open mathematical problems, marking a potential shift in how advanced research challenges are approached, as per a report by TechCrunch.
The development came to light after a software engineer and startup founder tested the mathematical reasoning abilities of OpenAI’s latest model on an unresolved problem. After an extended reasoning period, the model produced a solution that was later formalised and verified using mathematical proof-checking tools. The outcome suggested that newer large language models may be advancing beyond previously observed limitations in open-ended mathematical reasoning.
The model was found to rely on established mathematical principles and existing literature before arriving at a proof that differed meaningfully from earlier known solutions. The result addressed a variation of a problem originating from the extensive collection of conjectures compiled by Hungarian mathematician Paul Erdős, a set that has increasingly been used to assess AI-driven mathematical progress.
Recent updates to the Erdős problem archive indicate that several problems have shifted from “open” to “solved” status in recent weeks, with many solutions crediting AI systems as having played a role. Earlier progress had been attributed to models developed using Google’s Gemini technology, while newer experiments suggest OpenAI’s latest models show improved performance in high-level mathematical tasks.
Independent assessments within the mathematics community have pointed to cases where AI systems made autonomous progress, as well as others where models assisted by identifying relevant prior research. While AI has not reached the point of independently conducting mathematical research end to end, its scalability appears to make it particularly effective for tackling less prominent problems that may have simpler underlying solutions.
Parallel to problem-solving advances, AI tools are also accelerating the formalisation of mathematical proofs, a process that improves verification and reuse. Open-source proof assistants such as Lean, along with newer AI-driven formalisation tools, are increasingly being adopted in academic workflows to reduce the manual effort involved.
The growing use of these systems by professional mathematicians and computer scientists is seen as a significant indicator of their practical value. Rather than replacing human expertise, AI tools are increasingly positioned as collaborators that assist with exploration, verification and scale.
SPOTLIGHT
From purpose-driven work and narrative-rich brand films to AI-enabled ideas and creator-led collaborations, the awards reflect the full spectrum of modern creativity.
Read More“Confusion creates opportunity for agile players,” Sir Martin Sorrell on industry consolidation
Looking ahead to the close of 2025 and into 2026, Sorrell sees technology platforms as the clear winners. He described them as “nation states in their own right”, with market capitalisations that exceed the GDPs of many countries.