GPT-5.4 solves puzzle designed over 20 years to outsmart computers
What's the story
An artificial intelligence (AI) system based on OpenAI's new GPT-5.4 model has solved a complex mathematics problem that took nearly two decades to design. The challenge was created by Bartosz Naskrecki, a mathematician at Adam Mickiewicz University in Poland. He had long believed that AI could not handle such problems and described it as a "very advanced calculator" without the deeper understanding required for genuine mathematical insight.
Problem complexity
Challenge required advanced knowledge and reasoning
The mathematical challenge wasn't a simple puzzle. Its solution required some 13 pages of detailed reasoning, drawing on advanced fields like number theory, combinatorics, and algebraic geometry. Even experienced mathematicians would take weeks to find a viable approach. However, recent tests with GPT-5.4—a new generation AI model designed for advanced reasoning tasks—changed this assumption by solving the problem on its 11th attempt.
Unexpected outcome
Naskrecki was impressed by the AI model's output
The modest success rate of the AI model still surprised Naskrecki, who had spent years refining the challenge. After reviewing the output, he took to social media to express his "deeply impressed" reaction and called the solution "very nice, clean, and almost human." Despite this breakthrough, experts caution against overstating the capabilities of AI.