Przemysław “Psyho” DęBiak, a 42-year-old programmer from Gdynia, Poland, made history at the Atcoder World Tour Finals (AWTF) 2025 “Humans vs AI” contest held in Tokyo, defeating Openai's custom models. Considered one of the world's most prestigious coding tournaments, AWTF invites 12 of the top-ranked human programmers (for the first time, AI competitors) to tackle that tough challenge. After a 10-hour coding marathon, dębiak closed out about 9.5% AI, taking first place, with the model created by Openai settled in second place.
“Humanity has won (for now)!” dębiak wrote to X, confessing that he had only been sleeping for about ten hours in three days, pushing himself to the limit. Openai CEO Sam Altman responded appropriately, “Good job, Psyho.” Openiahc's branded AI model was widely expected to dominate the market. Still, instead of DęBiak's innovative, heuristic-driven approach – brute force calculations, he set the victory using problem-solving shortcuts and educated guesses. Contest administrator Yoichi praised his unique method, pointing out that AI “has not reached human creativity” while excelling in raw optimization.
Humanity has won (for now!) I am totally tired. I've had 10 hours of sleep in the last 3 days and I've barely lived. (To be clear, these are tentative results, but my leads should be big enough) pic.twitter.com/fimo0ifncdJuly 16, 2025
Atcoder World Tour Finals (AWTF) is considered the ultimate stage in a heuristic programming contest that focuses on “sufficient” solutions to complex and unsolved problems rather than perfect problems. This year's Challenge was tasked with competitors to plot the robot's paths on a 30×30 grid using as little movement as possible. With no access to external libraries or documents, success relies on intuition, creativity and adaptability. This may still represent human ingenuity and the speed and accuracy of the rawness of AI models.
DęBiak, a former Openai engineer who helped develop Openai Five (The Dota 2 AI), has admitted that he competed using only Visual Studio code in a basic autocomplete, pushing him to his limits. Near the end of the 10-hour marathon, he overtook Openaiahc, winning a victory and a 500,000 yen award. Debiak, a veteran algorithmic competitor, Mensa and four-time Top Coder Open Marathon champion, has never worked full-time and has even made a joke about considering a career from DJing to professional poker. His victory has symbolic weight. This is a human victory in an area where machines are usually superior.
Update: I'm alive, the results are now official, my lead over AI has increased from 5.5% to 9.5%. I didn't expect many people to be interested in programming contests. This means I need to drop here more often 👀pic.twitter.com/rsld8lecnqJuly 17, 2025
That being said, there is no denying the stable advancement of AI. At Stanford's 2025 AI index, the code benchmark jumped from 4.4% AI success in 2023 to 71.7% in 2024. However, this victory emphasizes that creativity, endurance and intuition reduce human strengths, especially in the long-term heuristic challenges.
dębiak is not naive about future challenges, saying, “It's easy to imagine another problem where AI wins and humans are far behind.” Still, this feels like a human John Henry moment. This means that we are not recreating the sparks of not only human will, but also the machinery of creativity, and (hopefully) cannot.
to follow Google News Tom Hardware Get the latest news, analysis and reviews in your feed. surely[フォロー]Please click the button.
