Exhausted Man Defeats AI Model In World Coding Championship
- Reference: 0178412832
- News link: https://developers.slashdot.org/story/25/07/18/2025211/exhausted-man-defeats-ai-model-in-world-coding-championship
- Source link:
> A Polish programmer running on fumes recently accomplished what may soon become impossible: [1]beating an advanced AI model from OpenAI in a head-to-head coding competition . The 10-hour marathon left him "completely exhausted." On Wednesday, programmer Przemysaw Debiak (known as "Psyho"), a former OpenAI employee, narrowly defeated the custom AI model in the [2]AtCoder World Tour Finals 2025 Heuristic contest in Tokyo. AtCoder , a Japanese platform that hosts competitive programming contests and maintains global rankings, held what may be the first contest where an AI model competed directly against top human programmers in a major onsite world championship. During the event, the maker of ChatGPT participated as a sponsor and entered an AI model in a [3]special exhibition match titled "Humans vs AI." Despite the tireless nature of silicon, the company walked away with second place.
>
> The competition required contestants to solve a single complex optimization problem over 600 minutes. The contest echoes the American folk tale of John Henry, the steel-driving man who raced against a steam-powered drilling machine in the 1870s. Like Henry's legendary battle against industrial automation, Debiak's victory represents a human expert pushing themselves to their physical limits to prove that human skill still matters in an age of advancing AI. Both stories feature exhausting endurance contests -- Henry drove steel spikes for hours until his heart gave out, while Debiak coded for 10 hours on minimal sleep. The parallel extends to the bittersweet nature of both victories: Henry won his race but died from the effort, symbolizing the inevitable march of automation, while Debiak's acknowledgment that humanity prevailed "for now" suggests he recognizes this may be a temporary triumph against increasingly capable machines. While Debiak won 500,000 yen and survived his ordeal better than the legendary steel driver, the AtCoder World Tour Finals pushes humans and AI models to their limits through complex optimization challenges that have no perfect solution -- only incrementally better ones.
"Humanity has prevailed (for now!)," [4]wrote Debiak on X, noting he had little sleep while competing in several competitions across three days. "I'm completely exhausted. ... I'm barely alive."
[1] https://arstechnica.com/ai/2025/07/exhausted-man-defeats-ai-model-in-world-coding-championship/
[2] https://atcoder.jp/contests/awtf2025heuristic
[3] https://atcoder.jp/posts/1516
[4] https://x.com/FakePsyho/status/1945444118924272018
It depends on the challenge (Score:2)
I am fighting with Claude 4 and ChatGPT 4.1 using them in VSCode with Copilot as an Agent (Gemini 2.5 pro fails every time), everything is fine, but they can't figure out a simple PHP application with everything explained in a nice file ,every url is wrong, uses differents patterns of design in every page, broke stuff that worked, iterates in useless testing code that never works and more. So, probably you don't need to test an AI in a precise exercise but in a real world application, an useful one, there's
Re: (Score:2)
From the summary: "The competition required contestants to solve a single complex optimization problem over 600 minutes."
And there you have it! It's a sufficiently-specific problem based on a knowledge base that is sufficiently-well-established in sources that can be used as training materials for AI. So, that's why it did so well.
Once you start putting the AI into real world situations, like the ones described in the parent post, it performs much worse. I literally just spent time fiddling with permissi
Re: (Score:2)
Thre are links on the AtCoder website for the challenge that take you to the exact problem to be solved. Its an interesting puzzle to read. At first, I was like 'there's no UI involved, of course that's why the AI did well', but then I read some more and realized that this was exactly the same kind of real world problem I worked on in the 90s to do with hard drive optimisation.
The problem they offered was not a common problem, but it was definitely mappable to some real world problems.
It demonstrates
Reminds me of old John Henry (Score:2)
John Henry told his captain
'A man ain't nothin' but a man
But before I let your steam drill beat me
Down
I'd die with a hammer in my hand. Lord
Lord
I'd dies with a hammer in my hand.'
Re: (Score:2)
You beat me to it. This guy is the modern John Henry!
Competitive coding is not... (Score:2)
...the same as solving real and complex problems, especially if they are not precisely defined
Re: (Score:2)
Yea, I'd even question the notion that 'top coders' go to these events. I've had the opportunity to work with a lot of people who pioneered modern open source developer tools over the past 20-years and I can't recall ever talking about coding competitions.
Humans always beat computers.. until we don't (Score:2)
If computing has taught us anything, it's that they tend to improve in capability
Re: (Score:2)
It's not really going to help if "AI" just makes mistakes faster than before. This is a technology that doesn't look like it improves significantly.
Re: (Score:2)
What is your basis for thinking LLMs don't improve? Genuinely, where does that come from. All the charts and benchmarks and records I see are going up every time there is a new one.
publicity stunt (Score:2)
Well done to openai for scraping all available code from the internet and amalgamating bits of it produce an apparently working program. We should also give credit to be many thousands of real people who manually curate their LLMs day after day. Let us not forget that there is minimal intelligence in current LLMs. Their only intelligence is that they can interpolate between datapoints to give outputs that are somewhere in between them. They have no ability to operate outside of their training data and ther
John Henry died (Score:2)
Just a reminder that the steam machine won in the end.
Re: (Score:2)
> [...] noting he had little sleep while competing in several competitions across three days.