Did I miss where they talk about how the AI "coder" worked? because based on what I've heard from programmers it just would lie the first few times. Was a human allowed to fix mistakes that the AI made?

Seems like the model was specifically tuned for this maybe?

I feel like I'm missing information as to whether or not I should be impressed by the AIs performance.

fucking articles these days, no link to nothing, just bunch of copy paste text hype from twitter
here's link to problem: https://img.atcoder.jp/awtf2025heuristic/en.pdf
here's link to live stream: https://www.youtube.com/watch?v=TG3ChQH61vE

edit: from stream 39:00 first solution by openai was after 15 minutes first human solution was after 38 minutes and it was 2x slower than initial openai solution

edit2: from stream 7:20:12 winner is telling live that he slept 8-10h over last 72h, openai models are crap given he don't know what he's doing there, his code is crap and he doesn't know why it's working so well

Then he laid down his hammer keyboard and died.

A Polish programmer running on fumes recently accomplished what may soon become impossible: beating an advanced AI model from OpenAI

No 🤣

So long as the protocol is predictive or algorithm based, it won't ever be better at the job.

Oml. As someone that can't wait for AI to advance well enough so I don't have to fucking code anymore, I'm so sick of this bullshit narrative.

The reality headline is "Barely functional expert outperforms extremely fast junior hire still in probation period."

We already have AI that can beat this guy, but it takes so much work that it's still years and years and years from being able to cover all the things we consider everyday tasks. Let alone the complexities of coding, which require the fundamentals of "intelligence", independent planning and decision making.

It's like that other recent news when media was flabbergasted that basic coding in a 1970s Atari chess game bested an LLM. "Yeah, no shit. It's an LLM. Do you even know AI is an acronym, or is it just synonymous with magic at this point?"

Common sense is as valuable as a computer science degree on this level.

Edit: OpenAI's route to doing what they say they can do is using what they can currently do to assist the work into actually getting there... BECAUSE ITS ALWAYS BEEN A BIG FUCKING JOB. We don't have any massive breakthroughs, at best we've gotten shortcuts, but until we can physically overcome processing limitations, figure out quantum processing or similar, NO. Just, no. Ffs it takes climate destroying levels of computing power to just be at this level of absolute shit we have rn.

I am SO sick of this general public narrative because they've just been given food, that's always been there, on a plate and have nfi about how the food got onto the plate. If they did, they'd realise there's no magic. It's the same it's always been but someone decided to improve it a little for the consumer.

/rant

Having just read the problem, I'm curious how o3 solved it (and the human too tbh). My experience with LLMs says they'd be absolute complete crap at this, it's a very hard and open-ended problem. Intuitively I'd say it would just end up doing random changes tryjng to improve its score.

I think I could write the "trivial" solution but anything beyond seems... difficult. Congrats to the winner!

The last human code master.

I don't read this as a win. One man finished in front of OpenAI and many, many, many finished behind OpenAI. If this is the future of coding, it's bleak indeed.

The top 1% of developers will probably be OK no matter what, it's the rest of the crowd who isn't an award winning developer that are probably in trouble.

Getting Kasparov v. Deep Blue vibes here.

I have yet to have an AI write code of more than one or two lines that doesn't have a breaking bug. Speed isn't useful if it's broken. And honestly I usually spend more time debugging AI code than I would have just writing it myself. It's nice sometimes for getting an understanding of syntax of a system I'm not used to, but beyond very generic scripts that don't depend on context, it's pretty useless in my experience. I have Copilot integrates with my IDE for work and it's more trouble than it's worth so far. Even just for code completion, the IDE does a better job most of the time even if it suggests much smaller chunks at a time. And the smaller chunks are actually better if I have to proofread every single word either of then outputs anyway.

"Humanity has prevailed (for now!)," wrote Dębiak on X, noting he had little sleep while competing in several competitions across three days. "I'm completely exhausted. ... I'm barely alive."

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, Dębiak'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.

So ...

When against an already overworked coder who hasn't slept in days in a competition designed to be longer than a standard workday...

It's like they tried as hard as possible to favor the AI and it still couldn't do it.

It's John Henry all over again. Dębiak should make sure he gets some proper R&R now, just to be safe.

I can't help but think of John Henry 😂

It looks like the challenge was a variation on multi-agent path finding. Neat.

https://en.m.wikipedia.org/wiki/Multi-agent_pathfinding

https://atcoder.jp/contests/awtf2025heuristic/tasks/awtf2025heuristic_a

The challenge: https://atcoder.jp/contests/awtf2025heuristic/tasks/awtf2025heuristic_a

It's similar to the kind of thing you do in Advent of Code, but instead of a single answer, you're competing for the most efficient solution.

Now, imagine what a well-rested man could do!

Clearly didn't involve anything related to hardware architecture, where AI consistently gleefully hallucinates and shits the bed .