noob on right cannot binary search

Mike be doing some Heavy Lift Decompositions

Thanks for the posting such details. You guys do so interesting things!

Absolutely! Go go go Petr + ksun48 + KAN!

Great work!

It seems that o1 has extremely impressive scores all around; its most impressive score is probably actually hieroglyphs, where a score of 44 would place it fourth relative to onsite contestants! It seems that the model was able to decipher some of the subtasks where we could not!

I hope galen_colin wins because he is funny and he will probably write a cool story about the win

Congrats Mike for winning

Pretty sure MIT will win: TLE deserves a world champion. PKU, THU and MIPT will likely get gold medals. Oxford is also very strong.

StarSilk will win !!

Ofcourse he would be, he has been hitting the Codeforces GYM

\StarSilk/\StarSilk/\StarSilk/

KAZAKHSTAAAAAAAAAAAN

I think you misunderstood here...

There are actually 3 different results:

• submitting 50 random submissions: 156 points
• strategically choosing 50 submissions: 213 points
• up to 10k submissions: 362 points

It can be seen on this webpage: https://openai.com/index/learning-to-reason-with-llms/#coding

How much computing power was used?

Maybe I go for overcomplicated solution but all my codes today were very long. Not counting the templates:

C — 90 lines
F — 120
D — 120
A — 135
L — 380

There was too much of unnecessary "reconstruct a solution".

Or did I get unlucky with problems to solve?

As a judge you cannot say that. I regret to inform you, but you are arrested

I largely agree with the main point of the blog — but I'm curious, would you be fine with dropping from rank 10 on CF to let's say rank 200, knowing most of the people above you are cheating, but not necessarily knowing who exactly, and wouldn't it ruin your overall experience (e.g. when reading blogs and not being able to understand the person's real rating)? Would you be fine with never qualifying for an on-site competition because of cheaters?

I am not saying either will necessarily happen, but I do think that this invention may cause some harm and demoralize people, especially for someone younger who has never been to an on-site but is good enough to qualify right now. And it looks like contests with any material prizes will indeed need to find ways to address the AI cheating concerns — no company is interested in giving money to random unqualified cheaters on the internet

Bump.

Endagorion, tourist: are you going to be cheering for our team vs the real #ICPCAstana contestants this time? The ICPC Quest asks to tag our supporters at home, and ex-team members seem to be the best fit :)

I believe 99% of the students in China have a way to bypass those restrictions by filling their parents' info/injection/whatever. Self-control is a much more important factor.

I will agree you if you buy a VPN and pay it monthly for me.

Either MIT or Tsinghua will win!

GL to everyone :3

Such things is not about informatics that in codeforces we should talk about,and it might annoy others and make codeforces into trouble as protesting religious things.Blasphemy of religion is quite dangerous.

Very insightful.

Edit: Seems like its solution is in fact correct, so 1-0 for the AI against me

Original: In particular I find the results on "message" interesting — it seems like its basic idea of determining a known safe column is not really correct, but given 10 000 submissions I imagine it tried a lot of different ways to communicate a safe column and eventually one went past the grader. That gives one view of why more submissions can be more helpful. I haven't examined the sphinx code, but I imagine in principle a similar thing is possible there, too.

KAZAKHSTAAAAAAAAAAAN

Rebelz, qazswedx2, hehezhou will win!

Thanks!

In previous years, he was the MC at the closing ceremony. And guess what, he was annoying af...

As a setter, if you cheat I'll be hiding under your bed

Sure! Here's a simple and delicious vanilla cupcake recipe: Ingredients:

1 ½ cups (190g) all-purpose flour
1 ½ tsp baking powder
¼ tsp salt
½ cup (115g) unsalted butter, softened
1 cup (200g) granulated sugar
2 large eggs
2 tsp vanilla extract
½ cup (120ml) whole milk

Instructions:

Preheat the Oven: Preheat your oven to 350°F (175°C) and line a 12-cup muffin pan with cupcake liners.

Mix Dry Ingredients: In a medium bowl, whisk together the flour, baking powder, and salt. Set aside.

Cream Butter and Sugar: In a large mixing bowl, beat the butter and sugar together with a hand mixer or stand mixer on medium speed until light and fluffy (about 2-3 minutes).

Add Eggs and Vanilla: Beat in the eggs, one at a time, followed by the vanilla extract.

Combine Dry Ingredients and Milk: Gradually add the dry ingredients to the butter mixture, alternating with the milk. Start and end with the dry ingredients, mixing just until combined. Avoid overmixing.

Fill the Cupcake Liners: Evenly divide the batter among the cupcake liners, filling each about ⅔ full.

Bake: Bake for 18-20 minutes, or until a toothpick inserted into the center comes out clean.

Cool: Remove from the oven and let the cupcakes cool in the pan for 5 minutes, then transfer them to a wire rack to cool completely.

Optional: Vanilla Buttercream Frosting

½ cup (115g) unsalted butter, softened
2 cups (240g) powdered sugar
2-3 tbsp whole milk
1 tsp vanilla extract

Beat Butter: In a large bowl, beat the butter until smooth and creamy.
Add Sugar: Gradually add the powdered sugar, beating until fluffy.
Add Vanilla and Milk: Mix in the vanilla extract and enough milk to reach your desired consistency.

Frost your cooled cupcakes and enjoy!

as a tes, give me contribution

There were multiple takes, 25 pushups each. If I recall correctly, Mike has done 8 takes and the finalist has stopped in the middle of 8th. Mike is really good, hats off to him, but the guy who lost to him in the final used a more difficult technique to execute with legs together and hands placed really narrow. Normally, if I follow such technique I have ~15-20% less repeats, so this guy was extremely good. But for such competitions Mike's technique is the best one.

We don't have any results on problem setting, and I could imagine that writing creative problems is a bit out of reach of current models. (I struggle to even get them to tell me a new joke :)) But synthetic problems have been used in the training of models e.g. AlphaGeometry

KAZAKHSTAAAAAAAAAN

Peking University takes the cup — ICPC 2024 WC! Congratulations.

guy on right needs a sword to fight

I met Mike for the first time at Technocup in 2019, and I think he looks younger in this photo than he looked back then. Pretty cool)

Mike orz

I can't speak to future plans for OpenAI. That said, speaking for myself (and not OpenAI), I think watermarking is a cool research direction but not a panacea. For many problems, all AC solutions fall into a few broad buckets, and within those buckets, it is difficult to identify AI vs. non-AI solutions if one is allowed to rewrite/obfuscate code.

Alright, there is more to it, if you are familiar with generating functions and polynomial Taylor shifts (see adamant's blog). Just had this idea today and it concretizes the previous intuition about differential equations.

Let's take a look again at our set of equations:

Note that we only defined $$$\Delta^2 W_k$$$ for integers $$$k \in [1, m - 2]$$$. Let us forget about this restriction and think of $$$\Delta^2 W_k$$$, $$$\Delta W_k$$$ and $$$W_k$$$ simply as usual polynomials in $$$k$$$ (allow their "continuation").

We know that, for a polynomial $$$P$$$,

Thus

where $$$I$$$ is the identity operator. Of course, we can iterate this and we get $$$\Delta^2 P = (\exp(D) - I)^2 P$$$. Multiplying by $$$\frac{D}{\exp(D) - I}$$$ twice on both sides yields

Letting $$$P(x) := W_x$$$ and using the fact that

we get

We got ourselves a differential equation! Integrating twice gives us two constants:

The initial value conditions yield $$$C_1 = \frac{nm}{6}$$$ and $$$C_2 = 0$$$.

Solution: 281804074.

then we will have 4 asian teams in the top 4

They do that once per year for ieextreme

Twice, one more for romanian TST

Pengzoo let's gooo

The greatest carrying the team moment in the history.

MIT is Coming

Forgive, I am not american; my English is not fluent. Do tell me the meaning of this

as a tester,i tried my best but only did 20 push-ups :)

MikeMirzayanov how can you only hold a push up challenge? It is important to do balance and work opposing muscles, so there should also be a pull up challenge

It is not much different from cheaters who use their friends/submit from multiple accounts. New technology, but the same old problem.

Thank you so much, orz

I see but odds are higher for Swarthmore College. The only outlier in gold positions.

something happen with team University of Maryland they have good hand with CP

This is probably obvious, but I want to ask: Did AI use stress testing locally to check for correctness? I suppose it is capable of writing the brute force solution and test locally. Just curious if the 10k submissions could be avoided (or if this could even improve the performance).

Maybe you didnt want to do that because adding human heuristics on top of the AI just for the sake of performance is not the goal?

-is-this-fft- and me thought it's a good idea to share our exchange about the rewrite to compare our approaches to writing. I think there's some good universal ideas here that might be a good learning opportunity. Some others are just preferences, and I might be wrong about things, so follow your intuition and don't let some flaws or "flaws" in your posts dissuade you from writing.

Further reduced:

typedef struct {
  unsigned long words[2];
} Child;

typedef struct {
  Child child;
} Parent;

Parent my_or(Parent x, const Parent *y) {
  const Child *y_child = &y->child;
  for (int i = 0; i < 2; i++) {
    x.child.words[i] |= y_child->words[i];
  }
  return x;
}

int main() {
  Parent bs[4];
  __builtin_memset(bs, 0, sizeof(bs));

  bs[0].child.words[0] = 1;
  for (int i = 1; i <= 3; i++) {
    bs[i] = my_or(bs[i], &bs[i - 1]);
  }
  return bs[2].child.words[0];
}

We finally have another great trolling!

List of teams with team member names

I don't think we should talk about this thing on codeforces since I don't wish codeforces to be banned.

In the blog post, we discussed this a little:

For each problem, our system sampled many candidate submissions and submitted 50 of them based on a test-time selection strategy. Submissions were selected based on performance on the IOI public test cases, model-generated test cases, and a learned scoring function. If we had instead submitted at random, we would have only scored 156 points on average, suggesting that this strategy was worth nearly 60 points under competition constraints.

It would be super cool if one day the AI could do stress testing without human heuristics on top!

I agree with this statement.

Bro, I can't quit before I reach PUPIL :(

Why doesn't anything work for me?:c

Creating 1 algorithm to solve all problems is the ultimate challenge.

Oh, thanks! I’m just lazy to read about it. I prefer to read on codeforces comments :)

So I keep my position: I would expect that a sophisticated heuristic on top of the model with stress testing would, in most cases, be as accurate as the real verdict. That is, score should not improve by allowing more submissions.

he is a dumb g**.

Could be a naive question but: Do you guys (OpenAI) plan on watermarking the code generated by future models? It could make the process of detecting AI generated code much easier.

I did not say "speaking English," I said "knowing":

show a complete lack of reading comprehension

I hope for a virtual participation.

For each task and the 10,000 submissions, if the score distrubution histogram can be shared, it will be more impressive!

You might find the second editorial of this problem helpful.

I am sorry, but what he has to do with final in Astana?

Thank you! Added link to text

Question: why "intelliger" instead of "integer"? Is it an implication that integers are intelligent and sentient beings?

A slight correction on 1: the growth is in fact just $$$\mathcal{O}(\log \log n)$$$ on average.

Why the University of Maryland scored so low? Anyone knows? I was rooting for them

Most sane AverageDiv2People Exceptionalism

I think StarSilk. Best of luck!

gcd(x, y) = min(x, y) or gcd(x, y) <= min(x, y) / 2. so we can say there are O(lg(n)) distinct prefix gcds.

It didn't aged well

Just be more Chinese. You simply are not Chinese enough.

Insert some Chinese genes into your DNA.

And how was the performance of the model on Codeforces problems measured? Did it participate in rated rounds? Is it possible to reveal a username of the model on Codeforces?

Just to follow up, here is a post sharing our submissions! https://codeforces.me/blog/entry/134091

Swarthmore College will surely win!

You can always use a new device, such that the Safe Exam Browser effectively becomes useless..

ZhouShang2003 chen_zexing stasio6 North America, USA, University of California San Diego

I've submitted a bug. https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116768

okwedook Hi! The issue is now fixed. All thanks to MANAV_2005 for fixing this issue.

Does the mirror contest start at a specific hour or you can start whenever you want?

Apparently, after registering for the Mirror Contest, it says that the contest is scheduled to start in 2025

i can confirm i am indeed a real person lol

oh wow, I totally forgot about references lol. I took a look at it, and it makes way more sense now. Thanks!

Your prediction was given when MIT was out of top 50. But in the end they got into top 12 and you was wrong, lol.

Out of curiosity: can you share if there are any endeavors in problem setting?

Auto comment: topic has been updated by 5k_sync_closer (previous revision, new revision, compare).

What's wrong with it?

Kevin114514 will surely win WF 2024!

Sorry,YouTube is banned in China(Don't ask how I know this).

MHC Round 1 conflicts with the 2024 ICPC North America Qualifier (NAQ) which happens at 11:00 — 16:00 PT on October 5th. Considering many universities in NA use NAQ for team selection for ICPC Regionals, will there be another bye system setup like there was last year? Not sure how this would work since the only round before round 1 is the qualification round.

And alternatively, can future MHC rounds happen on Sundays since it seems like there will always be ICPC conflicts on Saturdays in October/November every year.

Miscalculated, but where...