Codeforces Round 1000 (Div. 2) — Anti-LLM Evaluation Report As the main problemsetter of [contest:2063], I know that a lot of you were somewhat disappointed to see that it was not a Div.1 contest. Though I am not the one who caused a lack of Div.1 contests, I understand your feeling. So I present to you the great surprise: The **Anti-LLM Evaluation Report** for [contest:2063], the first of its kind on a Div.2! In this blog we discuss about how the round combats against LLMs (especially focusing on OpenAI o1, the greatest of its kind while we prepared and tested the problemset), by looking at the timeline of how the problemset changed. --- So, let us begin with the initial problemset we had when the testing began. (The task names are anonymized, unless they are released to the public before or in the round) In the beginning, we had a problemset that looks like this: - **A' — B — C0 — D' — E' — F0** (If you are confused with the meanings, (letter)' means it was not used in the final problemset but exis...

Full text and comments »

Codeforces Round 1000 (Div. 2) Hello Codeforces, and the legends of across $999$ and more rounds! ------------------------------------------------------------------ --- [user:wuhudsm,2024-12-25], [user:Yugandhar_Master,2025-01-15], and I are *beyond excited* to invite you to [contest:2063] at [contest_time:2063]. <b style="color:red;">Please note the unusual time of the round ($\color{red}{2.5}$ hours before the usual time)!</b> The contest contains $6$ carefully crafted tasks, one of them divided into two subtasks, to be solved in $2$ hours. You will solve tasks themed around Little John and his shenanigans aimed towards getting his own dream home (featuring, probably, galvanized square steel). This round could not exist without the thankful help of these so many people: - ~FairyWinx,2025-01-15 for <strike>rejecting more tasks than are used in the final problemset,</strike> coordinating the round and translating the problems; - ~rewhile,2025-01-15 for some very important "technical assistance"; - ...

Full text and comments »

Community Petition on Combating Cheaters Recently, cheating has become a more and more serious issue over time. Many community members have come up with various ideas for combatting cheaters better. I think I can make a summary. There are votes option for each point. Click the upvote button if you like it. ## Prevention - Enable "trusted participant" system for Div.1, Div.2, Div.1 + 2 and Educational rounds. - Use phone number verification for registration. (which has been rejected by Mike in his recent blog to some extent, but I'll let it be here anyway) - Use hardware level detection software. - Completely ban the usage of AI. - Use a new type of problem in which the usage of AI is marked as not cheating and there could still be much fun for human solvers, even with the assistance of AI. ## Detection - Train anti-cheating specialized LLM and use it for cheating detection. - Add a report option for each user for manual review. ...

Full text and comments »

A summary of what passes for an Indian ICPC Contest, apparently The online round for the India’s ICPC was held on 23rd. And also a week before that. The first time around, the organisers tried debuting a self-hosted DOMjudge instance. That didn’t go too well, so a recontest was scheduled on CodeChef. If you’re not too familiar with the Indian ICPC system — this online prelims is used to select teams that can participate in the offline regionals. From a single university, usually approx. 2 teams (for the top unis) or 1 (for almost all unis that aren’t in top-15) qualify. As you could guess, it’s quite easy for good teams to miss a spot. Looking at the first contest’s rankings, teams from rank 2 to 100+ all had 4 solves. Owing to severe system issues in the first contest, the time penalties weren’t too reliable differentiators, leading to the recontest this Saturday. In hindsight, we forgot to be grateful for the problemset and reliable testcases the first time. I’m sure everyone agrees that the basic requirements (ignoring all other stuff...

Full text and comments »

Déjà Vu: AI, Cheaters, and the Usual Chaos Hola CP-igos! As the name suggests _yet_ another blog an AI, cheaters etc. Introduction ================== Flashback to 1.5 years back... I just joined Codeforces, all contests were fine, no cheater blogs(and no cheaters)... I used to spend some time reading interesting blogs in catalog and the "Recent Actions" where I used to find blogs on other "non-educational" blogs, be it somebody's productivity plan, somebody reached GM, some new solutions on rounds, the "if u comment under this blog" blogs etc. All this was fun.. I could spend an hour or so, reading all this and the comments underneath them The Issue ------------------ Fastforward to now.... A round gets over.. And soon by the next day, all we have is tons of blogs on cheaters doing this, cheaters doing that, I found this, I found that and what not. And then people finding underneath blogs, seperate blogs being made for seperate cheaters, as though the cheaters have become kings, every set of cheaters with th...

Full text and comments »

Exposing a cheater + My opinions Some background &mdash; I use Python for CP and some benefits of using Python are clear and concise syntax, tons of in-built function, dynamic typing, FP-like operations (map, filter, accumulate etc.), list comprehensions, bool-to-int conversions and so on. These features make code-golfing very fun with Python. I discovered [user:jahnvisahni98,2025-03-10] in the solutions section of some problem few weeks ago (you can sort solutions by their size). If I find some user frequenting the section, I usually befriend them on Codeforces and even send a message saying I appreciate them. But this case was different. The solution's logic was still long, but the syntax had been shortened by giving a tab space of 1 character. I looked through some of her other solutions and she had a different code style in all of them. Somewhere she had formatted code with long variable names while at other places she had short undecipherable variable names with very poor formatting. This was still not a stro...

Full text and comments »

A short and simple anaylsis of Deepseek R1 <spoiler summary="Prompt"> Please think deeply about this problem, and write a solution using C++. Make sure to use map and set vs the unordered equivalents. Whenever you come up with a solution, ensure that it is correct and that it works within the constraints provided. Test your code on the samples along with generated cases + edge cases that could occur. If you are out of ideas, try and think some more, step by step. </spoiler> CF 800-1200 ----------- - [problem:1998A] - <span style="color:green"><b>AC in 199 seconds</b></span> [Submission](https://mirror.codeforces.com/contest/1998/submission/302525499) - [problem:2043B] - <span style="color:green"><b>AC in 147 seconds</b></span> [Submission](https://mirror.codeforces.com/contest/2043/submission/302525865) - [problem:2021B] - Initially misread the samples from screenshot (Thought for 410 seconds) - <span style="color:red"><b>WA on test 1 in 253 seconds</span> [Submission](https://mirror.codeforces.com/contest/2021/submis...

Full text and comments »

An Open-Source Retrieval Model and Service for Competitive Programming Problems Over the past few years, the number of *duplicate* or *nearly-duplicate* problems in competitive programming has grown noticeably. You can find several discussions about this on Codeforces, for example: * [https://mirror.codeforces.com/blog/entry/113064](https://mirror.codeforces.com/blog/entry/113064) * [https://mirror.codeforces.com/blog/entry/113251](https://mirror.codeforces.com/blog/entry/113251) * [https://mirror.codeforces.com/blog/entry/118883](https://mirror.codeforces.com/blog/entry/118883) To address the issue, we collected a variety of data — problem–solution pairs, duplicate-problem pairs, and pairs of full vs. simplified problem statements. With these data we built a dedicated **benchmark for problem retrieval** and trained a **special-purpose retrieval model** that currently achieves the best performance among open-source models in this domain. You can try the model-powered search service live at **[https://cpret.online/](https://cpret.online/)**. About two years ago, the project [yuantiji.ac](http...

Full text and comments »

Performance Ratings vs Ratings of Solvable Problems for the Recent GPT Models The GPT models have made great advancements in competitive programming recently. I saw from [user:Qingyu,2024-12-21]'s [blog](https://codeforces.net/blog/entry/137534) earlier that the new o3 model claims to achieve a 2727 performance rating. However, I'm slightly uncertain of how this 'rating achieved' should be interpreted. There is very little detail publicly available on how the performance ratings are computed. The only available detail I could find regarding their benchmark methodology is the following quote from the ['reasoning with LLMs' blog](https://openai.com/index/learning-to-reason-with-llms/): > Finally, we simulated competitive programming contests hosted by Codeforces to demonstrate this model’s coding skill. Our evaluations closely matched competition rules and allowed for 10 submissions. This does not really offer any detailed computation methods. Given the limited details, I'm hypothesising that the LLM doesn't need to solve problems anywhere close to the 270...

Full text and comments »

REPORT LLM CHEATER!!! [user:xiaowang5242,2025-05-25] is a **cheater**. 1. Do you belive a $\color{cyan}{\texttt{Specialist}}$ can reach $\color{orange}{\texttt{Master}}$ and get $\color{red}{\texttt{2550}}$ Perf in only **1 month**? <spoiler summary="Spoiler">   </spoiler> 2. He has already been caught cheating in Luogu(LGR-208) because of using AI <spoiler summary="Spoiler">  </spoiler> 3. He got an extremly **low** score in TianTi competition which just held **recently** ,according to this [post](https://www.zhihu.com/question/1900737340269401476/answer/1901413487093219761) in [知乎](https://www.zhihu.com/),(It is a Chinese website, you may not enter it) <spoiler summary="Spoiler"> ![ ](https://mirror.codeforces.com/predo...

Full text and comments »

About Cheaters on Round 1082 Oh yeah, I did notice that there are a lot of cheaters in Div2G/Div1F, and we know a few types of solutions that we believe are from either a strong LLM or a leaker. Fortunately, the number of ACs here are not in the thousands, so it shouldn't be hard to sort most of them out, and I think a majority of them will be skipped/banned in the usual plagiarism check. That being said, the CF admins (and myself) are busy people, and we can't just check all submissions on every problem, as well as every blog about cheaters. So I think it's better to group the reports in one place! Please write down in the comments here if you find either: 1. One submission/user that seems to have blatant evidence of LLM cheating 2. Multiple submissions that look abnormally similar to each other (either in code style, or code logic) Please note that case 2 doesn't include when it's about one solution that obviously almost everyone will implement, just like Div2D's intended solution. You don't have to...

Full text and comments »

Newbie here! N/A in other solutions code and cheating Hi there! I'm new here, and also to CP in general (started less than 2 months ago). I took the Round 1076 (Div. 3) which I enjoied :) I'm actually a Codechef refugee, and here rules are a little different... I looked around the blogs for info but I'm missing a few pieces so here are my two questions: 1. Why I still see N/A under other people submissions? (I can't see their solutions). Reading around, people were guessing that it was a measure to prevent crawling, but am I not supposed to see other people code at least during the hacking phase? 2. What measures are in place to prevent cheating (plagiarism, LLMs) and how can I report anyone violating the rules? Thanks! **Update:** for the people wondering about point 1, since It does not seem to be documented anywhere here's my guess: the code is visible if your contribution points are positive. That explains why I was not be able to see them in the beginning even during, after a context, and after being rated (had 0 cont...

Full text and comments »

Report AI cheater in Hello 2026 I find the guy [user:not-successful-yet,2026-01-07] is a AI cheater. ##Evidence 1 [submission:356808855] This submission contains a strange, unused function `vertree`, this function checks whether the input tree is connected. In fact, it was a trap specifically designed by the test creator for LLM. This is only checked in AI-submitted answers. ##Evidence 2 This guy submited completely different solutions for problem I.I don't think humans could come up with so many different approaches in such a short time. ##Evidence 3 His submissions were very strange; He couldn't even solve a D problem in goodbye 2025. Please ban the cheater!

Full text and comments »

We need to save codeforces Codeforces used to be the one place I felt truly challenged. Now, it feels rigged. Does anyone else remember Codeforces during last year's summer? I do. I remember waking up early to do the frequent and high-quality contests; I remember getting on calls with friends and discussing solutions post-contest and laughing at WA attempts; I also remember times where when I couldn't think of a solution and I would bang my head on the wall and draw trillions of random dots, parenthesis, letters and numbers on paper &mdash; all in the name of hoping that I would stumble upon an observation. I grew, the community grew, and we all thrived. We've all been there, fallen in love with competitive programming and problem-solving. That feeling of getting AC after debugging for an hour and then finding out it was a random global variable is ... indescribable. Codeforces used to be a haven for us problem-solvers. It used to be a place where we could *be ourselves*. Ratings weren't everything for us: we...

Full text and comments »