unalive's blog

Edu F vs GPT
By unalive, 4 weeks ago, In English

https://codeforces.me/contest/2026/submission/288600046
https://codeforces.me/contest/2026/submission/288593410
https://codeforces.me/contest/2026/submission/288601081

It's so over for contest enjoyers.

  • Vote: I like it
  • +72
  • Vote: I do not like it

»
4 weeks ago, # |
← Rev. 2   Vote: I like it +34 Vote: I do not like it

just feel that i'm so stupid that i cannot even win chatGPT :(

→ Reply
  • »
    »
    4 weeks ago, # ^ |
      Vote: I like it +10 Vote: I do not like it

    divided by rating, united by thoughts

    → Reply
»
4 weeks ago, # |
  Vote: I like it +851 Vote: I do not like it

Hacked :)

→ Reply
»
4 weeks ago, # |
  Vote: I like it +1 Vote: I do not like it

what do u mean

→ Reply
»
4 weeks ago, # |
  Vote: I like it 0 Vote: I do not like it

CP is Chess now.

→ Reply
»
4 weeks ago, # |
  Vote: I like it +108 Vote: I do not like it

I am impressed that people even attempted to use ChatGPT on problem F, something that is 1400 rating higher than ChatGPT. ChatGPT must feel flattered if it could feel.

→ Reply
  • »
    »
    4 weeks ago, # ^ |
      Vote: I like it 0 Vote: I do not like it

    is this the end?

    → Reply
  • »
    »
    4 weeks ago, # ^ |
      Vote: I like it -31 Vote: I do not like it

    meanwhile when chatgpt will get that much level CP will become dumb thing?? NO i don't want to listen this..these above all three are indian guy -_-

    → Reply
»
4 weeks ago, # |
  Vote: I like it 0 Vote: I do not like it

that's some next level prompting skills for sure

→ Reply
»
4 weeks ago, # |
  Vote: I like it +8 Vote: I do not like it

Feel like this is more of an L on the authors' part to not make strong enough tests, where even brute force works.

→ Reply
  • »
    »
    4 weeks ago, # ^ |
      Vote: I like it +28 Vote: I do not like it

    If you think about it, this might actually be the way to go, since offering full feedback makes it much easier for people who are unskilled at cp to gpt their way through problems.

    Does it worsen the experience for other people? Yes, but I’d prefer to have weaker pretests as compared to hundreds of gpt greys above me in the final ranklist.

    → Reply
    • »
      »
      »
      4 weeks ago, # ^ |
        Vote: I like it 0 Vote: I do not like it

      This actually is a great idea the authors can feed their question into gpt and design the testcases in such a manner that this solution passes the pretests. But another problems could arise that people could easily hack these solutions and get points.

      → Reply
    • »
      »
      »
      4 weeks ago, # ^ |
        Vote: I like it +60 Vote: I do not like it

      Maybe Hacker Cup was right all along...

      → Reply
»
4 weeks ago, # |
  Vote: I like it +36 Vote: I do not like it

So, is it acceptable to describe the grey and green as borderline retarded? This is clearly rude, insulting and retarded at the same time.
I don't know much about Codeforces Code of Conduct but it can't be that one can just insult others like that.
Whether the blog author meant to refer to the particular grey and green participants mentioned in the blog or to all people with these ranks, I think it's equally unacceptable and this blog should be edited or deleted.

→ Reply
  • »
    »
    4 weeks ago, # ^ |
    ← Rev. 2   Vote: I like it +76 Vote: I do not like it

    I think it's equally unacceptable and this blog should be edited or deleted

    No it's not. Cheaters are borderline retarded. Newbies and pupils are not.

    → Reply
    • »
      »
      »
      4 weeks ago, # ^ |
        Vote: I like it +31 Vote: I do not like it

      I disagree. Cheaters are just cheaters. If you prove someone cheats, then you apply whatever rules you have for that.
      Anyway, the author is clearly describing all the grey and all the green, not the two cheaters in the post.

      → Reply
  • »
    »
    4 weeks ago, # ^ |
      Vote: I like it 0 Vote: I do not like it

    womp womp

    → Reply
»
4 weeks ago, # |
← Rev. 3   Vote: I like it +8 Vote: I do not like it

I'm more sad about the fact that it took me 50 minutes to carefully implement D, although I've got the idea instantly. And o1-preview solves it in less than a minute. Guess the "borderline retarded" goes all the way up to the 1836 rating at least.

→ Reply
  • »
    »
    4 weeks ago, # ^ |
      Vote: I like it 0 Vote: I do not like it

    The glimmer of hope here is that there are people way above 1804 rating. Meaning AI still can't beat all of us and we have the potential to be better than o1.

    → Reply
»
4 weeks ago, # |
  Vote: I like it -19 Vote: I do not like it

Side note I think Educational Rounds should be unrated.

→ Reply
  • »
    »
    4 weeks ago, # ^ |
      Vote: I like it -7 Vote: I do not like it

    Give it as an unrated contest if you want to codeforces allows you to do that,why cry?

    → Reply
    • »
      »
      »
      4 weeks ago, # ^ |
      ← Rev. 2   Vote: I like it +26 Vote: I do not like it

      Because many people who does educational rounds to get $$$X$$$ rating does not have the capability to get $$$X$$$ rating with regular rounds, including me in the past (namely $$$X=1900$$$). That ruins the point of rating. Knowing some classical tricks does not mean you can solve real problems of the same difficulty.

      → Reply
      • »
        »
        »
        »
        4 weeks ago, # ^ |
          Vote: I like it 0 Vote: I do not like it

        omg, do u mean that edu rounds are easier ?

        → Reply
        • »
          »
          »
          »
          »
          4 weeks ago, # ^ |
          ← Rev. 2   Vote: I like it +20 Vote: I do not like it

          Their problems are more classical which means you can usually find similar techniques in other problems or even books or lectures. To be good at them you need to learn more classical techniques like binary search instead of improving your problemsolving mindset.

          → Reply
      • »
        »
        »
        »
        4 weeks ago, # ^ |
          Vote: I like it -20 Vote: I do not like it

        Again how does it affect you? Your rating is only dependant on contests YOU choose to give being rated. Dont give edu rounds and skip the inflation in YOUR ratings,simple fix.

        → Reply
        • »
          »
          »
          »
          »
          4 weeks ago, # ^ |
            Vote: I like it 0 Vote: I do not like it

          So what's wrong giving suggestions? Besides, it doesn't affect me either way because I'm Div.1 and forcefully unrated in those contests. And it's not about inflation. It's about these educational rounds serves more educational purposes than actual competition so we might want to exclude them from regular ratings.

          → Reply
      • »
        »
        »
        »
        4 weeks ago, # ^ |
          Vote: I like it 0 Vote: I do not like it

        I guess that's true, my average rating change in last 4 educational rounds is +81, but I usually loose rating in regular rounds.

        → Reply
»
4 weeks ago, # |
  Vote: I like it +18 Vote: I do not like it

Plz don't call me retarded I am trying to improve :(

→ Reply
  • »
    »
    4 weeks ago, # ^ |
      Vote: I like it 0 Vote: I do not like it

    Dont think so.The author used such word just to further satire those cheaters.You're not "the grey".You are a grey Nowbie who intend to improve by oneself.

    → Reply
  • »
    »
    4 weeks ago, # ^ |
      Vote: I like it 0 Vote: I do not like it

    He meant to say regarding the "grey" and "green" in the mentioned submissions, not the grey and green as a whole!

    → Reply
    • »
      »
      »
      4 weeks ago, # ^ |
        Vote: I like it +10 Vote: I do not like it

      No, he meant the grey and the green as a whole. And he meant to make what he meant so clear by not including the third cheaters (the blue one).

      → Reply
»
4 weeks ago, # |
← Rev. 3   Vote: I like it 0 Vote: I do not like it

ah , don't forget A as well , I hacked 7 submissions that works on $$$O(XY)$$$ , $$$O(\min(X,Y)^2)$$$ , $$$O(X^2+Y^2)$$$ which are obviously trash under current constraints.

Hacks:

Hacks Party
→ Reply
  • »
    »
    4 weeks ago, # ^ |
      Vote: I like it 0 Vote: I do not like it

    I don't think these people cheated though... I think they just couldn't think of a better construction...

    However there was also 288542915 this... this guy fully KNEW the construction, but decided to run some random nonsense loops before outputting the construction... and the saddest thing is, I couldn't even hack him with the worst case (when the nested loops run to 999 and 1000 respectively, and there are 5000 testcases)...

    → Reply
  • »
    »
    4 weeks ago, # ^ |
      Vote: I like it 0 Vote: I do not like it

    how can i be good at math like u

    → Reply
»
4 weeks ago, # |
  Vote: I like it +81 Vote: I do not like it

This post actually inspires a great way to combat cheating using GPT -- simply make the pretests weaker so that those brute force solutions by GPT will be allowed to pass pretests. As displayed by the previous OpenAI blog on CP, the performance of the model increases quite significantly when the number of allowed submissions increases; in addition, it is known that AI performs worse when the feedback it receives is not 100% accurate (i.e. pretest passed but FST). This really seems like a plausible way to reduce AI's effectiveness while affecting a genuine human solver much less (any competent contestant submitting an O(n^2) brute force to a n=10^5 question should know they'll FST anyway).

The above can be done in multiple ways, e.g. not including a max test in the pretests, which also helps to reduce the pretest judging time. A downside to this is that people can now hack all of these brute force solutions to get a lot of points -- maybe we can redesign the hacking system in some way. I'm sure there are other better methods than this, but this is just a suggestion for a starting ground.

That being said, if CF does want to take this path, it might be beneficial to make an announcement about this, mainly to protect the newer contestants at CF who have been familiar with the strong pretests these days so that they do not get frustrated unexpectedly.

→ Reply
  • »
    »
    4 weeks ago, # ^ |
      Vote: I like it +4 Vote: I do not like it

    There's another downside to the above example method -- people can now submit bruteforce to a difficult problem on an alt, lock it and copy a legitimate solution from the room on their main.

    Maybe it's time to reconsider in-contest hacking in the GPT era... But maybe someone can come up with a clever method that preserves the hacking system while still making the above cheat-combating method work.

    → Reply
    • »
      »
      »
      4 weeks ago, # ^ |
        Vote: I like it +28 Vote: I do not like it

      I don't think very many people want to preserve in-contest hacking.

      → Reply
      • »
        »
        »
        »
        4 weeks ago, # ^ |
          Vote: I like it +48 Vote: I do not like it

        How about just making pretests weak and not allowing in-contest hacking?

        → Reply
    • »
      »
      »
      4 weeks ago, # ^ |
        Vote: I like it 0 Vote: I do not like it

      is this water on her nose or something else ?

      → Reply
    • »
      »
      »
      4 weeks ago, # ^ |
        Vote: I like it 0 Vote: I do not like it

      Maybe it's time to reconsider in-contest hacking

      Yes, maybe it's even possible to create a separate short phase after the coding where people can challenge solutions of others and get points for that. Like, imagine being the top coder in your room just based on hacks. But on codeforces I don't think even a single round matches a format like that, not that I remember.

      → Reply
    • »
      »
      »
      4 weeks ago, # ^ |
        Vote: I like it 0 Vote: I do not like it

      If there is a concern that participants might hack GPT brute-force solutions, it could be possible to run system tests immediately after the contest and only open up hacking afterward.

      → Reply
  • »
    »
    4 weeks ago, # ^ |
    ← Rev. 2   Vote: I like it +24 Vote: I do not like it

    any competent contestant submitting an O(n^2) brute force to a n=10^5 question should know they'll FST anyway

    MrDindows will strongly disagree with you

    → Reply
  • »
    »
    4 weeks ago, # ^ |
      Vote: I like it +24 Vote: I do not like it

    I like the idea in principle, but there have been several problems where constant factor is a real issue and having max tests in pretests is our main line of defense to measure those things (I don't want to have to make random max tests and test those in custom invocation for every problem).

    For a very recent example, many solutions to 2035F - Tree Operations with the right complexity got TLE in pretests, as that problem requires a low constant implementation.

    → Reply
  • »
    »
    3 weeks ago, # ^ |
      Vote: I like it 0 Vote: I do not like it

    I don't really think so. You can counter that by local testing i.e. writing brute-force code to check the correctness of the produced program, and benchmark to see whether they can run within the time limit. I'm sure a green or a cyan will be more than competent enough to do those things. The only exceptions where I think this strategy would fail are those problems where generating strong tests is extremely difficult, such as graph problems, but they don't appear often enough to prevent cheaters from still having high performance. Codeforces best bet is probably just let them do what they want, cause they are gonna leave the platform after like 5 contests to get that juicy interview anyway.

    → Reply
»
3 weeks ago, # |
  Vote: I like it 0 Vote: I do not like it

GPT is such a noob doesn't know how to calculate complexity

→ Reply