Broke 2000 today :)
I argue that if you want to get to your goal rating $$$r$$$, it's decided by how well you can solve problems with rating $$$r-200$$$, and potentially not the classical wisdom of "really struggle with hard problems".
Here's a few examples from contests I've done well in:
- Round 1087. I got a 2047 performance with the hardest problem I solved being D, 1800
- Nebius Round 2. I got 1973 performance with D, 1900 (Maybe not the best example, since I was quite slow for C2 and D, spending an hour on each. But this still shows my point kind of.)
- Educational Round 190. I got 2269 performance with E, 2100
- Round 1101. I got 2252 performance with D, 2000 (according to CList so far)
A few older examples:
- Round 1085. I got 1871 performance with C, 1600 (and I took an hour to solve it)
- Round 1081. I got 1645 performance with C, 1300
Basically point being, solving a problem of rating $$$x$$$ can give you a performance near $$$x+200$$$, or more, if you're fast. While I think the classical wisdom of practicing hard problems to learn new ideas is good, it's also worth emphasizing the skills that will actually get you good performances in round.
Additionally, all my bad contests have been from failing early problems which messes up your momentum for the hard problems. If I try to skip the early problem and jump ahead, I end up jumping back and forth between the two problems instead of sitting down and really thinking about one, and making no progress on either. Getting good at easy problems helps with this.
Also, I think practicing implementation speed and accuracy can be very helpful. You lose a lot of time and mental momentum if you spend 25 minutes debugging an out of bounds error or overflow or some stupid mistake on problem C. I grinded through like 100x 1600 and 100x 1700 problems and I think I am noticeably better at it. I also don't have the "mental aversion to coding" that I used to where I would dread having to implement and WA2 and debug.
TLDR: Don't forget to practice problems slightly below your rating. And also get good at implementation.
Stop gatekeeping the real secrets like leaving beautiful comments like this that a real human boy would definitely make
EDIT: since I replied saying this but it got pushed to the bottom, I'll put it in this comment
"sorry guys I post stupid things late at night then regret them the next day
I probably sound like I'm begging for forgiveness but I don't know what else I'm supposed to say
I'll just shut up for now"
Yes, I used to write comments when I started. If you're not convinced you're welcome to go check my ICPC results next year.
also which submission even was this btw i forgot
edit 3: oh yeah this comment is definitely ai
(Kentq is a professor at my university)
Edit 4: I put this problem through chatgpt for fun and it seems to be spewing bs. Do let me know if it actually solves the problem correctly
You are attempting to prove by example that you don't use AI in contest, and prove using the future that you don't use AI in contest
Also, if you really weren't using AI, would you have edited this 5 times to desperately try to prove your legitimacy?
Would also like to note that "when you started" is approximately 10 hours ago as of writing this, you've been starting for quite a while
How can I prove a negative?
maybe three months ago you could prove you weren't ai by doing like 2k performance but idk anymore man ai getting too good. Although the burden of proof is upon the person accusing you so all you have to do is disprove the evidence against youy which you seemed to do
Burden of Proof
I'm pretty sure that's because you copied it wrong, and it removed the exponent in 2^n
Making comments is now AI? Ok I guess I have been using AI since 2020 if that is true.
If you are typing code without thinking, you will not solve the problem. You must think about the problem first, figure out the main observations, then find a solution.
no I've used comments too, but not things like "There is an answer if there is an answer"
That is not what his comment says. Reread the problem statement.
tbh i tend to like to ramble in the code editor. you can look at the 200 submissions ive done in practice
Maybe that's why you are stuck on pupil
> usually gets specialist performance
> loses rating ONCE
> "stuck on pupil"
Because you're the authority here for taking 33 contests to get to specialist
Haha just keep coping, kid. Dont face the reality. Keep living in that sweet dream of you are a specialist because you "usually get" spec perf.
Few tips, maybe you should spend your time training rather than accusing someone cheating out of pure jealousy and blaming "cheaters" for your rating and not realizing you have skill issue.
Ok thank you now I know never to interact with you again
Attending more contests -> solve more problem -> getting better at cp.
idk I (and I imagine enough other ppl) sometimes write pointless comments lol
dunno how that makes it AI lol
Ah yes, because comments are made to distinguish AI and human not to actually put an explanation of what a code blocks does
will this count in AI too..?
Yes.
(Also how tf did a blog about solving lower rated problems being better for contest performance turn this heated)
I think that is because for one legit case (that you probably are), there are multiple cheaters with similar rating dynamics.
Moreover, some cheaters, for some reasons, tend to make similar but more AI sloppish blogs and comments about "their" achievements, "their" standpoint on practice, problem solving, algorithms etc. My theory is that they do it to build credibility.
I'm surprised that you expected a different reaction given your conditions. Namely, your rating graph, which is CM level performance from the start in 2026.
Also, speaking about the actual content of your blog. I think that you were already quite good enough when you started. Therefore, in my opinion, the actual improvement (i.e. the difference between 2 states: the start and the current state) is more modest than your blog suggests. But I'm glad that we are both on the in my opinion team, so take it with a grain of salt.
However, I agree that solving easier problems is helpful. I also try to include them in my practice. Not only it improves time from opening statement to AC and routinizes the process to get AC, it also improves your guessing ability, which is, I think, the core of Codeforces contests.
Moreover, I think that for many people mastering speedforcing skills is a very good strategy to gain rating EV wise (but in short term). Specifically, for people who initially are not good enough (both problem solving skill and talent) to approach harder problems.
PS: Adding TLDR in the end of the blog is trolling.
Take this blog, for example. https://codeforces.me/blog/entry/153992.
At first glance it looks legit. Yet, the author RainRecall is very likely a cheater. I will not provide evidence to not make cheating life easier, though.
Btw, you can watch how this two comments are going to be downvoted by cheaters.
why
I don't think you can assume someone cheated by using AI just because their coding style resembles AI-generated code.
I can't help you if you miss basic text comprehension skills.
I may understand what you mean. This article does appear to have been generated by AI, but in reality, due to my poor English, I wrote it in Chinese and had ChatGPT 5.5 instant translate it into English (yes, I have already noted this at the beginning of my blog), which makes this article appear to have been generated by AI.
If you still don't want to believe it, I can put the Chinese version of this article and the Chinese versions of several other blogs I have compiled under this comment.
I'm sorry I have to use ChatGPT translation due to my poor English.
Barrett Reduction 通常可以更快的解决下面的问题：
直接做取模运算（如 C++ 中的
x % m）通常会比较慢，因此考虑对其进行常数优化。Barrett Reduction 算法会先预处理出 $$$\frac 1m$$$ 的近似值。选择精度参数
Unable to parse markup [type=CF_MATHJAX]
\mu=\lfloor\frac Bm\rfloor$，则此时有 $$$\frac xm=\frac{\mu x}B$$$。因此可以估算商 $$$q$$$ 为 $$$\lfloor\frac{\mu x}B\rfloor$$$，此时显然有余数 $$$r=x-qm$$$。 因为商 $$$q$$$ 是估算得到的，因此余数 $$$r$$$ 也需要做一定程度的偏移。具体细节参见代码实现。使用方式如下：
调用
bt.reduce(a)可以得到a % mod的值，调用bt.mul(a, b)可以得到a * b % mod的值。The reason why I am writing this blog is that when I was solving this problem in O(n^4) time complexity(I know n^4 cant pass it in theory, but my friend told me that it can actually pass it), my code obtained a Time Limit Exceed judgment due to a large number of repeated modulo operations. After inquiring in the discussion forum, someone told me that Barrett Reduction can be used to avoid the problem of slow mold taking.
Is it enough?
Solving 3000 is orz
Yeah, orz indeed. If only they had solved it themselves. Sadly, you didn't bother to inspect that account more closely. Not that you had to do it, no. But it's still sad. It means that cheaters can blatantly cheat, and most people won't care even when they're given hints and evidence is right before their eyes.
Ok, I was joking but TBH the cheating witch hunting stuff is getting out of hand. I see this in sports too where people are like "everyone better than me is using drugs". I don't think this mindset is productive, especially when we already know that 3000+ LGMs are possible since they have existed even in the pre-AI era.
I also can't help but notice that it's never Div 1 people doing these things. Not trying to throw shade, but I do think your (limited!) time and energy is better spent doing other things.
By all means, expose cheaters if they have fjString or gdCode in their submission, but don't go witch hunting.
You contradict yourself. How it is possible to have LGM cheaters when "it's never Div 1 people doing these things".
I meant to say Div 1 people witch hunting
Oh, I see. Then I can only recommend you to check, say, the last contest announcement blog and see for yourself if "it's never Div 1 people doing these things".
People shouldn't care about ts. Rating is just some online numbers dude, not that big of a deal. If people cheat, then let them be. It's their loss anyway
I don't fully agree with this because people who cheat do somewhat devalue the value of my rating if I decide to put it on my resume.
That said, just because murder is bad and illegal does not mean you should say everyone who has comments in their code is a murderer. In the US legal system we have this idea of "innocent until proven guilty" where in order for you to get punished for a crime the prosecutor has to prove beyond a reasonable doubt that you did said crime.
Elysia :flushed:
Which website is this?
"I will accuse you of cheating, but i wont tell you why. If you disagree with me then you are also cheater."
What kind of logic is this man?
The worse thing about cheaters are it create these "cheater-hunter"s who would accuse everybody performing well are cheating because they cant perform that good
I won't publicly tell why because providing such information helps cheaters to hide their cheating.
I would tell it to any legit contestant, who is smart enough to understand the reasons for not publishing such information, or better, to admins (but they don't care enough). You don't fall into either of these categories.
Also, this part " If you disagree with me then you are also cheater." is purely your imagination. I've never said that.
Anyway, I'm done with this discussion. Enjoy upvotes by cheaters.
I think you're a cheater. If you provide evidences, then you're hiding cheating. If you don't provide evidences, then you cannot prove you aren't a cheater. I won't provide evidences because they will help you cheating. So you are definitely a cheater.
"I won't publicly tell why because providing such information helps cheaters to hide their cheating."
https://en.wikipedia.org/wiki/Habeas_corpus
Wait this wikipedia article sucks. Basically if you're accusing somebody of something they have a right to see your evidence against them, regardless of whether or not this helps future criminals know what tools you have. Otherwise you can just become a tyrant who has the diary of a wimpy kid lie detector that's just a printer with a guy hidden inside.
After inspecting more of RainRecall's submissions to the point of getting limited for "crawling" (I reviewed submissions on my phone), I came to the conclusion that I got a false positive here with high probability.
RainRecall, I'm really sorry about that.
An unsolicited advice: get better at using AI agents, so that the text looks less AI, as there are a lot of people here who post AI slop and don't know what they are talking about. Speaking of the blog, it wasn't my main evidence. Nevertheless, I've just found a reasonable explanation to other patterns I discovered 16 hours ago.
I don't consider my actions witch hunting, as I always ready to privately provide evidence to interested people, who are not cheaters and who understand that publishing findings benefits cheaters.
Also, both false positives and false negatives are the inherent part of binary classification. Meaning that such errors are part of the game, which I don't plan to stop playing.
sorry guys I post stupid things late at night then regret them the next day
I probably sound like I'm begging for forgiveness but I don't know what else I'm supposed to say
I'll just shut up for now
Auto comment: topic has been updated by greateric (previous revision, new revision, compare).
greateriorz
Yes, I too observe that and it's quite a real thing — I often stalk the best 4th year CPers of my school and see the graph of problems solved by rating, and problems solved by them corresponding to their actual rating are often less than expected.
But i think it depends on the next problem you dont solve. Like if the next problem rating is 1700, no matter how fast you solve it your perf would be capped around 1700 (this is purely just my guess tho)
You've got a point there, but since more and more people are using LLMs to compete on Contests, it's getting increasingly tough to boost one's rating.
https://codeforces.me/contest/2193/submission/359868555
These cheating accusations lowkey pretty amusing to read :yayy:
Besides I think writing comments is lowkey a good practice since it makes tracing your implementation much easier. There's a lot of times where I completely forgor what my DP state was cuz I didn't write it down. This is especially true for harder problems where implementation can get as long as 200+ lines. I understand the stigma for commenting, especially in Comp Prog where speed is an important factor, but it doesn't necessarily means we are antagonizing every single person leaving comments in their code.
Ok Mr "I define dp[i][j] as" go take 381 and learn to write recursive spec it will help you solve Div 2 D :)
Bro I just solved 3k4 earlier today...
my rating is 6 points higher
Son, flaming people rating is bad :(
Halzyonnn orz
Strongly agree, I think that what entails reaching a rating r is becoming extremely consistent at problems r-100 to r-300 or lower. If you randomly spend 1 hour solving a problem -200 below your rating, this will cancel out any positive effect of solving a problem +200 your rating in contest. Maybe the floor is too overlooked. I discovered this independently and I started practicing differently (on my main)