As some of you might be aware, I decided to try pretty hard for Candidate Master last quarter. The main thing I did was try to take as many virtual contests as possible, particularly at the start of the quarter, kinda following from what lnsuyn did here. What I realized in hindsight was that it's hard to commit to regular virtuals and actually upsolving problems when you're a full-time college student. Anyway, that's beside the point. I did 17 virtual contests, and I found the results quite interesting.
I took progressively newer contests over time, and I made a fake rating for myself starting at 1600, so I could see how my virtual rating would change. I used carrot to calculate the performance rating.
Obviously, there are some limitations, as I included Div.2s as rated contests even if I had a rating over 2100, but as a current Div. 2 contestant, I think it gets my point across.
It's pretty easy to see that even in contests as late as the 850s, I had maintained candidate master. In the very last contest that I took (which was unrated later), I solved ABDEF, and would've ranked 15th officially (although I took the virtual without problem C, which was impossible to solve correctly, so this shouldn't be taken too seriously).
It's hard to tell how much of this comes from contestants being better, AI cheating, or from virtual contests being a better environment for performing well.
Another interesting thing is that in the earlier contests, I was consistently getting master to grandmaster performance, so you can see that contest difficulty has increased significantly since then.
Auto comment: topic has been updated by 0mar (previous revision, new revision, compare).
Sheesh, thats quite a difference (Of course I'm just a newbie so this rarely applies :sob: but still)
professional newbie larper ahh
I'm LarpQ ong
wait yooo larpq wsg
yo wsg whats ur discord cuz idk u from cf handle
larp larp larp sahur
im djikstran india in male cs on usaco server
I’m Um_larp bro
Problems haven't gotten much harder since then. This AI cheating is just throwing rating out the window. I know for a fact there are a lot of ppl who are WAY better than newbie and pupil but are stuck cuz of ts. And I also know that there is a ridiculous amount of ppl who are expert/CM and don't know shit. Moral of the story: I don't trust rating anymore, and neither should you.
I think rating inflation can also be explained this way: as the time passes more and more people learn about CP and thus more and more talented people emerge. To maintain the same challenging diffculty of contests, people come up with much harder problems, since competition becomes higher. Also I think I belong in the category of people, who know nothing about the real product development.
Modern problem ratings for the same difficulty are actually lower than older problems because so many cheaters solving them reduces their calculated rating
Well, that's what I meant, I might have rephrased it poorly (about difficulty part)
Also, most cheaters aren't "talented people"
I kinda wanted to give an alternate version of what's happening, leaving off cheaters part
https://codeforces.me/blog/entry/154620
Since the rating deflation is so high, if you are able to maintain mid-expert, you might be CM skill. If you want to be CM, you probably need master skill nowadays tbh
Yeah , you are right! Btw , congratulations for authoring an upcoming contest in codeforces. Hope the problems will be interesting!?
Just a reminder to the readers of this post that there is an ongoing effort to create a separate rating for trusted participants: Codeforces Community Rating
As one of the problemsetter I'd like to tell you it's simple as we found that previous tasks are used to practise and more educational materials provided. So just after a few years we're facing participaints much stronger and have to make harded tasks to keep them from getting bored. Both participants ability and problem difficulty rocket up faster then our thought.
May I ask about the name of the extension/website you're using to get the metrics in the photo?
I’m manually adding contests to google sheets. If you’re interested I can dm you a link
Can you send me too pls
Anytime cf. Automatically analyses your perf in VCs and updates rating
big brother his IQ is just insane i know him personally
Hmm, now you with your friends are creating contest i hope it will be nice
I do think it is a mix of all the aspects you mentioned. the later contests in the photo are 850s, which are from the beginning of 2023. at that time, i guess that llm cheating wasnt really so big of a thing (I may be wrong, but i used llms back in the day for some basic programming stuff and most of the time i could outperform almost every free model, even being only an specialist), so i guess that llm didnt do so much of a difference then. And we also have to keep in mind that codeforces is relatively new. it was created in 2010, and the 550s contests are from around 2018, so i guess the participants have gotten a lot better since. And in my personal experience, VCs are way better to perform, i guess the pressure is a lot lower.
Great statistics by the way, loved them
brazil > argentina pele > maradona
I agree with you with the conclusion, But I think using ancient contest before 2020 is not a scientific way of measuring you true ability, because these tasks are now very classic and participants knows very well how to solve them.
If you want to measure your real ability, you should mock with contests that took place in the year of 2024 and 2025. These two years although we still have AI cheaters, but they are mostly stupid and has been ruled out during the rollback.