Am I right that CP is slowly dying?
Today I got little rating bonus placing so high.
For example, compare Codeforces Round 1015 (Div. 1 + Div. 2) with today's round.
Back then, the performance of 38th place was 3046:
Now, the performance of 38th place is only 2862:
The performance of 2nd place back then was 3980:
Now, the performance of 2nd place is only 3660:
I also checked the participant lists. In today's round there were only 15 LGMs and 183 reds, while back then there were 32 LGMs and 261 reds, even though today's round had 24284 participants, compared to only 21544 back then.
Could it be because 30 minutes after the round the top of the standings is already full of gray accounts, and strong participants are simply losing motivation to compete in the LLM era?
What do you think?
P.S. Maybe it’s just a Thursday.
Could be that a lot of people no longer feel motivated to practice because CF rating is no longer useful for stuff like job applications because ai
...
I think sadly things have taken a hit. This round didn't feel so bad, but codeforces rounds 1097 and 1093 felt very sad. Record low div 1 participation from whatever I've seen, and the performance per rank changed noticeably.
I have hope. Cheaters are slowly getting handled better. In today's round, even though there were so many cheaters, many got cleaned up. I believe in the community to fight for whatever is left to fight for.
tbf, the actual skills were never really that useful for any real job.
let me guess, only good IQ is useful for real job, right?
well there are other things, but yes, the only reason why there would be (is) a correlation between job performance and competitive programming skill is cuz they both load on g to some extent. and for that reason, leetcode interviews probably aren't going anywhere just cuz of $$$AI$$$.
Transforming a useless array to another useless array
I think CP skills are useless from the very start, not because of AI. I don't think anyone would like to use a segment tree or something in real work.
You got +85 and you said "little rating bonus"?
And yes lots of GMs didn't participate simply because it's a Thursday.
And to explain why higher-rating participants dropped out (when it's Thursday) while lower-rating participants didn't, the reason could be that some CF users are in job markets that value CF rating and they are "forced" to participate in CF contests more eagerly, and are "forced" to cheat due to job-seeking pressure. I'm not defending cheaters, what I mean is that when you do CP for a number instead of to improve your CP skill, cheating is more likely to happen.
Sadly, AI cheating is becoming more and more serious. Three years ago, I wrote a blog discussing anti-plagiarism in LLM era and back then we didn't expect AI can reach 3400 rating on CF. And nowadays anti-plagiarism is much harder with false-positives (and on Luogu, too, I dealt with one in the previous week).
Hey, PinkieRabbit. I’m sorry you only got +58
And I'm happy with that!! You know, CF is a place where you easily get large negative delta and so hard to get large positive delta
I think As long as LLMs continue to make mistakes, strong fundamentals in algorithms and computer science will remain essential, if not even more valuable than before. They will be a major differentiator between people who can prompt their way to working code and engineers who can reason about correctness, scalability, trade-offs, and reliability.
I recently decided to stop competing because although I previously thought that it’d be somewhat likely for me to hit GM if I just got lucky a few times, I don’t think so anymore due to cheaters :(
Looking at solve counts vs how difficult I think problems are, I don’t even think I could sustain my current rating
I’m quite pessimistic about the situation in general, and think catching cheaters is hopeless. I think it should be pretty much undetectable unless someone doesn’t care that much about getting caught or is stupid
I find it especially sad that imo problem quality has gone way up in the last 5 or so years, and the ad-hoc meta has been great. All for the competitive integrity to largely go away due to some retard shareholders/CEOs pleasuring themselves to the idea of AI solving Olympiad problems
Olympiad problems are a very reasonable way to benchmark AI performance, so you can hardly blame them. Plus, AI has shown that it can do much more than solve just Olympiad problems in recent times.
I see, it's fair that it's a reasonable benchmark. I guess a large part of my negative sentiment against shareholders/CEOs is more against the massive investments that are put into AI in general
TLDR: One can still compete, but it requires more engagement, and it would be less global.
I think we can still keep competition enjoyable if the competition is replicated in a less cheating prone smaller set of participants with its own unofficial Elo system.
Say, people compete in some sports in real life and in online competitive games. All these activities often give almost none direct material gains, and in fact people spend money to engage in many of such activities (in contrast to Codeforces official rating, which is still valued, especially in certain regions).
An unofficial CP ladder for the sake of competition is somewhat similar, since it is likely to remove the value the official Codeforces rating has. I even implemented self-moderated system, however, it requires proactive community for it to work. Also, I don't claim that the proposed idea is the best/only solution for the problem.
Also, I don't share your pessimism about the complexity of the detection problem. The main problem with Codeforces is that it either doesn't have resources to match the issue scale, or doesn't care enough about cheaters. I think both. Because, say, this person XVIII who was flagged by the community in numerous blogs, still was contacted to receive material prizes (they got banned only after a month or so when they became LGM). By the way, NanYan is also in the prize winner list, yet a user with similar handle is banned on Atcoder.
Thanks for making that! I applied to join lol
Regarding detecting cheaters, I feel pessimistic because I suspect that there's a large number of them who are undetectable no matter how hard anyone tries. For example, it's pretty easy to just ask it for how to solve the problem, and then implement it yourself/review its implementation for anything that looks sus, etc.
You can only definitively say that someone is cheating if they do some dumb shit like submit some consecutive problems inhumanly quickly.
I don't even understand why you would use AI for CF contests. It's already undermined the value of CF ratings for hiring/college. Is it just for ego and street cred? It's not like you are learning anything valuable, simply harming the community and wasting your own time.
In my opinion yes, CP is slowly dying. I had an urge to participate in contests lately on CF but then I checked top200 for last couple of contests and checked distribution of lgm-igm-gm-master-cm-lower than cm for last couple of contests and for random couple of random contests in 2022 and everything became quite clear to me, I lost every bit of motivation to participate in online contests ever again.
Other point is I believe I(and almost every pre-2024 red person) can easily reach 3000 with using almost any kind of cheap AI and without getting caught which is kinda sad(and I believe there are already enough people doing it). I already saw what AI can do when I was writing Midnight Code Cup 2026 and I am quite sure everybody can just ask LLM to obfuscate their sloppy code to their style with enough effort to not get caught.
Somebody can say "you can just solve problems for fun", but that defeats the Competitive in Competitive programming.
Somebody can say "rating is just a number", but that number is basically what defines Competitive in Competitive programming.
You can still solve problems for fun when upsolving and you will not get frustrated with hundreds of gray-blue accounts above you that are using LLMs and can't be 100% proven as cheaters.
Wow, I had exactly the same thought about the word Competittive in Competitive Programming the other day.
The false positives were mentioned several times under this blog alone. I don't think a proper anti-cheat system absolutely needs 0 false positive rate. There are much worse things in life, like justice systems, which have non zero FPR. I brought up such example not as a normative argument, but to illustrate that some FPR can be generally well tolerated to some extent, and, some fraction of falsely banned participants could be outweighted by the increased overall integrity.
It's like competitive running vs. running. The second is the hobby, the first is the sport that gets Olympic medals.
Can't you just compare yourself to your friends and others who you know aren't cheating and still get almost all of the fun from the competitive part?
...
Competitive programming is a broad term. Your comment proves that online contests are dying. But I cannot say the same about onsite competitions, like IOI, ICPC, and many other national olympiads.
I feel like to explain your exact data (since you only gave 3 points; I'm sure you'd be able to discover a more convincing trend if we looked at more data) it just ended up being a weekday where less strong people were able to participate.
The perf of second place would go up significantly if there were more LGMs/GMs participating. If there are 100,000 cyans getting 2nd is probably like 2500 or so perf, but if you're 3000 and there are 100,000 GMs then getting 40000th or so would be 2500 or so perf.
Also having just a few GMs jacks up the perfs a lot — a 3000 is expected to beat a 2200 99% of the time.
Noooooooo. why r u writing this ??
Saturday for 1015 vs Thursday for last round does matter
Tourist only 3660?
As a gray coder, could this simply mean that there are more beginners, so possibly in a few years there could be a record amount of red coders?
I guess it wasn't just a Tuesday, today's contest looks pretty similar.