Hello,
I wanted to share some thoughts on the code golf leaderboard.
Roughly 3 years ago, I put some effort to work on code golf puzzles. I spent time one stack overflow for code golf tricks, read some topics on this forum, and with some efforts and quite a few hours spent I managed to climb up to 3rd place with ~25k points, behind @Eric_EGU which was leading with an almost perfect 27k score, and @Regulus136, me and @5DN1L behind.
I moved on, as I had less time to try to improve my score. Recently I came back to see the new leaderboard, and it seems that the battle is really raging since the last few weeks. After reaching 27k for a few days, @Eric_EGU lost the first place to @cedricdd, and more and more top players from global leaderboard (@Zylo , @Pduhard @Waffle3z @Wazemo) . I was also surprised that one some challenges, some players took the first place overnight with really impressive scores in ALL the languages.
So, just as an experiment, I took one problem where I lost all my first places yesterday (Mayan calculation), and asked AI to improve the solution. Starting from 271 chars in Perl , a simple ChatGPT prompt “reduce this code” manage to make it down to 214 caracters. But that was not enough to regain the first place. So I used my work Claude subscription, Opus 5.5 with max reasoning, and fed it with the 214 chars from chatGPT. It spawned few agents, quickly reached its token quota, but it managed to shrink to 180 chars, which validates all test cases, allowed me to get the 1st place in Perl (for now). But now, I cannot explain how this code works given the insane pile of tricks it contains. Of course I could ask Claude to explain, but since I’ve not done it myself, I’d probably forget and couldn’t re-explain it tomorrow.
So I’m left wondering, is this the new standard from now on? Just ask AI, burn some tokens, and copy-paste the output without being able to explain? Where’s the fun in that? I know there already was some concerns regarding “wrapping” the code from one language to another, and reversing/hard-coding the solutions (which cannot apply on Mayan calculation given the large outputs), but with AI this reaches a whole new level.
Also, some may have noticed that we previously we also had some informal “non-aggression pact” between most of the top players, where one would pad their solution with comments to make a submission equal to the leader, and avoid taking the first place for nothing (in, case of tie, 2 or more players can get the 200 points / language each). I think this is kind of a prisonner dilemma, of course one can be tempted to take all the first places to get 200 while reducing opponent scores, but in the end this will make the general rankings even more unreachable for “casual” code-golfers (not using AI).
To conclude, I’m not whining about my lost 3rd place, I’m just worried that with AI, the challenge will solwly evolve in a token war, and I’m not sure I still find some fun in this. Of course, no one can prohibit anyone to use AI, but maybe just leave a chance to those non using AI to still have fun and some 1st places.
Let me know what you think!
but I doubt that would be enforceable in any way. I guess we could only stop caring and keep coding, or join the dark side. In my experience the most of the “bad” solutions are where I let AI get too involved (the free one, though), but now the subscription ones may already be just “too good” not to rely on it 