APIO 2025
I just returned to Singapore last Wednesday after spending an entire month overseas. When I returned to Singapore, I attended the Explore HRT event (seems like a very cool place to work) then immiedately went to NUS to attend the NOI prize giving ceremony and APIO.
Jane Street was sponsoring NOI and I saw that there was an estimathon scheduled. Since I was returning from Explore HRT, I thought it would be very funny to wear a HRT Shirt to the event.
I ended up winning the estimathon with my team. The scoreboard is below. We were team 4.
Out of the top of my head, I remembered:
- Question 2: How many words are there in the Harry Potter books in total?
- Question 3: What was the world box office gross of the highest grossing film of 2024?
- Question 4: How many sheep are there in New Zealand?
- Question 5: How many citations does the most cited paper have?
- Question 6: How many 250ml cans of red bulls do you need to drink to exceed FDA limit?
- Question 10: How many places are called San Jose (the most popular place name in the world)?
So I did an extremely stupid thing for question 3 where I did not read the word 2024. And apparently do did my team members. Because I told them endgame was about 3B and the highest grossing all time was avatar. So we guessed $[2B,5B]$ intially or something. Then it was wrong. I was quite adamant that endgame was about 3B. So I guessed $[5B,50B)$ and it became wrong. So I thought ok, maybe its slightly lower, but there no way it is $<1B$. So I guessed $[\frac{5}{3}B,5B)$. But I was too lazy to write $1.667B$ and wrote $1.7B$ as my lower bound instead. It turns out that the actual value is $1.698B$ which is Inside-Out 2. BRUH. So after that got wrong. I thought maybe avatar was super popular and just guessed $[50B,500B)$ which came out wrong again.
Then I thought there is no way a movie has a higher box office than the entire valuation of Maersk, maybe the graders made a mistake. So I went down to the grader’s table and told them we guessed everything between $[1B,500B)$ and they are all wrong. Please check (because last year they messed up $1$ OOM, so I thought they did that again).
It turns out I social engineered the answer because he thought they made a mistake when I said $[1B,500B)$ when we actually only guessed $[1.7B,500B)$. Lol, so we just guess $[1B,2B)$ for free $1$ point. After using $5$ attempts…. I actually only realized I did not see 2024 after the estimaton…
For question 5, I was actually very lucky, because I knew Adam optimizer paper had about 200k citations. And I thought that was probably one of the highest citations. So I guessed $[100K, 1M)$ and it seems most people highly underestimated it. Turns out the stupid “transformers is all you need” paper also has around 200k which is maybe expected too? But it turns out the paper with the actual most number of citations is “Protein measurement with the folin phenol reagent” by Oliver Lowry and colleagues, published in 1951. But I expect Adam optimizer paper to have more citations soon since it is kind of obligatory to cite the adam optimizer paper.
On the topic of citations, I had this conversations with a ex-acedemic in HRT about why he made the career switch. From what I understand, he felt that in acedemia, they were not really contributing to science in the way that they imagined, but were just optimizing for the citation count. They could not focus on writing difficult papers as it was more viable to write easy papers that were related to the current trendy topics in the hopes that others would cite them.
Perhaps until a few months ago, I still felt that becoming a acedemic was a noble and interesting profession that could be my “ikigai” or whatever. Where as people in trading (or generally industry) were those that “sold out” to money. I had applied to Trinity to study Mathematics and was seriously trying to pursue this path. I was kind of fascinated with this path after reading A Mathematician’s Apology and knew that I wanted a career where I could just solve hard problems all day, since I liked competitive programming so much.
But there were like too many signs so far that I really should not consider an acedemic career path. For example:
- O4-mini getting an accuracy of 17% on frontier math private set. By the time I get my PhD (which earliest would be in $8$ years I think?), what would become of the state of math research?
- I got rejected by Trinity mathematics and was pooled into St Edmunds. In the world of acedemia where the dropout rates in each step of “phd -> postdoc -> assistant prof -> tenure track” was really high, I felt that if I did not get into one of the best programs, it was basically over already. I think everyone who wants to get into mathematics research dreams of solving the hardest and most attractive problems. But realistically, there is probably really only very few interesting problems. So I am not sure what everyone else is researching?
- I spoke to multiple math phd students, they kind of all wanted to leave acedemia too. I also recently watched this youtube video about why the author left acedemia. Actually I never spoke to Lim Jeck before, I wonder what his opinions on acedemia is now?
Btw, I authored problem 2 for APIO 2025 this year, permutation game.
You can see my original pdf submission for the problem here. You can observe there were way more subtasks than the final version. The SC actually asked me how to allocate points to the subtask and I suggested cutting down the subtasks to these constraints. Generally, not having “no additional constraints” for such case-worky OI problems make sense, because then contestants will not need to “stitch” their solutions for each case.
According the perliminary scoreboard, no one got AC on this problem. No one got AC on this problem. And only a single person got positive score on the final subtask where $e=m$. Maybe I should have split the final subtask into the even and odd case?
I was a bit dissapointed that in the final version, the SC decided that no points should be given to contestants who were able to calculate the optimal score. I felt that was an important part of contest strategy, to be able to verify whether you were able to calculate the optimal score correctly.
Another thing that could be improved was to give some token score to contestants who have a valid construct but used way more moves than $12n$. I should probably have mentioned that to SC.
Anyways, I hope you enjoyed the problem.
Btw, for the story of this problem. Last year, there was a Codeforces round that needed a d1C urgently. KAN asked me if I had a spare d1C lying around and I came out with permutation game. But the subtask where $e=m=3$. This can be more naturally stated as: Alice chooses $3$ integers, Bob can swap any $2$. And surprisingly, the result was extremely surprising!
Unfortunately, that problem ended up being much more difficult than a d1C should be. I decided to try and generalize the problem with Si Jie where $G$ could be any graph. After a few weeks of work, we were able to solve it for $G$ being any connected graph! We tried thinking about the case where $G$ could be unconnected but did not make much progress.
Since I would be going to IOI as the coach of Singapore team, we decided to send the problem to EGOI because I wanted a free trip to Bonn.
But actually EGOI rejected the problem. Their feedback was like “The description was hard to follow in some places. For example it takes place on a graph, but it quickly becomes clear that the structure of the graph is not very important. We think it is not an ideal EGOI problem as there is quite heavy casework involved and the description of the game is quite involved, making it not very accessible for weaker participants, which we aim for even in harder problems. Therefore, we decided to not use this problem for EGOI.”
I felt kind of offended by the reply from EGOI. Yea, I guess the problem has a hard-to-read statement, but the underlying idea seems quite natural once you parse it beyond the surface level. But like “The description was hard to follow in some places.” and “For example it takes place on a graph, but it quickly becomes clear that the structure of the graph is not very important.” is a total non-sequitor. It felt like they wrote some nonsense to justify why they rejected the problem. I would totally be fine if the feedback was just “Your problem is too hard” (fair) or “There are better problems” (can’t argue). But this reply feels extremely offensive for some reason because it was so illogical.
Anyways after the problem was rejected, I kind of realized that the problem was insanely hard, so decided to send it to APIO to test the difficulty against Chinese contestants. Amazingly, no one got AC on it!