The "Diamond Sponsors" of the event are Huawei and OpenAI. I found the welcome note from OpenAI [1] quite curious:
"[...] Eventually, AI will be able to solve even the hardest contest problems that we’ve seen yet. It will work alongside us and help drive the discovery of new knowledge. What you take from this week - the sense of being stuck, the thrill of progress, and the practice of building together - will remain critical as you shape your community and the future you build. [...]"
By Chief Scientist of OpenAI, Jakub Pachocki. Who happens to be an incredibly accomplished Competitive Programmer (2nd in ICPC World Finals, Winner of Code Jam, 2nd in Hacker Cup).
That curious statement comes across as inappropriate. These talented students are among the best in the world. Yet, openai chose to make it about and praise themselves. That was not classy.
eunos 2 hours ago [-]
Lol and the Huawei note wrote otherwise. AI still hallucinates a lot so pursue mastery over algorithms and data structures to improve the technology
Mr-Frog 36 minutes ago [-]
Chinese century incoming.
brcmthrowaway 33 minutes ago [-]
How many chief scientists (aka career climbers) does OpenAI have at this point?
toonewbie 4 hours ago [-]
I was onsite today watching the contest live, and great atmosphere all around. One surprising outcome: the team in 17th place solved the same number of problems as the team that won gold in 4th place. Hopefully that isn't too demotivating to any team and we can see better separation in the future. After all, it can only mean that the problemsetters underestimated the contestants ;)
Congratulations to all the teams!
tshaddox 4 hours ago [-]
> the problemsetters underestimated the contestants
Except for problem C, which was only submitted by 4 teams, all unsuccessfully.
cperciva 4 hours ago [-]
I don't blame them. That problem statement seems to be deliberately confusing.
tshaddox 3 hours ago [-]
Yes, I meant to imply that the problemsetters are to blame.
hatthew 3 hours ago [-]
I doesn't seem that unclear to me? I absolutely don't have the skill to solve it, but it took <2 minutes to understand the problem and goal.
orlp 1 hours ago [-]
It doesn't seem that hard to solve to me either. It's solvable with basic linear programming.
1. Add a variable for each node, and a variable for each output edge from stations.
2. For each reservoir add equality constraints to the sum of incoming edges with the coefficients given in the problem.
3. For each station add equality constraints between the weighted sum of its inputs (which is 1 for the root station) and its outputs (which are the variables we added).
4. Add an out_edge >= 0 constraint for each output edge on stations to forbid illegal negative flows.
5. Add a variable m which is constrained to be less than all the output station variables.
6. Maximize m.
amluto 51 minutes ago [-]
This happens every now and then. Basic LP isn’t conceptually so bad, but the ICPC environment doesn’t come with CPLEX or Gurobi, let alone a less fancy open source tool. And there is definitely not a copy of cvxopt around to make this easy. Even if you want to quickly kludge up an interior point solver, you don’t have a linear algebra package available (sorry, no numpy or BLAS).
What you can do is to submit a 25 page PDF that the organizers will print and stick on your desk for the competition. And you could put a careful implementation of a very basic simplex solver using dense matrices that is optimized for ease of transcription, taking up, say, half a page. You would hope not to use it because it’s absurd, but then if this problem C shows up, the fastest typist on the team can type it in verbatim.
If I, personally, did this and won the contest as a result, I would feel slightly bad. In my opinion, the contest organizers should either provide an LP solver or refrain from giving obvious LP problems like this.
Obviously OpenAI could kick everyone’s butt by typing faster than any human and by effectively having a large memorized library of pre-written code. Honestly, LLMs vs humans in the ICPC feels a bit like IBM’s old Jeopardy stunt where the machine had a huge advantage in its ability to push the button.
I'd just like to clarify that I'm not saying this is necessarily the solution the problem writers were looking for, or that it will run within the allocated time. Just that it's a feasible solution.
The headers A-L are links to the problem PDFs if you want to see.
cptroot 3 hours ago [-]
Does anyone know how long it will be until the input/output data packet is made available? I'm interested in taking a crack at some of these but know I always miss an edge case.
3 hours ago [-]
coolThingsFirst 1 hours ago [-]
Great memories of ICPC. Their problem statements are so creative it’s insane.
Nothing like CF garbage, solving imaginary patterns that no one cares about.
akamaka 2 hours ago [-]
In the TV show Silicon Valley, there’s a joke that Nelson “Big Head” Bighetti, the perennial underperformer, did his undergrad at ASU. But I guess that’s one thing the show got wrong, because Arizona State University finished among the top three American schools in the world finals.
riku_iki 2 hours ago [-]
I expected fast response from AI labs about what kind of medals top models could win.
3 hours ago [-]
Rendered at 00:54:35 GMT+0000 (Coordinated Universal Time) with Vercel.
"[...] Eventually, AI will be able to solve even the hardest contest problems that we’ve seen yet. It will work alongside us and help drive the discovery of new knowledge. What you take from this week - the sense of being stuck, the thrill of progress, and the practice of building together - will remain critical as you shape your community and the future you build. [...]"
By Chief Scientist of OpenAI, Jakub Pachocki. Who happens to be an incredibly accomplished Competitive Programmer (2nd in ICPC World Finals, Winner of Code Jam, 2nd in Hacker Cup).
[1] https://icpc.global/community/history/brochures/world-finals...
Congratulations to all the teams!
Except for problem C, which was only submitted by 4 teams, all unsuccessfully.
What you can do is to submit a 25 page PDF that the organizers will print and stick on your desk for the competition. And you could put a careful implementation of a very basic simplex solver using dense matrices that is optimized for ease of transcription, taking up, say, half a page. You would hope not to use it because it’s absurd, but then if this problem C shows up, the fastest typist on the team can type it in verbatim.
If I, personally, did this and won the contest as a result, I would feel slightly bad. In my opinion, the contest organizers should either provide an LP solver or refrain from giving obvious LP problems like this.
Obviously OpenAI could kick everyone’s butt by typing faster than any human and by effectively having a large memorized library of pre-written code. Honestly, LLMs vs humans in the ICPC feels a bit like IBM’s old Jeopardy stunt where the machine had a huge advantage in its ability to push the button.
I'd just like to clarify that I'm not saying this is necessarily the solution the problem writers were looking for, or that it will run within the allocated time. Just that it's a feasible solution.
Nothing like CF garbage, solving imaginary patterns that no one cares about.