World-class competitive programmers will compete against AI in a heuristic programming contest on July 16, 2025 in Tokyo. The match will be part of the AtCoder Heuristic division finals (9:00-19:00 JST).
Key details:
Contest type: Heuristic optimization (not algorithmic)
Participants: Top 12 human programmers vs AI system(s)
Sponsor: OpenAI
Note: Organizers emphasize this is an "assessment" rather than pure competition, though concrete scores will be produced
See this https://prtimes.jp/main/html/rd/p/000000059.000028415.html for
more details.
Rules are not very clear for now but I expect results to be very clear. However, if they will not be clear, it will be resolved to my best judgement (e.g. may resolve N/A)
Update 2025-07-04 (PST) (AI summary of creator comment): In the case of a tie, the market will resolve to N/A.
Update 2025-07-16 (PST) (AI summary of creator comment): In response to concerns about the timing of the official results, the creator has stated:
The market's close date has been extended by one week.
The market will be resolved as soon as the final results are available.
1,000🏅 Top traders
| # | Trader | Total profit |
|---|---|---|
| 1 | Ṁ161 | |
| 2 | Ṁ118 | |
| 3 | Ṁ82 | |
| 4 | Ṁ60 | |
| 5 | Ṁ55 |
People are also trading
@jessald I moved market closing date a week away but I will aim to resolve as soon as results are in
More context:
Seems like OpenAI has some kind of model or system that is tuned for tasks of this kind (heuristics). Source: this came to me in a dream
The best currently existing AI (https://sakana.ai/ale-bench/) took 21th (top 2%) place in last public competition of this kind
@CraigDemel good question. I have not considered this because in human competitions there are no ties (time is tiebreaker) but in this case there might be a tie. I think N/A probably makes sense as a "neutral" outcome in the case of tie