For a bit of background, I've always found these types of problems interesting. The really difficult optimization problems where techniques like gradient decent don't work, or where even finding a feasible solution is difficult (e.g. square packing). I've worked on these independently for fun and have created this project as a way to share the problems I find interesting with the community. I am curious to see how far others can push these.
Each challenge is defined by its verifier, a program that will enforce the constraints, and evaluate the objective function. The verifiers are open-source per challenge, so users can check their solutions against it before submitting. The platform keeps a record of submissions and will update the leaderboard each time someone finds a new record. Because the verifier is the spec, any loopholes in its implementation are open game to exploit to reach the top of the leaderboard.
So far I've shipped two challenges: one for implementing the standard normal CDF in Solidity where the objective is to find the implementation with the smallest error, the second is to pack a 351,049-word dictionary of words into the smallest crossword possible.
These are just the first few ideas I've had with lots more coming soon. The next one might on finding an optimal algorithm (think matrix multiplication) which is going to be a bit tricky to design the verifier for.
Any feedback on this would be amazing to hear, especially around the verifier model, whether these types of challenges seem interesting, or if anything about how the platform works seems unclear.
Show HN: Argmin, challenges for optimizing hard problems
argmin.dev