Sounds quite easy to come up with such tasks. E.g. prove collatz conjecture. or write an O(log N) worst case sort. So there must be some boundaries.
The pay clearly isnt by hour so how is it worked out?
Tasks must be coding problems to complete in a set of open-source repositories. We have eight repositories now and will add more in the future. You can source tasks by identifying merged commits into these repos, and making the following: - A well-defined task description (what should the AI do?) - A golden solution (can be sourced by the implementation in the merge commit) - A test patch (can be sourced from the merge commit, this is a testing suite that verifies whether the AI's solution is correct).
If you can make a task that is hard enough for our AI, you will get paid a fixed amount. That's it!
Once you are approved onto the site, there will be a more detailed tech spec + payment info + a tutorial.
This seems like trying to get free code contributions with a weird sort of gambling mechanic attached.
We also definitely do not require you to write new code into the repo. As I said, you can adapt existing merged commits into coding tasks.
For example I have a bunch of closed PRs on my Vibe-Prolog project that took multiple attempts to get right: https://github.com/nlothian/Vibe-Prolog/pulls?q=is%3Apr+is%3...
As a specific example: https://github.com/nlothian/Vibe-Prolog/pull/214 which implements https://github.com/nlothian/Vibe-Prolog/issues/204
I'd be very interested if it is eligible!
You can't know for sure what is going to be the right difficulty in advance; however, you can definitely develop intuition for it as you make more tasks and better understand the model's strengths and weaknesses!