🥇Reward Model
We created a reward model based on the bonuses paid by AI entities for labeling tasks and the number of makers and checkers assigned to each case. This model will allow users to clearly see the income they will earn for each label they complete, along with the points and token rewards they will receive.
Reward Model
Assume a labeling project has a total value of A USDT. The AI Entity assigns m Makers and n Checkers to participate in voting. The Maker with the highest score receives a USDT, the Checker with the highest score receives b USDT, and Makers and Checkers who do not receive the highest score will receive c USDT.
Constraints:
If a user's error rate across all Maker tasks in the project is <= x/m, then all Maker rewards for the user under this project will be zeroed out.
If a user's error rate across all Checker tasks in the project is <= y/n, then all Checker rewards for the user under this project will be zeroed out.
The highest Maker reward a, the highest Checker reward b, and the error reward c are based on empirical proportional relationships.
Ultimately, this model estimates the rewards a user will receive for each Maker and Checker task, while also penalizing users who deliberately make mistakes. Even if a user completes very few tasks, they can still earn rewards.
Last updated