2 ms·

Since the model doesn't get to observe the actual situation being reported on, such an improvement in accuracy would only be random chance and should not be rew

Since the model doesn't get to observe the actual situation being reported on, such an improvement in accuracy would only be random chance and should not be rewarded.

walkabout 1y ago
Oh, agreed, they don’t get points for failing the task but accidentally reporting something more-correct in the process.