3 ms·
Assuming you’re in control of the test data set, you do know if a task is unsolvable. At that point you can reward the model based on how quickly they give up.
by jurgenburgen 13d ago
Assuming you’re in control of the test data set, you do know if a task is unsolvable. At that point you can reward the model based on how quickly they give up.