Not sure task review deserves to be set aside that fast. Two of the chains here start in the same place: a model gets an assignment that can't be finished, and a system trained to keep pushing goes looking for another door. Checking that a task is solvable before it's assigned is the one control on this list that doesn't depend on the model holding the right belief about where it is.
Not sure task review deserves to be set aside that fast. Two of the chains here start in the same place: a model gets an assignment that can't be finished, and a system trained to keep pushing goes looking for another door. Checking that a task is solvable before it's assigned is the one control on this list that doesn't depend on the model holding the right belief about where it is.