He's right, there is a fitness cost to seeing reality accurately, but there is also a benefit. the optimized solution is somewhere between disregarding reality & accurately seeing it.
In the model he presents, there are cases where agents knowing more about reality drive to extinction those that know less. Depends on the cost of information.
...will you be more competitive in natural selection if X? seems like this can depend on lots of things
yeah, almost so many things you could make models for any case you wanted.
