Isn't sudoku bit problematic problem for GA as there is just one correct solution and the rest are just wrong, compared to a problem that has solutions that are better or worse, but potentially no solution is absolutely correct?
A fair point (though technically sudoku can have multiple solutions, they are all equally correct; there is no range of correctness, it's just binary). However, I think something like this is useful as a proof of concept for a GA system, because it's easy to judge its solutions. Many textbook examples of GA I've seen involve evolving a best fit line for a given line, and the simple examples use functions like y = x^2+1 because its easy to judge completion.