The complaint as I understand it is that if it's just supposed to be harder/more academic 'FizzBuzz', then it ought not to be something that requires revision, cramming prior to the interview (probably sufficient in itself! So what's it testing?) because it's not memorable/derivable after years of doing something different or that's built on top of abstractions on top of abstractions on top of abstractions on top of the question.
But I've never actually had an interview like it (or I have I didn't realise/object to it myself) and discussions about it tend not to be particularly concrete. For example in the spirit of 'harder/more academic FizzBuzz', I think 'given this data structure, write a function to transform it or others like it into one that looks like this' is perfectly reasonable, but say 'in big-O terms, what's the best an algorithm to invert a binary tree can be? Write it' is a bit unfair for a job writing CRUD apps.
But I've never actually had an interview like it (or I have I didn't realise/object to it myself) and discussions about it tend not to be particularly concrete. For example in the spirit of 'harder/more academic FizzBuzz', I think 'given this data structure, write a function to transform it or others like it into one that looks like this' is perfectly reasonable, but say 'in big-O terms, what's the best an algorithm to invert a binary tree can be? Write it' is a bit unfair for a job writing CRUD apps.