In practice, what I've seen too often is that code that was duplicated but should have been factored out ends up evolving with people having forgotten that there were duplicates. Bug fixes that should have been implemented in all instances of the duplication only end up in a few of them. When you do go to factor that duplication out, you're stuck with 5 to 10 different versions that all have different bug fixes, and theoretically they all need all of them.

On the other hand, if someone finds that one instance of the duplicated code really is different, they either turn it into a new function (the right answer) or add a new argument. Even if they take the wrong path by adding a new argument, it's easier to turn that one function into two functions than merge 5 to 10 different versions of what should have been the same code.

The problem typically comes down the road when those abstractions touch so many pieces of the code that they become extremely expensive to change. The problem isn’t the hasty abstraction, the problem is the lock-in it creates.