If AlphaFold really is brute force on known protein problem space, would it then be usable as a model for novel proteins?