There is valid concern on how predictive the implicit bias test is in non-test conditions. But "implicit bias" is only the start of the conversation - it is one easy thing to measure in a sea of difficult to measure issues. If you want to jump into this rabbit hole, I have found the researches that work in this field to be eager to describe their more up-to-date work on addressing them (I do try to play a devil's advocate in such conversations and have gained much respect for their rigor). Regrettably, as usual in academia, the easies way to be exposed to those conversations is not that easy: going to talks given by those researchers.