The first is a multiple-choice exam. Basically you categorize which AI answer is better based on some categories. If you want to prepare, for example, look at what Outlier tasks look like, it is very similar. Second is a proctored written assessment where you pick a topic (question/prompt) of your choice and answer it like a good AI would, you have 60 minutes for it and you have to provide good sources. Finally, there is an interview. Cannot say much about this.