One of the major discoveries in attitude semantics over the last thirty years has been the fact that certain types of attitude reports require interpretation de se. This finding has prompted a move among semanticists to treat attitude verbs as uniformly quantifying over centered worlds (typically modeled as triples of worlds, individuals, and times), rather than merely over possible worlds, and likewise a move to treat attitude complements as uniformly denoting sets of centered worlds, rather than mere sets of possible worlds. Thus, for instance, "A believes P" is true iff P holds of all triples <x,t,w> such that A believes that she might be x in w at t. Proponents of a Uniformity Thesis of this type include Schlenker (1999), Ogihara (1999), von Stechow (2003), Anand (2006), Pearson (2015), and Grønn and von Stechow (2010). In this talk I present evidence against the Uniformity Thesis, drawing from my fieldwork on Nez Perce (Sahaptian). I show that dedicated de se devices (shifty 1st person indexicals, relative tenses) are possible in one type of attitude report in Nez Perce, but not in another type, and argue that the difference between the two types of attitude report crucially reflects the semantics of the attitude verb and its complement. I argue in particular that some attitude verbs quantify over centered tuples, making it possible to include dedicated de se devices, whereas others quantify merely over possible worlds, ruling such devices out. [Note that a paper version of this work is available for download on my website.]