OpenAI is at it again. Looped transformers are no more opaque than regular ones, and should be run to whatever depth is needed. (And chains of thought ensuring transparency is fiction to begin with.)