It's amazing how many of the "bullshit insane" things in neuroscience make perfect damn sense from a Bayesian brain standpoint. If you're saccading to update a

It's amazing how many of the "bullshit insane" things in neuroscience make perfect damn sense from a Bayesian brain standpoint. If you're saccading to update an approximate posterior model of the visual scene, you don't need to "pause" or "blank out" the blur of the saccade. Each fixation is a sampling of the environment under generative assumptions, so yeah, your previous posterior becomes your new prior, and as long as the saccades don't take too long, your new prior is accurate enough to make decent predictions! Since what we consciously experience is the predictions themselves, we don't notice any gaps until the bottom-up prediction error signals become huge.