retinotopy

When a paper avoids testing its own model's spatial limits, we did

TRIBE v2 claims to predict brain activity from video, audio, and text, but its own paper never checked whether it reproduces the spatial maps that define sensory cortex. Jakub Strawa ran the classic experiments, retinotopy, tonotopy, motion selectivity, through the model. Retinotopy and tonotopy came back null. Motion selectivity, however, landed precisely in MT+ with extraordinary enrichment. The eccentricity signal turned out to be a confound from image statistics, not real position coding. This is the kind of rigorous stress test every brain model needs.