#visual representations

'Study finds DINOv3 models trained on billions of natural images align with human cortical responses across space and time, driven by model size, training amount, and image type'