Visual Anagrams Reveal Hidden Differences in Holistic Shape Processing Across Vision Models | Read Paper on Bytez