#brain test

Researchers gave top AI models a classic attention test used in psychology and found a major flaw. While the models could correctly name colors in short lists, their performance deteriorated sharply as the task became longer and more complex. Some leading systems fell from over 90% accuracy to nearly complete failure.