AI & Enterprise
ChatGPT, Claude stumble in attention test, raising questions for AGI
ChatGPT and Claude, among the latest large language models, performed worse than expected in the Stroop test, a psychology experiment used to gauge human attention and executive control, a study found. Researchers tested OpenAI\'s GPT-4o and Anthropic\'s Claude 3.5 Sonnet and saw performance fall sharply when word meaning conflicted with ink colour, with accuracy declining further as tasks increased. Follow-up tests on newer models showed limited improvement, the study said.
