A new study shows that large language models like GPT-4 can match humans in detecting sentiment, political leaning, emotional intensity and sarcasm in text. While AI struggles with sarcasm, its speed and consistency make it promising for journalism, research and public health, though questions about reliability and bias remain.