Researchers found that six generative AI large language models outperformed real people in emotional intelligence tests, scoring an average of 81% compared to humans' 56%. The AI models were able to generate new assessments and tests that were as reliable as the original ones, which took years to develop.