Researchers at MIT’s Computer Science and Artificial Intelligence Laboratory tested current large language models against thousands of real-world labor tasks and found that AI “workers” succeed only ...