Arjun Romani

Generative AI: what is it good for?

Alok Jha, Tom Standage, Abhi Burtix, Arjun Romani

Following the 2017 Transformer architecture and the historic 2022 adoption of GPT-3.5, large language models now process unlabeled internet data to excel at tasks ranging from medical licensing exams to code generation. Despite these capabilities, the technology remains a "black box" with complex, opaque weights that lack the transparency required for high-stakes fact-finding or full automation. Consequently, while economists project that generative AI will impact half of the daily tasks for 20% of the U.S. workforce, current deployment is expected to evolve as a human-AI collaborative workflow rather than an immediate replacement of human roles.