Which Increases Productivity More : The Advent of Personal Computer or a Large-Language Model?
“Our analysis indicates that approximately 19% of jobs have at least 50% of their tasks exposed to GPTs when considering both current model capabilities and anticipated GPT-powered software.”
That’s the conclusion from OpenAI’s recent paper “GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models.”
How much might US GDP grow assuming large-language models enable US workers to do more?
The BEA estimates US GDP is $26.2t. If we assume the OpenAI study implies workers spend 50% less time on the tasks impacted by large-language models, the US should benefit from a 4% GDP improvement which is about $1.05t.
Taxed at the US corporate income tax rate of 21%, LLMs would produce about $220b in Federal government income or about 15% of the projected 2023 deficit of $1.4t.
The Congressional Budget Office projects the US economy will to grow 3.1% this year so if this study is directionally correct, LLMs’ contribution would more than double the GDP growth rate.
This surge could parallel the personal computer’s doubling of the US labor productivity rate from 2005-2015 when computers penetrated most business operations which correlates to Moore’s Law according to a research report by the Federal Reserve Bank of New York.
None of these calculations take into account changes in staffing or personal income tax collection on gains in productivity or new jobs/industries that might grow or shrink.
Second-order effects like these are challenging to accurately predict - like monarchs & monsoons. But given the pace of Microsoft, Salesforce, & Adobe launching LLM-enabled products, the speed with which these productivity gains accrue to GDP may be rapid.
Within the paper, the authors reveal the professions in which 100% of the work will be impacted by LLMs : mathematicians, tax preparation, financial analysts, writers, & web designers. Insurance appraisers, financial managers, & search marketing strategists will see less than 15% of their work impacted by AI.
What do you think? Will large-language models produce greater productivity gains than the personal computer?