How did Anthropic measure AI’s “theoretical capabilities” in the job market?

This month, a striking graphic has emerged from an Anthropic report, sparking discussions about AI's influence on the job market. The illustration compares the current "observed exposure" of various occupations to large language models (LLMs) with their "theoretical capability" across 22 job sectors. The red area of the graphic depicts the present engagement of these jobs with LLMs, while the blue section reveals a startling prediction: LLMs could potentially handle over 80% of tasks in a wide range of professions, including fields like Arts & Media, Office Administration, Legal, Business & Finance, and Management. This projection suggests that LLMs may eventually take on a significant share of the workforce. However, a closer examination of these theoretical capabilities paints a more nuanced picture. The blue section, while impressive, relies on outdated and speculative estimates regarding AI's potential to enhance human productivity, rather than a definitive forecast of job displacement. Anthropic's baseline for these theoretical capabilities does not stem from its own model evaluations or measurable forecasts of performance improvements. Instead, it references a report from August 2023 titled "GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models," co-authored by experts from OpenAI, OpenResearch, and the University of Pennsylvania. This context is essential for understanding the complexities of AI's anticipated role in the workforce.