U.S. Occupations Treemap

One emerging use for large language models is turning unstructured qualitative data into structured categories: customer-support emails sorted into product areas, survey free-text bucketed into themes, medical notes coded to diagnoses. The classification can also run on more bespoke texts — including four-hundred-year-old ones.

In one chapter of Pantagruel, the 16th-century satirist François Rabelais sends his character Epistemon on a tour of the underworld, where he finds that the damned — kings, poets, philosophers — have been reassigned to humble trades. Alexander the Great patches old breeches; Cleopatra sells onions. Rabelais uses these assignments to puncture worldly status, but the list also happens to be a proto-occupational dataset: every figure gets a job.

I asked an LLM to read each entry and assign it a modern Standard Occupational Classification (SOC) code — the same taxonomy the U.S. Bureau of Labor Statistics uses to track wages, workplace injuries, and employment projections. The treemap below shows where Rabelais's damned land in the present-day labor force, and the sidebar panels surface the BLS data for each occupation: wage distributions, interest profiles (RIASEC), employment type, and day-to-day work context.