Research

What the job-market data shows, and how we work with it.

Our own analysis and notes. Where a piece uses a number, the number comes from a query anyone with a key can run, and the piece says which one.

  1. Every number links to the Casebook page that reproduces it, with its query and what it cost.
  2. People are counted, never named. A group of fewer than 10 is written <10.
  3. Costs are measured from the key's balance before and after, not estimated.
  1. InsightProfiles name the field, rarely the tools50% of US machine learning engineer postings ask for PyTorch. 7.3% of the 3,161 machine learning engineers in San Francisco list it. A search that filters on a tool keeps only the people who wrote it down.50%postings ask for PyTorch
  2. NotesEvery number with a receipt: how the Casebook is madeEach case in the Casebook ships the prompt that produced it, the queries behind every figure and what rerunning them costs. Here is the process, and the rules that keep it honest.15published cases