OpenAI's 'Project Stagecraft'
OpenAI is reportedly paying thousands of specialists to simulate and demonstrate tasks across 400+ professions to create highly contextual training data—what the coverage calls 'Project Stagecraft.' This shift toward domain‑specific task simulations means labs are valuing researchers who can formalize real workflows, not just build general models. (businessinsider.com) (startupfortune.com)
Reported contractor pay ranged from at least $50 per hour for general contributors to as much as $500 per hour for domain experts, according to documents reviewed by reporters. (startupfortune.com) A 439‑row spreadsheet obtained by journalists enumerated specific job titles contractors were asked to model, including commercial pilots, emergency‑medicine physicians, sculptors, pharmacists, and agricultural managers. (startupfortune.com) Handshake AI has moved to deepen its data‑quality stack via an acquisiton of Cleanlab announced January 28, 2026, an acqui‑hire that folded nine Cleanlab researchers — including co‑founders Curtis Northcutt, Jonas Mueller, and Anish Athalye — into Handshake’s research team. (techcrunch.com) Contractor briefs required development of professional personas and prompt templates that replicate real decision‑making in a role, signaling effort to convert tacit occupational judgments into structured training tasks and evaluation checks. (startupfortune.com) OpenAI’s public hiring pages and recruiting practice guides show active hiring across research, engineering, and “emerging talent” pipelines, while interview resources emphasize technical breadth and building agent/tooling systems — a combination that favors candidates able to operationalize domain workflows in code and data pipelines in addition to strong ML theory. (openai.com) Reporting estimates the Stagecraft program’s contractor spend is likely in the millions given published hourly rates and the documented headcount, and media coverage framed the outlay as a material capital allocation toward embedding occupational depth in models. (startupfortune.com)
