Memorandum
- From
- Delaney Sawyer via Fast Company
- Date
- Filed
- Business·4 min to read
- Re
AI Training Startups Pay Experts Up to $125 an Hour to Automate Their Own Jobs
ReAI Training Startups Pay Experts Up to $125 an Hour to Automate Their Own Jobs
A new white-collar gig economy is emerging as AI labs pay domain experts to train the models that may eventually replace them. Mercor alone pays out up to $5 million daily to more than 100,000 contractors, with demand far outstripping supply.
A fast-growing white-collar gig workforce is being built around a single, uncomfortable bargain: highly educated professionals are being paid handsomely to train the artificial intelligence systems that may one day make their own expertise obsolete. Startups like Mercor, Scale AI, Turing, and Ethos are recruiting doctors, lawyers, bankers, research mathematicians, and game theorists to probe the most advanced AI models, find where they fail, and teach them to perform better.
Mercor, founded three years ago in San Francisco, has become one of the most prominent middlemen in this space. The company pays out as much as $5 million a day to experts on its books, according to CEO Brendan Foody, with average pay around $125 an hour—roughly the same as a physician and almost twice the average for U.S. executives. More than 100,000 people are now on Mercor’s roster, including software engineers, scientists, architects, and cinematographers. Even that supply of world experts is not enough: Foody says demand from the AI labs Mercor works with runs at three or four times what the company can supply.
The work itself varies widely. Lazaros-Antonios Chatzilazarou, a PhD student in game theory who also works at EY and teaches at the London School of Economics, was recruited by a Mercor recruiter over LinkedIn during a late-night video game session. Within weeks he was developing prompts to test how AI models apply game theory and grading their responses. He describes the pay as «a very competitive salary» and says the workload can amount to a full-time job. Academic curiosity and a desire to be involved in a world-changing technology drew him to the role, a sentiment echoed by other professionals doing similar work.
The broader industry is expanding quickly. Scale AI runs Outlier, a contractor network whose website invites workers to «become the expert that AI learns from.» Turing has listings for video content creators to produce short «walk-and-talk» clips in public places, speaking naturally about parks and landmarks to help AI systems better understand the world. Ethos, a London startup, raised $22.75 million from Andreessen Horowitz in May to build an expert network with voice-based onboarding, framing its mission around the idea that AI labs are «pointing a giant capital gun at every economically valuable occupation in the world.» Wirestock, which once helped photographers license stock images, pivoted to AI data supply in 2023 and raised $23 million in May from investors including Sheryl Sandberg’s venture capital fund. It now claims more than 700,000 creators. Even home cleaning is being drawn in: a startup called Shift offers New Yorkers free cleaning if they allow cleaners to wear mounted cameras that generate training data for embodied AI systems.
The irony is not lost on observers. The International Monetary Fund has said generative AI will touch some 40% of roles worldwide, and as many as 60% in developed countries, including many white-collar positions. Every problem experts hand to a model and every blind spot they help it close makes the machine a little more capable of doing without them. The expertise that took a career to acquire becomes easier to hire by the hour—or eventually not to hire at all. For now, the money is good and the work is intellectually engaging. But the professionals training these systems may be accelerating the arrival of a future in which their own skills are no longer needed.
2
