A doctor gets a LinkedIn message at midnight. A lawyer takes an online test. A math professor is asked to invent problems nobody has solved yet. All three end up doing the same side job: training AI models by finding where they fail, then correcting them.
The company behind many of these messages is Mercor, founded three years ago by three college dropouts who were still in their early twenties. It just closed a 350 million dollar funding round that values the company at 10 billion dollars, with Felicis leading the round and Benchmark and General Catalyst joining in. Mercor now pays out as much as 5 million dollars a day to the roughly 100,000 professionals on its books, with average pay around 125 dollars an hour, close to what a working physician earns in the United States.
This did not happen by accident. For years, training AI meant paying low wage contractors in Africa and Asia to draw boxes around stop signs or label photos of cats and dogs, through companies like Scale AI. That changed in June when Meta invested 14 billion dollars into Scale AI and hired its founder, which pushed rivals like OpenAI and Google to pull their business from Scale over conflict of interest concerns. That single deal sent a wave of work toward competitors, and Mercor and a rival called Surge AI, now reportedly valued between 15 and 25 billion dollars, picked up the overflow.
The work itself has changed too. Instead of labeling images, the new gig economy asks real professionals to stress test models in their own field: a physician builds a tricky medical case to see if the AI gives the wrong advice, a mathematician invents a problem no model can solve yet, a lawyer checks whether a contract clause was read correctly. It pays well because the skill is rare and the demand from AI labs is bigger than the supply of experts willing to do it.
The uncomfortable part is obvious once you say it out loud. Every mistake an expert catches becomes a mistake the model will not make again. The same people correcting the AI are the ones whose expertise becomes less scarce every time they do it. The International Monetary Fund has estimated that close to 40 percent of jobs worldwide are exposed to generative AI, rising to 60 percent in richer countries, and white collar fields like law, medicine, and finance sit near the top of that list.
Not everyone doing this work feels it. Some say the pay is good enough and the window is long enough that it does not matter. Others treat it as a controlled way to understand what is coming before it arrives, which is not nothing.
What is clear is who benefits regardless of how this plays out. The founders running these training platforms collect a fee on every hour billed, whether the expert's job is still around in five years or not. For professionals weighing whether to take the work, the honest framing is that it is good money today, built on a clock that the worker does not control.