The AI industry is currently being floated by an unprecedented level of venture capital. And while the leading labs, like OpenAI and Anthropic, are starting to see revenue figures in the hundreds of millions, it pales in comparison to the money they need to make to be profitable. It’s possible AI companies will get there, but in the meantime, there are a handful of oddly specific companies that are benefiting the most from the AI investment boom: AI data annotation companies. Chances are you’ve never heard of Mercor, which has become the fastest growing company of all time. It supplies data and workers to OpenAI, Anthropic, and the like, at an estimated cost of $10 billion a year. But Mercor and its rival, Handshake, are arguably the only businesses making real money from AI.
We’re submitting a pair of complementary stories about Mercor by Josh Dzieza (one of which was co-reported with Hayden Field) in this category because they tell the reader about how AI works on a very mechanical level. So much writing about AI is about the technology’s potential—as a social good or as a labor destroyer—but there is a lot less out there about how it’s being developed, and on whose backs it’s being built. Dzieza draws on years of experience explaining AI in laymen’s terms. (For the most clear-eyed description of reinforced learning by human feedback—RLHF—you can refer to Dzieza’s 2024 OJA-nominated feature AI Is a Lot of Work.)
In exploring the staffing and annotation pieces of model training, Dzieza pulls back the curtain on the very human cost of AI.
