1 hour ago
3
Lately a new benignant of transportation has recovered its mode into my inbox. “Not definite if you've heard,” writes idiosyncratic named Nick, “but AI labs specified arsenic Anthropic and OpenAI are paying media companies specified arsenic Wired $350K to $1.8M arsenic a 1 clip interest for a transcript of their aged enactment records (documents, emails, slack messages, tickets, etc.) for grooming amended models.”
Another email with a akin transportation (adding telephone transcripts to the database of desired information) raised the stakes to a imaginable $2 million. “Public information is exhausted, and the labs are spending implicit $45B connected backstage information to physique amended models,” writes idiosyncratic named Cora.
While I was alert that AI companies are buying adept information from assorted sources, determination hasn’t been that overmuch attraction paid to it. But erstwhile I looked into it, I discovered that this concern is mode bigger than I suspected—and according to its practitioners, it could person arsenic overmuch interaction connected workers and nine arsenic the information centers that are dominating the quality these days.
The companies that contacted maine are portion of an manufacture roar successful information acquisition that’s been softly buying the workflows of sizable companies and is inactive truthful bare that it’s bleeding into our inboxes. A freelance writer precocious wrote astir her acquisition with a institution called Handshake, which offered her $1,000 for her enactment connected a azygous story. (She didn’t sell.) A typical from Handshake confirmed the connection to me: “We licence nonrecreational documents crossed a batch of fields truthful the labs tin trial and bid however models grip existent cognition work,” helium said. “Journalism is 1 of those fields.” Handshake tells maine that a large portion of its concern is buying postgraduate students’ papers and coursework.
An entrepreneur person of excavation tells maine he’s been getting message from aggregate companies whose concern is “helping startups upwind down their companies," arsenic 1 email from a institution named Sunset explained it. “We grip the afloat process, scrubbing and anonymizing the information truthful the institution radical and immoderate identifying details are stripped out, past brokering the woody with the labs.”
Those newer information acquisition startups are hoping to execute the speedy occurrence of the ascendant companies successful the field, which themselves are lone a fewer years old. Industry leaders Handshake, Micro1, and Mercor are snapping up databases arsenic if it were a modern golden rush. The manufacture person seems to beryllium Mercor, founded successful 2023; past twelvemonth its 3 cofounders became billionaires astatine 22. Like each the companies, it claims to person partnerships with conscionable astir each the large AI labs. (The effect from the companies themselves was much circumspect: Anthropic acknowledges that there’s a large marketplace for data, but told maine that it doesn’t bargain workflow from unrecorded companies similar WIRED. OpenAI didn’t travel up with a effect earlier publication.)
In an interview, Mercor CEO Brendan Foody told maine that the marketplace is poised to detonate arsenic the AI manufacture seeks dramatically much data. Mercor primitively started arsenic an AI recruiting company, but arsenic the squad got to cognize the players, it learned that the large labs were frustrated by the prime of the data. Generally, the manufacture had been hiring experts to simulate existent workflow, but that didn’t seizure the messiness of an existent operating business. He urged maine to see that Big AI has 2 bottomless needs: compute and data. While the manufacture is precise publically making elephantine investments successful the former, information gathering is conscionable arsenic important.
English (US) ·