How Freelancers Are Quietly Training ChatGPT

Somewhere between a farm field and a flight deck, ChatGPT is learning how the real world works—one freelance task at a time.

Behind the scenes, a growing network of contractors is feeding the AI highly specialized knowledge. Not broad, generic data—but the kind of expertise you’d expect from a soil scientist, a commercial pilot, or a medical professional. And they’re getting paid well to do it.

Documents reviewed by Business Insider reveal an OpenAI-linked initiative—internally dubbed Project Stagecraft—where thousands of freelancers are helping shape how AI understands real jobs. The work is coordinated through Handshake AI, a San Francisco startup that has quietly pivoted from a career platform into a major player in the data-labeling economy.

Teaching AI to Think Like a Professional

This isn’t your typical data-tagging gig. Contractors are asked to step into a role—say, a nurse practitioner or a music director—and build realistic scenarios from scratch. They write prompts that mirror actual workflows: drafting reports, analyzing datasets, or preparing deliverables that reflect real-world expectations.

The twist? Everything must be done as if assigning work to a human colleague, not an AI. No simplifying. No “dumbing it down.” The goal is authenticity.

One training guide puts it plainly: focus on knowledge work, not manual labor. That means simulating decision-making, critical thinking, and domain expertise—skills that define modern professions.

In one example, contributors are asked to act as financial managers for a theater company, building payroll spreadsheets based on union agreements. In another, they might generate a 10-page medical literature review. These aren’t toy problems. They’re slices of real jobs.

A Workforce Scaling Into the Thousands

Roughly 3,000 to 4,000 freelancers are believed to be involved in the project so far. Pay starts at around $50 per hour, but for highly specialized expertise, rates can climb dramatically—sometimes up to $500 per hour, according to listings on Handshake’s platform.

The range of professions is striking. Think commercial pilots, geoscientists, pharmacists, farmers, HR specialists, even sculptors. A leaked spreadsheet tied to the project listed hundreds of roles across industries, along with contractor contact details—raising quiet concerns about data exposure.

Each submission goes through multiple layers of review: first internally, then by subject-matter experts, and finally by OpenAI itself. The idea is to ensure the model doesn’t just sound smart—it behaves like someone who actually knows the job.

The result is a shift in how AI is trained: less about general intelligence, more about professional credibility.

But there’s an uneasy undertone. Some contributors are fully aware of the irony. As one contractor put it, they’re effectively training systems that could one day replace them.

That tension has only grown with recent reports of payment disputes. Several contractors claim they were denied thousands of dollars for completed work after allegedly violating platform rules—decisions they say came without appeal. Handshake has declined to comment publicly, and OpenAI has remained silent on the matter.

Still, the momentum is unmistakable. As AI systems push deeper into specialized domains, the demand for real human expertise isn’t shrinking—it’s evolving. For now, the people closest to the work are shaping what these models become.

And in doing so, they may also be defining the future of their own professions.

Related Post

Read More

OpenAI’s Hiring Surge Signals a New AI Power Play

Layoffs have become a familiar headline across Silicon Valley. OpenAI, though, seems to be reading from...