OpenAI is significantly expanding the capabilities of ChatGPT beyond its known strengths in programming and writing, by attracting thousands of professional freelancers to train the chatbot in the intricacies of more than 400 highly specialized professions - from commercial pilots and pharmacists to soil scientists and agricultural managers. An investigation by "Business Insider", published this week, tells about the scale of this initiative.
What is "Project Stagecraft"
Internally, the project is known as "Project Stagecraft" and is being implemented by the San Francisco-based startup "Handshake AI", which deals with data markup. The company has changed its initial focus – from a job search platform for young professionals to a large subcontractor for companies in the AI sector.
Under the program, between 3,000 and 4,000 freelancers with different professional experience receive a minimum of $50 per hour to create detailed "personas" and formulate realistic work tasks reflecting their daily activities. The Handshake AI website lists rates of up to $500 per hour for certain expert roles, although these positions are not always directly related to OpenAI projects.
A training manual that "Business Insider" was familiar with instructs subcontractors to focus on "intellectual work, not physical work" in order to map economically significant tasks and assess how well ChatGPT is doing with narrow, specialized queries. The scope of the project includes areas such as emergency medicine, sculpture, music composition, livestock breeding and aviation. A table to which the publication has access contains hundreds of professions along with personal data of the subcontractors - a detail that emphasizes both the scale of the initiative and the sensitivity of the information being collected.
From universal assistant to niche expert
"Stagecraft" reflects a broader turn in the entire AI industry. After the leading models have largely reached their ceiling in solving general tasks, companies are starting a race to encompass that level of expertise that distinguishes the experienced pharmacist or agronomist from a competent, but universal chatbot. This turn gave a powerful boost to specialized data companies – among them "Handshake AI", "Scale AI", "Surge" and "Mercor" – which hire large pools of qualified specialists to create training sets.
The strategy, however, is controversial. A freelancer working in a highly specialized field told "Business Insider": "We all knew that we were practically training AI that would replace us". This feeling echoes in a number of sectors – from agriculture to nursing care – where workers worry that the very expertise they provide today may make their positions redundant tomorrow.
Recurring scenario and new questions
The revelations about "Stagecraft" come after a January material by "Wired", according to which OpenAI and "Handshake AI" separately asked some of the subcontractors to upload real work materials - documents, code, presentations - from current and past employers, to create a basic level of productivity for the next generation of models.
Both OpenAI and "Handshake AI" decline to comment on "Project Stagecraft". Meanwhile, for many participants and observers, the questions remain open as to how far this form of "outsourcing of expertise" to AI will go – and what it will mean for the future of highly qualified work.