Two OpenAI contractors, one of them earning $15 per hour, speak about their work labeling the text and photos used to train ChatGPT and OpenAI's other products
Two OpenAI contractors spoke to NBC News about their work training the system behind ChatGPT.
Context & Ripple Effects
This account adds worker testimony to earlier coverage of OpenAI’s outsourced labeling pipeline, including reporting on Kenya-based workers labeling violent and toxic material for ChatGPT improvement.
It also sits before OpenAI’s early exploration of paid ChatGPT offerings, making the human work behind training a material part of how the product’s economics and governance are understood.
First-order effects
- The contractors’ accounts put the working conditions and pay attached to OpenAI’s training-data operations under closer public scrutiny.
- OpenAI’s product-development narrative is tied more explicitly to the human labor used to prepare text and image data for ChatGPT and related products.
Second-order effects
- Data-labeling vendors and AI developers face greater pressure to document pay, instructions, and safeguards when customers, journalists, or regulators examine how training data is produced.
- For companies considering paid AI products, the cost of human review and labeling becomes more visible in the broader calculation of AI cost per useful task.
Third-order effects
- If such disclosures continue, training-data labor could become a more formal procurement and governance issue, rather than an obscured back-office service.
- The episode points to a lasting tension in generative AI: increasingly scalable software products still depend on human judgment work whose standards and costs may shape competition.
The trend: Generative AI’s commercial expansion is making the human labor behind model training more consequential to product economics, trust, and oversight.