Two OpenAI contractors, one of them earning $15 per hour, speak about their work labeling the training data used for the company's products, like ChatGPT
Two OpenAI contractors spoke to NBC News about their work training the system behind ChatGPT. — Alexej Savreux, a 34-year-old in Kansas City … Tweets: @ndiakopoulos Tweets: Nicholas Diakopoulos / @ndiakopoulos : “Two OpenAI contractors spoke to NBC News about their work training the system behind ChatGPT” https://www.nbcnews.com/...
Context & Ripple Effects
The account adds a U.S.-based view of the human work behind OpenAI’s systems after reporting that the company used Sama in Kenya to label violent and toxic material for model improvement, documented in an earlier investigation into OpenAI’s toxic-content labeling pipeline.
It also follows reports that OpenAI had expanded its remote contractor base, with a majority assigned to data labeling, making this less an isolated job account than evidence of a rapidly scaling training-data operation. The reported contractor hiring push gives the individual accounts operational context.
First-order effects
- The contractors’ accounts make the human labor underpinning ChatGPT’s training data more visible, putting OpenAI and its labeling vendors’ pay, working conditions, and quality controls under closer scrutiny.
- For the workers doing this task, labeling remains a distinct, paid production role rather than an invisible byproduct of automated AI development.
Second-order effects
- Data-labeling providers and AI developers face greater pressure to demonstrate that their safety and quality claims are supported by workable labor practices, since the same workforce helps determine what models learn to reject or recognize.
- Labor costs and reviewer capacity become more consequential inputs to model development: expanding training and moderation datasets requires either more contractor work or greater investment in tooling that makes reviewers more productive.
Third-order effects
- If scrutiny persists, training-data supply chains may become a more visible part of AI governance, alongside model outputs, with customers and policymakers asking who performs high-risk annotation work and under what standards.
- The case points to a durable split between AI’s automated interface and its labor-intensive production system; competitive claims about capable, safe models will increasingly rest on both.
The trend: Generative AI is scaling through an increasingly industrialized human-data supply chain, making the cost and governance of annotation part of the product itself.