Employees from Kenya worked on the ChatGPT neural network and were paid less than two dollars per hour. They studied materials about murders and sexualized violence - Time

The ChatGPT neural network has a built-in “detector” that filters out unethical content before the user sees it. For his “training”, OpenAI, which developed the neural network, hired low-paid employees from Kenya, Time reports.

The publication found that since November 2021, OpenAI, through the mediation of Sama, sent tens of thousands of text fragments to an outsourcing firm in Kenya, some of which detailed sexualized child abuse, bestiality, murder, suicide, torture, self-mutilation and incest. Employees had to find and mark such texts. A Sama worker told Time that he read a graphic description of sex between a man and a dog in the presence of a child while on the job. In a conversation with the publication, he called it "torture."

According to the publication, employees received from 1.32 to 2 dollars per hour. They had to read 150-250 text fragments in a nine-hour shift and mark them up. Each fragment was from 100 to more than 1000 words.

Sama employees were able to attend one-to-one consultations, but all Time interviewees admitted that these meetings were "rare and useless."

In February 2022, OpenAI and Sama ended their partnership ahead of schedule. According to Time, before that, OpenAI started working on another project and asked to process 1,400 unethical images. After that, the management of Sama announced the completion of joint projects.

Another reason why Sama stopped working with OpenAI was a February investigation by journalists about Facebook content moderators in Africa who had to view unethical content for $1.50 an hour, Time said.

Source: Time