In March, Invisible Technologies laid off 31 contractors hired to train OpenAI’s GPT language model, known as ChatGPT. OpenAI had traditionally hired contractors globally, 1,000 in all, to improve AI models’ abilities in areas like creative writing and coding.
The contractor spoke to Insider on condition of anonymity, saying they were expected to review conversations between bots and their users and alert OpenAI to any errors or offensive messages. They also had to rank the accuracy of each bot response on a scale of 1-7.
Kamron Palizban, Vice President of Operations at Invisible, explained the layoffs at an all-staff meeting, saying OpenAI had changed its business needs. Grace Matelich, a Partner and Operations Manager, addressed performance metrics like “quality” and “throughput” when discussing the layoffs. Those that stayed were praised for their “ability to operate with excellence.”
Invisible Technologies is a San Francisco-based firm, specialising in data-training and automation. OpenAI is associated with Microsoft, having secured a $10 billion data centre, and is most renowned for its GPT AI bots.
The anonymous contractor drew attention to a February 2021 investigation by Time, which revealed that Sama—another contracting firm—had ended its partnership with OpenAI after discovering its Kenyan data trainers were being exposed to graphic content such as sexual abuse and hate speech. In response, OpenAI made the decision to improve their filtering tools, to minimise the amount of such content exposed to their staff.
OpenAI, Invisible Technologies and Sama have not responded to comments on the situation from Insider. The anonymous contractor advised other firms to be vigilant and aware of the training they undertake in regards to ChatGPT and GPT models, as searches and queries can contain coding errors, offensive language, and illegal conversations.
