AI trained AI: OpenAI fires contractors

TVS News AI, editorially supervised, presents AI trained AI: OpenAI fires contractors.

Whoever is supposed to train an AI should not, of all things, use an AI for it – obvious, really. But that is reportedly exactly what happened with several contractors who were working on OpenAI models on behalf of service providers. They were fired.

OpenAI relies on human trainers to fine-tune its models – people who rate answers, correct mistakes, and provide feedback. When these contractors instead let an AI do the work, the quality of the training data changes fundamentally – and with it, what eventually reaches users.

What is known so far

According to Golem.de, service providers have dismissed several contractors who were supposed to improve OpenAI models. These contractors had used AI themselves for their task – in other words, they used the very tool they were supposed to be training.

How many people are exactly affected, which service providers sit between OpenAI and the trainers, and what consequences the AI usage has for training quality does not emerge from the report so far.

Why the case is likely to make waves

When the ratings that are meant to improve an AI model themselves come from an AI, the entire training process gets caught in a loop. That human oversight is precisely what OpenAI and its service providers are trying to ensure with external trainers.

The incident shows how tempting it is for contractors to cut corners on their tasks using AI – and how hard it remains for the clients to reliably detect that.