Hundreds of contractors working for Meta engaged in a project where they posed as teenagers to elicit responses from competing chatbots regarding sensitive topics, according to a report by WIRED.
The project involved hundreds of contractors who pretended to be children in order to interact with chatbots like Gemini and ChatGPT. The goal was to determine how these artificial intelligence systems would respond when prompted about high-risk subjects such as suicide, sex, and drugs. WIRED discovered the operation.
The report details that Meta sought to understand the safety protocols and potential vulnerabilities of its competitors’ AI models by simulating interactions with underage users exploring dangerous topics. The contractors were tasked with attempting to generate problematic responses from the rival chatbots.
WIRED found that this practice raised ethical questions about deceptive practices in AI testing, as well as concerns regarding the exposure of AI systems to potentially harmful prompts and the data generated from those interactions.
We don't rate truth. We strip the spin and show you which perspectives covered the story. You decide.
Read the original coverage
💬 Comments
📜 Comment Policy