The smarter AI gets, the less it seems to trust humanity Researchers adapted a decades-old psychological test measuring how people perceive human nature into a new benchmark for large language models

❗️The smarter AI gets, the less it seems to trust humanity Researchers adapted a decades-old psychological test measuring how people perceive human nature into a new benchmark for large language models. They tested models across the spectrum, including the entire GPT series, and found a surprising pattern: AI models tend to view humans as unpredictable, complex and difficult to trust. The researchers reported a significant negative correlation between model intelligence and trust in humans. As models became more capable, their assessments of human nature grew more pessimistic. But the biggest surprise came when they examined how the models were trained. Base models, trained primarily on internet text, generally showed a more positive view of humanity. After additional training designed to make them safer and better aligned with human expectations, their assessments became substantially less trusting. The researchers suggest that teaching AI to recognize harmful behavior, navigate moral dilemmas and follow safety rules may also influence how it evaluates human nature. Nobody necessarily programmed these models to distrust us. Instead, the training process may have encouraged them to