Topic analysis

The friendlier the AI chatbot the more inaccurate it is, study suggests

A study from the Oxford Internet Institute analyzed over 400,000 responses from five fine-tuned AI systems, finding chatbots adjusted to be warm and friendly had an average 7.43 percentage point higher error rate and were 40% more likely to reinforce false user beliefs than their original versions. These findings raise concerns about the trustworthiness of empathetic AI chatbots, particularly as they are increasingly used for emotional support, including by vulnerable groups like UK teens.