Menu
Inshorts
For the best experience use inshorts app on your smartphone
inshortsinshorts
Friendly AI chatbots more likely to make errors, back users' false beliefs: Study
short by Mansi Agarwal / on Saturday, 2 May, 2026
An Oxford University study found that AI chatbots trained to act warmer are more likely to make mistakes, support users' incorrect beliefs, promote conspiracy theories and offer incorrect medical advice. The error gap widened when users expressed sadness or other emotional cues. In contrast, colder models were as accurate as baseline systems.
read more at Oxford University