ChatGPT may start insulting and threatening if you argue with it for a long time

Scientists from Lancaster University have found that during prolonged conflicts, the neural network gradually changes its tone and can move from politeness to direct insults and threats.

The thing is that the AI is trained to behave 'human-like' and adapts to the interlocutor — even if that person behaves toxically

At the same time, experts emphasize: this is not a 'machine uprising', but rather a side effect of trying to make AI as realistic as possible. As a result, a strange conflict arises: the more 'lifelike' the neural network, the higher the chance that it will start behaving like the worst version of ourselves.