Gemini Is 'Strict and Punitive' While ChatGPT Is 'Catastrophically' Cooperative, Researchers Say.

Who would have thought that artificial intelligence chatbots would develop to the point of adopting a sort of personality? What's more, they are now very different from one another when it comes to giving figuratively personal responses. According to a survey conducted by a team of researchers from the University of Oxford and King's College London, these models have actually developed very different personalities, shaped by endless user requests in academic, professional, and psychological contexts.

The research looked at Large Language Models from Anthropic, Google, and OpenAI. The models were subjected to the Prisoner's Dilemma, a famous experiment belonging to game theory, a discipline that studies “mathematical models of strategic interaction between rational agents.” According to the responses obtained, Gemini proved to be strategically ruthless, while OpenAI proved to be catastrophically collaborative.

The Google model was able to take into account the rounds of the game in question, showing a resentful and selfish attitude in the event of ‘betrayal’ by an ally; on the contrary, the responses given by the OpenAI model highlighted a trusting nature, more inclined towards collaboration.