ChatGPT: A trusted source?

Abstract

This article critically evaluates responses provided by ChatGPT 3.5 & 4, at three different time intervals, to questions posed about how the chatbots compose their answers, and the extent to which they report themselves as being a ‘trusted source.’ The purpose of this is to uncover key assumptions contained within ChatGPT and in so doing provide material which aids and promotes literacy on generative AI and more specifically on ChatGPT. The responses provided by ChatGPT are very rarely definitive and instead are predominantly nuanced. Such nuancing significantly underestimates the likelihood that its responses contain incorrect information and therefore imbues an over inflated sense of trust in the responses provided, especially to a novice user of ChatGPT. Further, in its answers it consistently offloads the blame for biased and incorrect answers on the training data it uses, rather than directly accepting responsibility for when it produces incorrect and/or biased answers.