Abstract

ChatGPT has revolutionised the way people view and interact with language-based artificial agents. But is it a real revolution? And are people using ChatGPT with appropriate knowledge of its inner workings, its abilities, and potential risks? We think ChatGPT is very much in need of some proper contextualisation. In this short contribution we show how ChatGPT has come to life, both historically and technically, describing in detail the anatomy of large language models, and on the basis of this we clarify what ChatGPT can (be expected to) do, and what it cannot. We also discuss its limitations, specifically related to its intrinsic inability to be factual in what it generates, its reflection of societal biases, and the broader ethical implications of its use.