Tens of millions of people have already used chatbots in the past few months.
Image: Lucas Bäuml
ChatGPT’s success was surprising, its limitations are not. How do the language systems work, how can they get better – and what are the challenges?
Man may have to go back to the introduction of the iPhone: Few things have garnered as much attention in the computing world in recent years as the introduction of ChatGPT. “Try talking to him,” Sam Altman, chief executive of IT firm Open AI, tweeted in late November. Since then, around 100 million people are said to have tried the chatbot with its surprisingly well-formulated texts. A poem in the style of Goethe about Covid-19, a summary of the “Odyssey”, training plans for bodybuilding, a rejection of an invitation or software codes: ChatGPT provides automatically generated texts for almost all purposes, which the bot also adapts based on the interaction with a user . “Please with more humor”, “please something more formal” – he fulfills many wishes in a very short time.
The promises are great, so is the hype: the mission of Open AI is to ensure that artificial intelligence (AI) “benefits all of humanity,” explains the US company on its website. At the same time, he wanted to avert damage from her. That doesn’t always work: According to research by the US magazine Time Through a service provider in Kenya, low-paid workers have identified passages in sample texts that contain hateful language and descriptions of sexual and other violence – traumatizing work.