ChatGPT has attracted much attention in recent months. Indeed, ChatGPT’s performance is impressive – and together with solutions like Midjourney or Lensa, it has made the general public aware of AI as a content creation tool.
ChatGPT impressively demonstrates how human-centric design and low-threshold access can get people to open up to the use of AI – and what it can make possible. Emails, applications, essays, program code, word games: There are almost no limits to the versatility of ChatGPT.
💡 Know what you’re dealing with
Behind ChatGPT is a language model that outshines everything known so far. But it is important to bear in mind that it is designed to be a generalist: ChatGPT’s text completion relies on a wide range of content in its training data. We must therefore be aware that we are not dealing with quality-assured data in ChatGPT. The model aggregates speech-aware output from probabilities and patterns present in its comprehensive training data. It does not, however, guarantee factual accuracy, and so, in turn, users do not know whether a piece of information was “invented” or has a reliable source. Human expertise is needed to identify possible misstatements – and one strand of the current discourse around ChatGPT centres on the urgency of developing literacy and skills in this space.
➡ As with all tools, users must learn to use ChatGPT and be aware of its weaknesses.