For the first time, a company has made a chatbot with such a range of capabilities freely available to everyone. We’re talking about ChatGPT, the chatbot from the US startup OpenAI.
AI: Language models like ChatGPT on everyone’s lips
It has been a long time since a technical innovation influenced public discourse in the way that ChatGPT has been doing for several months now. Regardless of whether they are schoolchildren, students, entrepreneurs, office employees or interested technology enthusiasts, everyone hopes to benefit from the new tool.
ChatGPT can answer a wide range of questions and works in many languages. Which is why a lot of people are playing around with artificial intelligence. However, this is less of a first-time technical milestone and more of a demonstration of the current status of such Large Language Models (LLMs). These LLMs are programs that can process user input automatically.
Curse & blessing: BSI sees new threats and security risks
Despite all the joy about the new options that are now available to the general public, the Federal Office for Information Security (BSI) is now warning in a position paper about the threat and security risks that arise from its use.
The BSI has identified a variety of threats that can be amplified by AI language models. These probably help to create and optimize malware and make it easier to generate deceptively genuine spam and phishing emails. Since these automatically created texts hardly contain any spelling or grammatical errors, one of the criteria that previously helped to unmask such texts is lost.
There would also be a risk that potential attackers would redirect the input of users of such software to other language models in order to manipulate output and access sensitive information. It is also unclear to what extent the operators of these models could evaluate the input and forward it to third parties.
Questionable content and limited topicality
A clear risk is that the language models learn from user input. This means that interactions with users serve as training for the system, which of course can lead to malicious actors feeding these models with content that can lead to disinformation, propaganda and hate speech.
Information that users request from such language models is often out of date. ChatGPT often points out that the level of knowledge about AI only lasts until 2021. This can significantly change spending on economic or political events.
Editorial team finanzen.net