Providers of large language models claim that their tools can not only reduce the burden of repetitive tasks, but also support more complex business processes. To assess whether these promises are realistic, an article on Heise online compares language models with the different levels of autonomous driving.
Level 2 means that a human must continuously supervise the system and intervene when necessary. At Level 3, a human still needs to be present, but no longer has to monitor the system constantly. The author currently places language models somewhere between these two levels.
The article concludes that large language models are strongest when working with text and summarising content. Their performance is weaker in research, particularly when it comes to specialised questions, where the models are often less successful. For the time being, experts are therefore still needed to answer highly specific questions reliably.
By Markus Hoffmann
Source: Heise online, 4 August 2025, by Philipp Steevens
