The current economic situation is challenging for many companies and the cost pressure is high. German guiding industries see themselves in a tough competition for markets. In many places, the supply chains are no longer as stable as they were. In view of this, solutions are urgently needed so that companies in Germany also remain internationally competitive in the long term.
In a high -wage country – like Germany – the path to increased competitiveness can actually only lead to better products and more efficiency. So it is about higher productivity, shorter development times and the acceleration of manual activities through their automation. “German Engineering” still has a good reputation, but competing products are often also good today and are often cheaper.
Peter Liggesmeyer is the head of the Fraunhofer Institute for Experimental Software Engineering IESE in Kaiserslautern and owner of the Chair of Software Engineering at the Computer Science Department at the RPTU Kaiserslautern-Landau. In addition to the focus topics of artificial intelligence, autonomous systems and Industry 4.0, his research focuses are primarily concepts in the area of safety and security.
Support of development activities through large voice models
Large voice models are able to create texts on a given topic. Software – that is, computer programs – are basically written in a programming language. Therefore, large voice models are in principle able to automatically generate those parts of software that are intended to take on standard functions and are therefore in a similar form in the training data of the language models.
A software developer would not completely write these program parts themselves, but would further develop the solution generated by a large voice model, which ultimately saves time and therefore of course money. The creative performance of the developers is used sensibly. Productivity is increasing and development times are shortened. The challenge is also reduced to win suitable workers. Companies can become more competitive in this way.
However, the expected advantages are likely to have a disadvantage: large language models “hallucinating” occasionally. This means that you create incorrect results, but often in such a plausible way that it is not easy to recognize. The developer will therefore have to invest a certain effort in the review of the generated results, which reduces the achievable productivity and time gain. I am still convinced that the benefits will predominate in many situations.
Automated production of digital twins
In principle, digital twins are virtual images of a real or logical body – for example a machine, a person or an order. Depending on the purpose, you can provide information, simulate behavior or enable predictions. In the future, digital twins will play an important role in many industries – for example in the production technology of the fourth industrial revolution (Industry 4.0,) in medicine or in pharmaceuticals. I consciously say “in the future”, because although digital twins are already used in many industries, this is far from being succeeded in all areas.
One reason for this is the often high manual effort for the generation of the twins, because the data required for this is currently still in very unstructured form in many companies. Companies understandably shy away from the effort of transferring this information “by hand” into digital twins. And now LLMS come into play: You can structure the data and texts in a uniform way and extract the relevant information from you. With these bundled information, digital twins can then be “fed”. This is not perfect, but of course it is a profit compared to a purely manual approach.
Research mandate: generate evidence
The examples show that large voice models enable certain advantages that can help to open up new potential for companies. Due to the rapid development in large voice models, we are only at the beginning. We see opportunities, but have fewer secure insights into where they are, how great the achievable benefit will be and what risks we take. To do this, we will have to generate evidence.
Although the current market environment is challenging for many companies, in my view you are well advised to deal with these topics at an early stage and to consider the extent to which advantages can be opened up here in the future. At the same time, it is the task of research to generate the necessary evidence for this topic – that is, well -founded knowledge that show the increase in efficiency with the help of large voice models. Finally, the more specific the basis for decision -making, the easier it is to make companies the famous jump into the cold water. And it is precisely on this extraction of the evidence and the accompaniment of companies in this way we are currently working on Fraunhofer IESE.
