LARGE LANGUAGE MODELS IN THE AUTOMOTIVE INDUSTRY

FROM ASSISTANCE SYSTEMS TO SMALL TALK EXPERTS

Generative AI and large language models are also shaping the automotive industry. Where are they already being used? What challenges does integration pose? What will these technologies (literally) bring to the road in the future? Dr. Christoph Endres, Managing Director of our sister company sequire technology, answers these and other questions in an interview with Alexander Pfalzgraf, founder and CTO of SemVox.

Short biography:

Alex Pfalzgraf studied computer science in Lübeck and Saarbrücken and founded SemVox GmbH in 2008 with three other scientists as a spin-off of DFKI (German Research Centre for Artificial Intelligence). As CTO, he is responsible for the technical implementation and was instrumental in transferring his own technology to series production in the automotive industry and other sectors.

Christoph Endres: Hello Alex, thank you for taking the time to talk to us. We share a common past as researchers at the DFKI (German Research Center for Artificial Intelligence). Back then, you brought a research idea to industry and are one of the four founders of SemVox. Looking back, you were already super innovative in 2008, when you founded the company and built natural language assistance systems. What was so revolutionary about your semantic approach?

Alexander Pfalzgraf: If you compare the status quo in 2008 with today, you can easily see the dynamics in IT, especially in AI. When we launched our technology on the market, there were no big tech players on the playing field – Apple Siri, for example, came afterward. Recently, this market has been shaken again, especially with the emergence of OpenAI. Everyone is talking about Large Language Models (LLMs) and their impressive performance today. Our approach back then was different: the goal was efficient modeling of language dialogs based on semantic abstraction. Despite the complexity of natural interaction, this also resulted in good manageability regarding the computing power required in embedded systems and the comprehensibility of the algorithms. Today, we are working on integrating both approaches and their coexistence.

Christoph Endres: And as a result your company has also rapidly grown as a result. I remember meeting you at IFA in 2010 (International Radio Exhibition Berlin) – a lot has happened since then. How big is the company today?

Alexander Pflazgraf: In 2008, the four of us founded the company. At our first sale ten years later, we were about 80 people, and today we are around 180.

Christoph Endres: You are now part of a large group of companies and, therefore, strongly represented in the automotive sector. How do you see the future of driver assistance systems – especially in the current hype surrounding LLMs? Will we get an annoying Knight Rider at some point? Will my car be making small talk with me in ten years? It’s all very difficult for me to imagine how this will develop.

Alexander Pfalzgraf: Small talk is certainly possible with today’s LLMs. But especially in the vehicle, it’s more about relieving the driver of tasks and supporting them – gimmicks wear out quickly. LLMs are also very promising here, although reducing interaction will be an important goal. I think assistance functions will work in the background and rely on getting to know their users better – at least to the extent they want to allow.

Christoph Endres: I have another professional perspective: the security of Large Language Models.
We published extensively on this topic last year. What is your point of view on that? Are there safety concerns about such systems in the automotive industry?

Alexander Pfalzgraf: Like other parts of the software stack in the vehicle, voice dialog is generally subject to all cybersecurity requirements, and threat and risk analysis are also part of this. In this respect, integrating LLMs is also subject to these security requirements. However, as LLMs are comparatively new and some potential threats are still unknown, the corresponding integration must be planned with foresight and equipped with preventive measures. Security and data protection must not fall victim to the pressure to bring LLMs onto the market.

The interview with Alexander Pfalzgraf offers a fascinating insight into the future of automotive technology and the role that generative AI could play in it. Thank you for the interview.

cropped-christoph_endres.png

Dr. Christoph Endres
managing director
sequire technology

Alexander Pfalzgraf

Alexander Pfalzgraf
Founder CTO
SemVox