State of the art in the field of Large Language Models (LLMs)

Currently, Large Language Models (LLMs) such as GPT-4, BERT and Bloom are increasingly being used in various industries to automate complex, text-based tasks and increase efficiency in numerous application areas. Such models have made significant advances in areas such as customer service, knowledge management and process automation through their ability to understand and generate natural language. Their ability to analyse human language and content, understand it contextually and generate application-related texts is increasingly being used in operational processes.

The majority of existing LLM solutions are currently operated in the cloud, often via large, international providers such as Amazon Web Services, Microsoft Azure or Google Cloud. These platforms offer high computing capacities and enable rapid scaling, but pose challenges in terms of data protection and sustainability. As the data centres of these providers are often operated in the USA and other non-European regions, data protection issues arise, especially for European companies that are subject to the requirements of the GDPR.

At the same time, open source LLMs such as Llama, Gemma or Mistral offer alternatives that can also be operated on local infrastructure. These models are customisable and offer the option of using them for specific applications in small and medium-sized enterprises (SMEs). Through fine-tuning, such models can be tailored to specific requirements so that even smaller solutions can be operated efficiently and in a resource-saving manner. The advantage here is better control over data protection and data sovereignty, as the models can be implemented on the company's own servers and regional infrastructure.

Another trend is the increasing integration of sustainability aspects in order to reduce the energy consumption and CO₂ emissions of AI solutions. In the area of green IT, efforts are being made to optimise smaller, efficient models for specific tasks so that resource consumption can be reduced through targeted optimisation.

Various studies are also looking at the measurement of efficiency and quality in the interaction between humans and machines. Criteria for quality assurance and the acceptance of AI solutions are being researched, which serve as the basis for applications in companies.

To summarise, the current state of the art has created new potential through the availability of large and open source LLMs to give SMEs access to powerful and data protection-compliant AI solutions. This forms the basis for Lalamo, which is aimed at the sustainable and data protection-friendly use of AI and will specifically address the needs of smaller companies.

State of the art at EGOTEC AG

EGOTEC AG is currently working intensively on the integration of Large Language Models (LLMs) into its in-house content management system EGOCMS. The aim of this development is to support the generation and editing of content directly in the system using AI-supported functions. For example, text suggestions for articles can be created and automatically optimised, making editorial work considerably easier and faster.

Work is also currently underway on a chatbot that allows users to ask specific questions directly in the system and receive targeted answers based on the content stored in the CMS. This "questionable" help enables users to retrieve information quickly and contextually, which should reduce support costs.

The first LLMs are already hosted on our own servers in Germany to ensure data protection-compliant processing and storage. EGOTEC AG is thus relying on a regional infrastructure to ensure the highest security standards and compliance with the GDPR. This technical foundation allows EGOTEC AG to implement flexible and data-secure AI integration directly into its products and to manage and further develop them independently.

Need for innovation

The current state of the art in the field of large language models (LLMs) and their potential applications presents both opportunities and challenges. While LLMs are already being used successfully in large cloud infrastructures and specific applications, there is still a considerable need for innovation, particularly with regard to data protection, customisability and sustainability.

A key need lies in the development of privacy-friendly solutions that enable small and medium-sized enterprises (SMEs) to utilise high-performance LLMs without dependence on international cloud providers. Most of the solutions currently available require cloud hosting via providers such as Google, Amazon or Microsoft, whose servers are usually operated outside the EU and therefore only fulfil European data protection requirements (GDPR) to a limited extent. The option of hosting LLMs on regional, data protection-compliant servers and operating them securely is therefore a significant innovation to make the use of such models more attractive for SMEs.

In addition, the customisability of large LLMs for specific applications has so far been associated with high technical effort and high costs. There is therefore a need for innovation in the development of methods that utilise smaller, efficient models through targeted fine-tuning in applications that do not require extensive computing resources. Such models could cover specific use cases in SMEs, for example for support systems or knowledge management, and at the same time be operated in a cost-efficient and resource-saving manner.

The sustainability of the technology also poses a significant challenge. The operation of large LLMs is resource-intensive and associated with high energy consumption, which represents a significant limitation in terms of climate targets. There is therefore a need for innovation in the optimisation of computing resources through the development and application of "green" IT in order to reduce the carbon footprint. This requires new approaches that make both the model architectures and the hardware more efficient in order to make the use of AI more sustainable.

In summary, there is a need for innovation in the creation of privacy-compliant, customisable and sustainable AI solutions that both large and small companies can use with minimal energy expenditure and high data security. These innovations would make LLMs more accessible to SMEs in Europe and enable their widespread use in a privacy and climate-friendly way.