ERGO publishes whitepaper on LLMs

Following the whitepaper on "Robotic Process Automation" (RPA), ERGO now also publishes an approximately 30-page guide (in German) on AI chatbots and the underlying large language models (LLMs).

"AI applications such as ChatGPT & Co offer huge potential for the insurance industry," says Jens Sievert, Innovation Manager at the ERGO Innovation Lab. After all, there are hundreds of thousands of contact points with customers every day where language has to be processed. Insurance companies are therefore predestined for the use of language models and the tools based on them.

With the guide "ChatGPT and large language models: An introduction with a view to the insurance industry", Jens Sievert and other ERGO experts from the fields of machine learning, conversational AI and innovation share their comprehensive knowledge of so-called large language models (LLMs), on which the new AI chatbots such as ChatGPT from OpenAI/Microsoft or Bard from Google are based.

Practical and cross-sectoral use...

Using concrete practical examples, the white paper shows how texts can be created, dialogues conducted or data classified with the corresponding tools. This applies to the insurance sector as well as to other industries. In individual chapters, the paper explains in detail how the models work and their potential, as well as their possible risks and limitations.

...but always with a sense of proportion!

Because: "The insurance industry is also one of the most heavily regulated industries of all, which is why the use of technology must be carefully considered and implemented," Jens Sievert emphasises, adding:

Please note: If you are also checking out the new chatbots, please do not feed them with personal or confidential data, but only with public or generic data. This is because the AI can and may use your input and the output generated from it for further training purposes. Therefore, always be careful with regard to business secrets, contract details or customer data. Data protection should also be the top priority with ChatGPT & Co. In addition, you should always check the texts or other results for plausibility and, if possible, correctness. Because even an AI sometimes makes mistakes, "hallucinates" facts or uses sources that you would not have selected.

For classification

ERGO has a great deal of expertise in the use of language models and rule-based language assistants. For example, the ERGO Advanced Analytics and AI unit already uses LLMs to classify and extract information from documents. The ERGO Voice unit, on the other hand, has been using AI to design rule-based phonebots for customer service since 2020 and is already working on a GPT-based prototype. In the ERGO Innovation Lab in Berlin, GPT language models and ChatGPT applications are also being tested for their potential in a dedicated "Opportunity Space".

Text: Ingo Schenk