German start-up, OpenAI innovates with AI-powered customer service

By Christoph Dernbach, dpa l Monday, February 20, 2023

 

BERLIN – Despite the lead of US developers in artificial intelligence, a humble German start-up is on their heels with an AI system that can also provide individual customer service.

The language model Luminous, pioneered by Aleph Alpha in the southern city of Heidelberg, scored similarly well against pre-eminent AI specialist OpenAI or the model developed by Facebook parent company Meta, the company’s CEO and founder Jonas Andrulis said on Monday.

The standardized performance comparison included tasks for classifying, evaluating and creating texts, as well as answering questions about specific text content.

Luminous was the first example of a European AI language model challenging the US tech giants and even outperforming them in some respects, Andrulis said.

The AI model has about half as many (70 billion) parameters as its rivals, meaning it has twice the efficiency at the same level of performance, Andrulis said, making it a viable rival to the big-hitters.

“Luminous is a strong alternative in many situations and thus an important step towards Europe’s technological sovereignty,” the developer said.

Californian start-up OpenAI has won broad attention with its text robot ChatGPT – developed with support from Microsoft – since many millions of users are able to try out the AI application.

Aleph Alpha does not offer a mass product for private consumers, partly because it cannot afford to operate millions of queries every day. However, part of its strategy is to focus on solutions for companies or public administration.

In October, its “Lumi” citizens’ assistant, which is based on Luminous, was put into operation in Heidelberg. The system is able to respond to very individual, non-pre-programmed questions.

This makes it possible to make information accessible to people who, due to poor spelling or other language problems, may have had problems communicating with authorities in the past.