Dusseldorf Just a few days ago, Feiyu Xu was on stage for SAP at the Sapphire customer conference to talk about the use of artificial intelligence (AI) at the software manufacturer. But the scientist who is currently in charge of the department will soon be leaving the DAX group: she is founding the start-up Nyonic with several well-known players from the tech scene.
The company wants to develop a foundational model for AI that is suitable for special applications in business and supports several European languages, it announced this week. A first version of the system should be available early next year and will then compete with the products of leading providers such as OpenAI.
In addition to Aleph Alpha, another start-up is working in Germany on the development of generative AI, which has created a real hype. The systems are capable of automatically processing speech and generating content. So far, they have mainly been developed in the USA – by OpenAI, Microsoft and Google, for example.
Feiyu Xu drives the topic. She has “hardly been able to sleep properly for months” and is constantly thinking about the potential of this technology, the computer linguist said in an interview with “Zeit” in mid-March: “AI will change many processes in companies and affect almost all areas of society. “
Nyonic wants to develop the basic technology for this. The founding team includes other experts who are well-known in the AI scene. The management is shared by Vanessa Cann, currently head of the AI Federal Association, and the serial entrepreneur Han Dong, whose previous positions include the start-up company Rocket Internet.
>> Read here: Programming code in a few minutes – How SAP wants to use artificial intelligence
Other co-founders are the physicist Johannes Otterbach, who worked at OpenAI for almost three years, and the computer linguist Hans Uszkoreit, who until recently held a managerial position at the German Research Center for Artificial Intelligence (DFKI) – and is married to Feiyu Xu.
Artificial intelligence for the factory
At the beginning of the year, ChatGPT widely publicized the possibilities that AI offers. The program can write texts of all kinds and answer, translate and program questions. Numerous companies are working on incorporating the technology into their applications.
There should also be many possible uses in industry – for example, through targeted answers from manuals and machine documentation or when operating a system using spoken language. This is where Nyonic comes in, which wants to focus on the requirements of companies in various industries.
“AI language models offer such a variety of possibilities that they cannot be covered by one or two companies,” said co-founder Vanessa Cann of the Handelsblatt. There are differences in both the technical approaches and the data used. The aim of Nyonics is to initially address the top-selling companies in Europe, such as the Dax groups.
AI language models offer such a variety of possibilities that one or two companies cannot cover them. Vanessa Cann, co-founder of Nyonic
According to Cann, the start-up will train its language model with patent data, process manuals and legal documents. “We quickly realized that corporations have their own language.” For example: “In the context of Deutsche Bahn, the word ‘Halt’ means ‘Haltestelle’, in general it is more of a synonym for ‘Stopp!’ used.”
Read more about artificial intelligence
Cann also sees an opportunity for herself and her colleagues in the market for large language models because many corporations and medium-sized companies are reluctant to share company data with Microsoft, for example: “Many companies are not allowed to use US language models according to their own rules.”
In addition, the founder points out that German companies cannot rely on the fact that services from a US company like OpenAI will be permanently available to them. “OpenAI boss Sam Altman himself said that his company would withdraw from Europe if the AI Act entailed too many restrictions,” she says, referring to the regulatory project for AI planned in Brussels.
Competitors with billions in financing
There is an expectation in the industry that, in addition to large providers such as Microsoft and Google, numerous other companies will be able to develop models – also thanks to open source technology that anyone can use for their own purposes. Startups like Anthropic, Jasper and Cohere are also developing enabling technology.
The Nyonic team competes with companies that have received millions and sometimes even billions in funding and have been developing language models for many years. Aleph Alpha, for example, has already received 28 million euros. Nyonic should have a similarly high capital requirement when it comes to training its own models. The start-up is currently trying to find investors.
However, size and financing volume may not be the only decisive categories. Kristian Kersting teaches and researches artificial intelligence at the Technical University of Darmstadt and has participated in the start-up Aleph Alpha. He says: “We see new models coming onto the market almost every day – including smaller and specialized ones.”
>> Read here: Aleph Alpha makes AI understandable
It is an open question whether you can get further with smaller models than with all-encompassing models. In his view, specialization is also becoming important with regard to European values.
Kersting welcomes the fact that there will be another ambitious competitor in Germany. “From my academic point of view, we need several players on the field, you won’t go ahead with a monopoly position.”
He also points to the heated debates, in which US entrepreneurs in particular are currently demanding research freezes on the one hand, and warning against over-regulation on the other and threatening their possible withdrawal from Europe: “What we are seeing right now is a trade war being waged with AI, and we should do our best to ensure that we have many, many companies in this area.”
Further cooperation with SAP
Feiyu Xu’s departure is a loss for SAP. The scientist, who worked at DFKI for many years, took over responsibility for AI in 2020. With the development teams, she “created both the technical and ethical foundations for embedding artificial intelligence on a broad basis in our solutions and business processes,” the group explained.
Feiyu Xu will stay until the end of September. During this time, they will be looking for a candidate for the role who meets the scientist’s “extraordinary standards,” according to an internal email.
However, Feiyu Xu should still have something to do with SAP. The software manufacturer integrates the AI models of various providers into its products. The more choice, the better. The group is looking forward to a good working relationship, “because Feiyu will remain connected to SAP in her new role.”
More: Partnership with Microsoft – SAP integrates AI language models into its products