AI Scene grows: Berlin Start-up Nyonic plans Generative AI from Europe

Inhaltsverzeichnis

(Diesen Artikel gibt es auch auf Deutsch.)

A Berlin-based start-up plans to build AI foundation models from scratch and offer them to industry players. According to its website, Nyonic will develop Large Language Models (LLM) with an industrial focus and a local flavour that are multilingual to meet European use cases and whose development process up to delivery will comply with ethical and legal requirements in the European Union.

A CEO duo presides over the start-up: The business for Europe will be led by Vanessa Cann, who is still the executive director of the KI-Bundesverband until the end of June and is considered an influential AI networker through her activities in business and politics in Germany; AI researcher and serial entrepreneur Han Dong will lead the global business. Cann will give up her role at the KI-Bundesverband (Germany's largest AI network association, which aims to connect deep tech companies with politics) to devote herself entirely to Nyonic. heise online reached out to her for insights into the background of the new enterprise: US providers are strongly focused on end customers, she said. The models that Nyonic plans to develop are more intended to meet the needs of the European industry.

Foundation Models to be launched in early 2024

Currently, the team is preparing, cleaning, and merging the data to start training a basic AI model as soon as possible. Training is scheduled to begin in Q3 of 2023, with a release planned for Q1 of 2024. The company is based at the AI Campus in Berlin; Nyonic does not operate its own computing clusters. It uses the computing capacities of commercial providers available in Europe for training. "We would like to train our models ourselves on a sovereign infrastructure," Cann explained, "and with the availability of GPUs, we notice how dependent a European company is on US hyperscalers." If Europe is to succeed in developing powerful models, a robust ecosystem of industrial companies, AI start-ups, investors, academia and the public sector is needed. The start-up Nyonic is a member of the German AI Association and will therefore "actively support" the LEAM initiative in the future, according to Cann (LEAM stands for Large European AI Models).

Founding team knows SAP, OpenAI and Palantir from close up

The five-member founding team of Nyonic unites other well-known personalities in the German AI scene: For example, it was announced at the time of the founding that Dr Feiyu Xu, currently still Global Head of AI at SAP, will join Nyonic as Chief Innovation Officer in October 2023. She has 20 years of AI experience, had most recently designed the AI strategy at SAP and is considered an expert in business-driven AI applications.

The technical development is led by AI researcher Dr Johannes Otterbach, who heads the AI unit at the Berlin-based company Merantix Momentum. Otterbach had worked on the API at OpenAI in the days of GPT-3 and also worked for Palantir, among others. As an academic researcher, he had mainly worked on theoretical physics and quantum computing. The combination of quantum computing and machine learning could point the way for future progress in the field and become tangible for industrial application in the near future (from 2025/ 2028).

Discussion with potential investors ongoing

According to Vanessa Cann, Nyonic is currently in talks with potential investors. The interest is great. Three of the five founding members have previously founded start-ups (Dong, Uszkoreit and Xu). CTO Otterbach was involved in building OpenAI and is said to have a comprehensive understanding of modern AI technology and how the ChatGPT provider works.

Since the founding team is partly active in the German AI Association and the company itself is a member there, it was obvious to ask about LEAM, the initiative for cooperation on large-scale AI models in Europe launched by the association, whose goal is, among other things, dedicated AI computing infrastructure in Europe (the Hamburg-based open-source network LAION, for example, is pursuing a similar goal with proposals for international supercomputing centers, but rather for open source and research-oriented large-scale AI models).

In a public statement, Thomas Saueressig, a member of the Executive Board at SAP, expressed regret at the loss of Dr Feiyu Xu to Walldorf and at the same time emphasized the importance of a vital start-up scene and cross-industry cooperation. At the end of May, SAP announced at its annual in-house conference SAPPHIRE in Orlando that it would cooperate with a number of other providers of foundation models in addition to Microsoft. Among others, Databricks and the German AI company Aleph Alpha were named. At a press briefing last week, it was said that SAP has about 80 Foundation Model projects underway with different partners and use cases.

European AI industry: competition and common goals

According to Vanessa Cann, Nyonic has "many new focal points" that distinguish the start-up from other companies. At the same time, there are interests overlapping with other projects. Since foundation models will transform almost all areas of society and the economy, "even five or ten companies in this field are still too few," she stressed.

European sovereignty is the overriding common concern, so they will "openly enter into talks with players in the European ecosystem" about possible cooperation. What unites the existing providers is the common goal of freeing Europe from its growing dependence on a few US providers. In Europe, European jurisdiction, copyright law, and data protection must be considered, posing particular challenges for companies. Developing industry models is the focus, she said. In addition to the usual big data sets, which come from similar sources for all providers, Nyonic will "include additional and different data" to, in Cann's words, cater to the multicultural and multilingual European society and industry

European sovereignty: getting down to business

The scientific leadership at Nyonic is provided by the fifth co-founder, Professor Hans Uszkoreit of the German Research Centre for Artificial Intelligence (DFKI), a computational linguist with more than thirty years of experience in AI. According to the Nyonic website, he continues to work part-time in an advisory capacity for the Berlin office of the DFKI. Uszkoreit is a member of the European Academy of Sciences and has long warned that Europe needs to get into action on generative AI, particularly the basic technology of foundation models, as he had expressed at the Rise of AI conference in Berlin in May.

In his lecture, Uszkoreit had pointed out the looming dependence on Chinese or US foundation models and identified Aleph Alpha as the only provider of basic AI technology from Germany at present, and a few isolated offers for Europe as a whole, such as the French model BLOOM of the BigScience initiative (which in turn is not a purely European project). The words are now apparently being followed by deeds in the form of a company foundation, which can be described as courageous given the forthcoming regulation of primarily generative AI in Europe through the AI Act. Jörg Bienert, board member of the German AI Association, welcomes the new company as a contribution to digital sovereignty in Europe.

Whether there are conflicts of interest in the case of Nyonic due to the close intertwining of association activities and economic activity of the founding members came up as a question from the open-source environment in internet comments. The editors are not aware of anything to this effect currently. Further information, the mission statement and more details on the professional and scientific background of the founding team, as well as on their voluntary activities, can be found on the Nyonic website.

(sih)