A European answer to large AI models like GPT-3 in order not to fall behind in the international AI competition - this is precisely the goal of the lighthouse project "Open GPT-X", which we won together with our nine project partners. In this blog post, we answer all the important questions about the initiative and classify its significance in the international context.
Reaching the goal together
Together with our project partners Fraunhofer Institute for Intelligent Analysis and Information Systems & Fraunhofer Institute for Integrated Circuits, 1&1 IONOS SE, Forschungszentrum Jülich GmbH, Technische Universität Dresden, Deutsches Forschungszentrum für Künstliche Intelligenz GmbH, Aleph Alpha GmbH, Control Expert GmbH, Westdeutscher Rundfunk and KI Bundesverband e.V., we applied for the funding call "Innovative and practical applications and data spaces in the Gaia-X digital ecosystem" - and were selected from a large number of consortia that applied. Associated partners are: BMW, Institut der Deutschen Wirtschaft, Aalto University, OPNTEC GmbH, Ströer SE & Co. KGaA, Eclipse Foundation, Covestro Deutschland AG, eco Verband der Internetwirtschaft e.V. (Service-Meister project) and Siemens. The project, for which the consortium may now submit a funding application and which has a term of three years, is expected to begin on 01.11.2021.
Gaia-X: Hyperscaler made in Europe
The project builds on the Gaia-X initiative, which was launched by the Federal Ministry for Economic Affairs and Energy. Gaia-X is to be understood as a European alternative to infrastructures such as AWS and Azure. Funded by the BMWi, standards for distributed high-performance centres are to be developed. This forms the basis for companies to develop their services in a European cloud - which has several advantages. On the one hand, there is less dependence on American providers, and on the other hand, it is easier for companies to migrate their applications to other providers. To achieve this goal, funding totalling around 175 million euros will be made available until the end of 2024 to support 16 lighthouse projects.
Open GPT-X: the European answer to GPT-3
In the "Open GPT-X" project, large AI language models are to be developed that are compatible with Gaia-X as Advanced Smart Services. This will enable data-based business solutions to be used for practical applications in the areas of mobility, media and finance/insurance. The language models are based, among other things, on GPT-3 (Generative Pretrained Transformer 3), the mega-model from the USA. With over 175 billion parameters, this was presented in August 2020 as the largest neural network in the world to date, but was replaced a few weeks ago by a model 10 times larger. Wu Dao 2.0 from China. Models of this scale will enable language processing applications such as answering questions, writing texts or creating code. "OpenGPT-X" is thus intended to become a Gaia-X node for large AI language models and innovative language application services.
Alexander Thamm GmbH builds data management and MLOps
A particular challenge for training such large AI models is data management and the establishment of stable and continuous operations models. Here, Alexander Thamm GmbH brings its experience from over 1000 data and AI projects to the consortium and will implement powerful processes and applications for data management and the automation of model creation.
A digital sovereignty for Europe
"Open-GPT-X" is more than a project that will enable groundbreaking advances in speech processing. It is the European answer to powerful models like GPT-3 from the US and Wu Dao 2.0. from China. Because Europe needs to build digital sovereignty to avoid becoming dependent on AI superpowers and falling behind in international competition. That is why we are pleased to take a first important step in this direction through the funding project together with our project partners.
We also support the initiative LEAM (Large European AI Models) of the German AI Association, which was recently presented in Brussels, through the project. This initiative also pursues the goal of uniting business, research and science in order to be able to develop competitive large European AI models with sufficient clout.