GPT-NL, the Dutch language model aims to strengthen European digital sovereignty
GPT-NL, the Dutch language model aims to strengthen European digital sovereignty
The Netherlands has initiated the project GPT-NL, an independent language model developed to ensure the digital sovereignty of the country and of Europe, placing concrete emphasis on the protection of public values. The initiative, supported by the Dutch government, aims to forge a credible alternative to non-European Large Language Models (LLMs), focusing on transparency, reliability, and ethical data management.
Language-based artificial intelligence has become an unavoidable element in every field, from the workplace to education, and through public services. This pervasiveness raises high-profile questions: Who decides how these models operate? What data do they use? And how are fundamental values such as privacy, copyright, and transparency protected? The increasing dependence on external providers outside Europe represents a strategic weakness that GPT-NL seeks to address.
The consortium driving the project includes TNO (The Dutch Organization for Applied Scientific Research), SURF (the ICT cooperative for Dutch education and research), and the Netherlands Forensic Institute (NFI). The financing, amounting to 13.5 million euros, comes from the Netherlands Enterprise Agency (RVO) on behalf of the Ministry of Economic Affairs and Climate Policy. This public investment underscores the importance placed on an independent Dutch language model that is future-proof.
GPT-NL, the language model that aims for European digital sovereignty
The creation of GPT-NL is based on well-defined pillars, aimed at building a language model that is not only high-performing but also deeply rooted in the Dutch context and values. Sovereignty is at the forefront of this initiative. Developed entirely in the Netherlands and Europe, GPT-NL ensures total control over the model, the data used, and the implementation choices. This approach allows for avoiding dependency on non-European providers and investing in a sustainable AI ecosystem, fully aligned with the laws, values, and social goals of the region.
In terms of openness and transparency, the project commits to exhaustively documenting the decisions made during the data collection and training phases. It proactively addresses risks related to biases and ethical concerns. The source code will be published as open source, and detailed information about the dataset will be shared. However, the model's weights will be made available under a controlled license.
The reliability of GPT-NL is ensured by training the model from scratch. This approach prevents the inheritance of issues related to the uncertain origins of data, copyright risks, or the presence of personal data from pre-existing models. Data collection adheres to stringent criteria: safeguarding intellectual property, removing and anonymizing personal data before training, excluding confidential information and harmful content, and eliminating duplications within the dataset.
The principle of reciprocity translates into a legal and clean data supply chain. GPT-NL collaborates closely with data providers, actively involving them in developing the model. A Content Board offers these providers and rights holders the opportunity to voice their opinions on the future of GPT-NL. A part of the revenue is intended to return to the creators, establishing a fairer innovation model, where value is shared rather than simply extracted.
Attention to resource efficiency is another fundamental aspect. The development of AI requires significant computational and energy resources, which is why the project actively focuses on energy efficiency and responsible resource use.
GPT-NL aims to demonstrate that generative artificial intelligence can coexist with public values. The goal is to build technology that makes the Netherlands stronger, more autonomous, and fairer, showing that it is possible to develop high-profile models with a fraction of the typical budget of large Silicon Valley companies. However, considering the multi-billion-dollar investments of overseas giants compared to the mere millions invested by the Dutch, this appears to be a truly tough battle for the European initiative.