DNB Contributes to Dutch Language Model GPT-NL

DNB provides public information from dnb.nl to build a new AI language model, GPT-NL. The project is an initiative of non-profit organisations TNO, NFI and SURF and offers a responsible alternative to existing language models. The model is being developed for the Dutch language and culture, based on qualitative Dutch data legitimately obtained. DNB is one of the Dutch data suppliers.

A transparent and responsible language model

The language model will be developed as openly and transparently as possible. Qualitative, Dutch data is chosen, and data is only used if it is legitimately obtained. The creators are transparent about which training data are used, and decisions and considerations are communicated openly. Part of the proceeds flow back to the copyright holders. With the introduction of GPT-NL, the Netherlands will have its own language model and ecosystem.

DNB provides public data

This initiative offers DNB an opportunity to actively contribute to the development of a responsible language model. DNB is contributing from within the financial sector by making public data from dnb.nl available. A Dutch language model is of great added value because it can better respond to the nuances and specific characteristics of the Dutch language and culture. This can lead to more accurate and relevant applications in various sectors, such as finance, education and government.

Planning of the GPT-NL project

The GPT-NL project has two phases. The language model is currently being developed. In the second phase, starting in the second quarter of 2025, the model will be trained. In the fourth quarter of 2025, the model will be further improved and work on its use.

More info: interview

Source (in Dutch): DNB

iStock credits: tondone