Silo AI: Why Körber, Philips or Unilever work with them?

Silo AI and TurkuNLP are developing a family of multilingual open
source LLMs with the aim of strengthening Europe's digital
sovereignty and democratising access to LLMs. Developing baseline
models that are in line with European values is crucial to this
effort to ensure that they are based on data and information that
accurately represents the different languages, citizens,
organisations and cultural landscape of the European Union. This
approach is not only in line with European values, but also enables
sovereignty over how downstream applications and value are created.
The success was attributed to the combination of the low-resource
Finnish language with resource-rich languages. The team worked to
determine the optimal frequency of data reuse for low-resource
languages during training, and integrated translated text pairs
from English and Finnish. This strategy, which relies on a
cross-linguistic signal to improve the model's understanding of the
relationships between the languages, proved crucial in achieving
excellent performance in the low-resource languages without
compromising performance in English.