Silo AI launches SiloGen – a world-class generative AI team for Europe’s digital sovereignty

Today, Silo AI, one of Europe’s largest private AI labs, is announcing SiloGen, a large-scale initiative on generative AI and LLMs. With a focus on more accurate, trustworthy and robust downstream applications of LLMs, SiloGen operates as a subsidiary of Silo AI and is run by Silo AI CEO and co-founder Peter Sarlin.

• Silo AI launches a large-scale initiative SiloGen with the aim of building generative AI technology for European digital sovereignty.
• In only five years, Silo AI has grown to be one of Europe’s largest private AI labs with a team of 250 leading AI experts and approximately 120 PhDs.
• With SiloGen, Silo AI is gathering some of Europe’s leading generative AI and Large Language Model (LLM) experts, as well as access to data sources, powerful computational resources and infrastructure to train, run and operate LLMs.
• SiloGen has been operational since late 2022 and is currently working on its technology at full speed with clients to improve downstream and domain-specific applications and to ensure companies can utilize trusted models for private, confidential and proprietary data.

The unique initiative gathers together Europe’s leading experts in language technology and multi-modal AI with a pedigree from leading companies and universities like Google, Meta, Apple, Stanford University and Columbia University, as well as LLM initiatives like LLM training on Europe’s most powerful supercomputer LUMI, AI Sweden's GPT-SW3 and University of Turku's FinGPT. SiloGen benefits from Silo AI's existing engineering team and taps into a group of more than 120 PhD-level AI scientists, such as Principal AI Scientist and University of Turku professor Filip Ginter.

Recent months have shifted close to all attention in AI to generative technologies, and particularly US-based initiatives on LLMs, such as those by OpenAI, Microsoft, Google and Meta. With a continuously widening gap to the forerunners, initiatives on European LLMs are imperative to ensure Europe’s digital sovereignty in AI and downstream applications, as well as overall market independence. Most importantly, we need independent LLMs that can be finetuned and instructed to be optimal for a wide range of applications and eventually allow for the processing of private, proprietary and confidential data.

Peter Sarlin, CEO and co-founder of Silo AI, says: ‍

“The development of independent, trusted and robust foundation models aligned with European values is imperative for Europe’s digital sovereignty. Generative AI is much more than an LLM with linguistic precision, as we need to ensure that the underlying data, and hence knowledge, is representative of the citizens and organizations in the region. And ultimately we need sovereignty on how downstream applications and value creation eventually happen. This requires trusted and secure approaches to independent foundation models and generative AI that enables finetuning models for domain-specific, bespoke needs.”

Tero Ojanperä, Executive Chairman and co-founder of Silo AI, says:

‍“In the past few months, most companies have had to re-evaluate how they leverage recent developments in AI. With SiloGen, we aim to not only contribute to European digital sovereignty but also help ensure that our clients succeed in tapping into recent developments in generative AI. I believe all companies, including Nordic and European, should consider how they best create sustained value with generative AI, while also ensuring they utilize trusted models for private, confidential and proprietary data.”

SiloGen has been operational since late 2022 and is currently at full speed working together with clients on a number of key initiatives to improve downstream applications and increase value creation with LLMs. SiloGen will release further details on its technology later in 2023.

A world-class generative AI team to develop Europe’s digital sovereignty

SiloGen is joined by some of the key contributors from the two first and currently only LLM initiatives in the Nordics. Dr. Magnus Sahlgren, who led the development of GPT-SW3 at AI Sweden and RISE, has taken a part-time role as a Principal AI Scientist at Silo AI. With AI Sweden, he led the team that recently published the GPT-SW3 LLM for the Scandinavian languages.

SiloGen also has strong ties to University of Turku’s TurkuNLP research lab through the long-time Silo AI employees Professor and TurkuNLP lab leader Filip Ginter and one of the contributors to FinBERT Jouni Luoma, as well as a mutually beneficial industry-academia collaboration with Associate Professor Sampo Pyysalo, the principal investigator behind the FinGPT family of models. The lab has published multiple LLMs, such as the first transformer-based Finnish model FinBERT and a Finnish generative model FinGPT, and is part of the EU-funded High-performance Language Technologies project with the aim to introduce open LLMs for all EU languages.

Given LLMs’ massive computing requirements, the SiloGen team is uniquely positioned with experience in training LLMs on and building infrastructure for the LUMI supercomputer, which is ranked as Europe’s most powerful and the world's third most powerful supercomputer. For the AMD-based compute hardware, the team has already built a unique infrastructure to train, run and operate LLMs at scale in production. Part of the team also has experience from the LUMI pilot project FinnGen and the LUMI Extreme Scale Project Foundation for Finnish AI.

The rest of the SiloGen team includes senior machine learning, natural language processing, computer vision and multi-modal AI experts, covering new and old Silo AI employees, with a pedigree from organizations like Google, Meta, Apple, Spotify, Gavagai, IBM, Nokia, Fujitsu, Xerox PARC, Yahoo! Research, Stanford University, Idiap Research Institute and Columbia University. The team is located in Helsinki, Stockholm and Turku, and includes leading experts like Niko Vuokko, Jaakko Vainio, Jussi Karlgren, Fredrik Espinoza, Aarne Talman, Vien Nguyen, Shantipriya Parida, Antti-Ville Suni, Johanna Malinen and Farrokh Mehryary.