We’re launching Lyria 3.5 in Google Flow Music, with advances across musicality, lyrics, vocals, and creative control
deepmind.google·6d ago
AI Engineering
Hacker News·Jun 21·5 min read
Apertus – Open Foundation Model for Sovereign AI
TL;DR
AI models often lack transparency and compliance with regulations. Apertus is a fully open foundation model developed by the Swiss AI Initiative, featuring open weights, data, and methods.
✦ Why It Matters
Engineers can leverage Apertus to develop compliant AI applications that prioritize transparency and user privacy.
Key Takeaways
- Explore Apertus for compliant AI development.
- Utilize open weights for reproducibility.
- Incorporate multilingual capabilities in projects.
- Ensure data privacy by removing PII.
Full Summary
AI models frequently face criticism for their opacity and inability to comply with regulations like the EU AI Act, which mandates data protection and user opt-outs. Apertus, developed by the Swiss AI Initiative in collaboration with EPFL, ETH Zurich, and CSCS, addresses these issues by providing a fully open foundation model.
It includes open weights, data, and alignment principles, ensuring reproducibility and transparency. The model is available in two sizes, 8 billion and 70 billion parameters, and has been trained on over 1000 languages from the outset.
Key features include the removal of personally identifiable information (PII) and mechanisms to prevent memorization of sensitive data. As a result, Apertus serves as a global foundation for building compliant AI applications.
This initiative positions engineers and researchers to create more responsible AI systems.