Swiss Researchers Launch Apertus: A Transparent, Privacy-Focused AI Alternative
Swiss researchers have launched Apertus, a suite of language models aiming to provide a transparent and privacy-focused alternative to commercial systems like ChatGPT. The project, developed by the Swiss AI Initiative, prioritizes multilingualism and data protection.
Apertus is the result of a collaboration between ETH Zurich, EPFL, and the Swiss National Supercomputing Centre (CSCS). Founded by Adrian Locher, also co-founder of Merantix, the Swiss AI Initiative is behind this innovative project. Key institutions involved include ETH Zurich and EPFL, renowned for their AI research.
To minimize data reproduction, researchers employed the 'Goldfish objective' during training. Apertus also uses a unique alignment approach based on the 'Swiss AI Charter', with another model serving as an 'LLM-as-judge' for controversial topics.
In terms of performance, Apertus-70B-Instruct shows solid results but trails leading open-weight models in most general categories. However, it excels in multilingual capabilities, with training data sourced from over 1,800 languages, including Swiss national languages like Romansh and Swiss German.
Apertus is available in 8 and 70 billion parameter sizes, aiming not to compete with leading AI companies' billion-dollar budgets but to create a secure and accessible AI system for science and business.
Industry representatives, such as the Swiss Bankers Association, see great long-term potential in Apertus, particularly in complying with local data protection and banking secrecy laws. Apertus' central feature is full transparency, making all development cycle artifacts available. Trained exclusively on publicly available data, Apertus respects website crawler restrictions and filters out copyrighted, non-permissive, toxic, and personally identifiable content.
Read also:
- AI Validators: Unsung Heroes Ensuring Trust and Compliance
- GEW Demands Survey and Funding for Hessian Schools' Renovation
- India's Sports Revolution: Bhubaneswar, Chennai, and Ahmedabad Pioneering Nation's Athletic Makeover
- "Blood tests could potentially enhance the accuracy of malaria diagnoses in research circumstances"
