Switzerland Unveils Open-Weight AI Model for Developers
▼ Summary
– Switzerland launched an open-source AI model called Apertus as an alternative to proprietary models like ChatGPT and Claude.
– Apertus is designed to set a new baseline for trustworthy and globally relevant open models, with its source code and training data fully available on HuggingFace.
– The model was trained on over 1,800 languages and comes in two sizes with either 8 billion or 70 billion parameters, comparable to Meta’s Llama 3 model.
– Apertus was built to adhere to EU copyright laws and voluntary AI code of practice, which some US companies have criticized as potentially curbing innovation.
– Its training data was restricted to public sources and respected website opt-out requests, avoiding any “stealth-crawling” practices.
Switzerland has introduced a new open-weight artificial intelligence model named Apertus, offering developers worldwide a transparent and legally compliant alternative to proprietary systems like ChatGPT and Claude. This initiative aims to foster innovation through accessibility while aligning with European Union copyright standards and ethical data practices.
Developed with a commitment to openness, Apertus means “open” in Latin, the model provides full access to its source code, training data, model weights, and development documentation on HuggingFace. Available in two versions with 8 billion or 70 billion parameters, it supports over 1,800 languages and is positioned as a competitive open-source counterpart to Meta’s Llama 3.
A key differentiator for Apertus is its strict adherence to EU regulations, including the voluntary AI code of practice. Unlike some U.S. firms that have expressed concerns about regulatory constraints, the Swiss team designed the model using only publicly available data while respecting website opt-out requests for AI crawlers. This approach ensures compliance and avoids unauthorized data scraping, reinforcing the project’s focus on trustworthy and responsible AI development.
(Source: The Verge)
