Switzerland releases open-source AI model trained only on public data
What's the story
Switzerland has launched an open-source artificial intelligence (AI) model called 'Apertus.' The new model offers an alternative to proprietary models like OpenAI's ChatGPT and Anthropic's Claude. The entire source code, training data, model weights, and development process of Apertus are available on HuggingFace, a popular platform for sharing AI models.
Model features
Apertus trained on over 1,800 languages
The name "Apertus," which means "open" in Latin, reflects the model's purpose. Its developers say it aims to "set a new baseline for trustworthy and globally relevant open models." The model has been trained on more than 1,800 languages and comes in two sizes with either 8 billion or 70 billion parameters. Notably, Apertus is said to be comparable to Meta's 2024 Llama 3 model.
Compliance
Apertus complies with European Union copyright laws
Apertus has been built to comply with European Union (EU) copyright laws and a voluntary AI code of practice. This is in stark contrast to some US-based AI companies that have reluctantly signed the regulations, fearing they may stifle innovation and deployment in the field. The developers behind Apertus ensured that its training data was limited to public sources only, respecting "AI crawler opt-out requests" on certain websites.