Switzerland has introduced an open-source AI model called Apertus, which aims to serve as an alternative to proprietary models such as OpenAI’s ChatGPT and Anthropic’s Claude. The source code, training data, model weights, and detailed development process of Apertus are accessible on the AI model platform HuggingFace.
The name “Apertus,” which means “open” in Latin, reflects its intended purpose. Developers state that the model aspires to establish a new standard for trustworthy and globally relevant open AI models. Apertus has been trained in over 1,800 languages and is available in two configurations, with either 8 billion or 70 billion parameters. It is considered comparable to the upcoming 2024 Llama 3 model from Meta.
Constructed with compliance to the European Union’s copyright laws and voluntary AI code of practice, Apertus addresses guidelines that some US-based AI companies have reluctantly accepted, citing concerns that these regulations may hinder AI innovation and deployment. The training data for Apertus was exclusively sourced from public datasets, and the model complies with AI crawler opt-out requests from certain websites, avoiding “stealth-crawling” methods.
This launch highlights the ongoing global efforts to create transparent, open-source AI alternatives within regulatory frameworks.
Source: https://www.theverge.com/ai-artificial-intelligence/770646/switzerland-ai-model-llm-open-apertus