Apertus-15, a project from ETH Zurich's AI Center, is developing open-source AI infrastructure focused on transparency and trustworthiness. Unlike frontier models from private companies, Apertus aims to provide a foundation that organizations can confidently build upon. The project emphasizes open access to its source code, training data, deployment methods, and model weights. Apertus's training methodology includes using exclusively open data, respecting privacy and toxicity filters, and employing the "Goldfish objective" to prevent data memorization while maintaining performance. A significant portion of its training data is also allocated to non-English content. AI
IMPACT Promotes transparency and trustworthiness in AI development, offering an alternative to proprietary frontier models for broader adoption.
RANK_REASON The item describes a research project focused on open-source AI infrastructure and training methodologies. [lever_c_demoted from research: ic=1 ai=1.0]
Read on Mastodon — fosstodon.org →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
