Odaily Planet Daily News Polyhedra partners with Berkeley RDI to launch the first verifiable zero-knowledge proof system for large language models (LLM). Just as HTTPS establishes security for internet communication, this system builds a verification infrastructure for the AI field. It efficiently verifies AI performance and compliance without exposing the model or data, achieving speeds 20,000 times faster than previous methods. Currently, it can quickly verify large language models including Llama 3-8B. Key highlights include: - Polyhedra's innovative technology: Zero-Knowledge Machine Learning (zkML) technology is akin to the HTTPS protocol in the internet, allowing verification of AI system performance and compliance without disclosing model or data details, thus ensuring privacy and security. - Outstanding performance: Expander significantly enhances verification efficiency, with single-thread verification speeds 20,000 times faster than existing methods, supporting VGG-16 at 2.2 seconds per image and Llama-3 at 150 seconds per token. - Developer-friendly: The zkPyTorch compiler seamlessly integrates with PyTorch, providing developers with user-friendly tools. Application scenarios: - AI service quality assurance - AI training data compliance - Fair data automated trading market - AI Agent accountability - AI Agent compliance - Financial service compliance verification This system not only addresses compliance challenges in AI model verification but also offers a new solution for AI safety and trustworthiness.