Nov 23, 02:43 PM

Prime Intellect Completes Decentralized Training of 10B Model Across 3 Continents; Stanford Researchers Develop AI Agents with 85% Accuracy

AI startup Prime Intellect has successfully completed the first decentralized training of a 10-billion-parameter language model. This significant achievement involved collaboration across computers located in the United States, Europe, and Asia. The model's training marks a milestone in the development of large language models, leveraging distributed computing resources to enhance performance and efficiency. In a separate development, researchers from Stanford University, Washington University, and Google DeepMind have created AI agents capable of predicting human behavior with 85% accuracy in social experiments, showcasing advancements in AI's ability to mimic human interactions.