This Startup Wants to Spark a US DeepSeek Moment

DeepSeek's January breakthrough sparked momentum around open-source Chinese AI models, prompting startup Prime Intellect to develop a new approach for building competitive AI without relying on big tech. The company is training INTELLECT-3, a frontier language model using distributed reinforcement learning that allows people worldwide to contribute specialized training environments for different tasks like math or legal questions. This democratized approach addresses the current AI landscape where researchers must choose between closed US models or open Chinese alternatives, with reinforcement learning becoming the key bottleneck for scaling AI capabilities.