Self-training AI slashes costs and eliminates human feedback dependency.
DeepSeek-R1-Zero was the first large-scale AI model trained purely with reinforcement learning, eliminating the need for human feedback. This means an AI can now train itself and improve its reasoning autonomously. The impressive part was the cost. DeepSeek-V3’s last training run cost just $5.576 million, 18 times smaller than GPT-4’s supposed $100 million cost. DeepSeek’s team improved how to effectively train and manage resources through smart resource utilization, mixture of experts, and multi-head latent attention.
— DeepSeek - New Economic Rules And Regulatory Challenge · Forbes