Chinese AI manufacturer, DeepSeek, introduces its new reasoning model, DeepSeek-R1. Available to the public and optimized for commercial use, R1 surpasses OpenAI’s o1 across multiple AI benchmarks, confirms DeepSeek.
R1’s prowess lies within its functionality; it checks its logic, preventing usual model-related snags. Although reasoning models might take slightly longer to solve problems, the results they yield are often precise, especially in subjects like math, science, and physics.
R1 boasts 671 billion parameters, an unprecedented number crucial for problem-solving ability. The more parameters an AI model possesses, the better it performs. To cater to different user needs, DeepSeek offers variants of R1, with parameters varying from 1.5 billion to 70 billion. Even the smallest version can conveniently run on a laptop, while the larger ones require more robust hardware, all at a much lower cost than OpenAI’s o1.
Despite R1’s achievements, it adheres to China’s internet regulation, implying some limitations. It does not engage in discussions involving certain sensitive topics, following similar protocols of many Chinese AI models.
R1’s launch comes amidst the U.S. government’s proposed restrictions on AI technologies from Chinese ventures. But this has not dampened the spirits in the Chinese AI labs, DeepSeek, Alibaba, and Kimi, owned by Moonshot AI, who have all produced models challenging o1’s superiority.
Original source: Read the full article on TechCrunch
