Lates News

On September 17th, the team led by Liang Wenfeng published a paper in the journal "Nature" introducing the training method of the open-source AI model DeepSeek-R1 which utilizes a large-scale inference model. The study shows that training large-scale inference models through pure reinforcement learning can effectively improve the reasoning ability of large language models and reduce the need for human input. The model uses a reward mechanism in reinforcement learning to reduce training costs and complexity in solving problems.