Search AI Products and News

Researchers from DeepSeek and Tsinghua University recently published a new paper exploring scaling methods for reward model inference, seemingly advancing DeepSeek R2. Reinforcement learning is widely used in the large-scale post-training phase of large language models, but faces the challenge of obtaining accurate reward signals for these models. The researchers found that using pointwise generative reward modeling (GRM) improves model adaptability and scalability during inference. To this end, they propose Self-Principle Calibration Tuning (SPCT) learning.

Today, Chinese AI company DeepSeek officially refuted rumors that its next-generation AI model, DeepSeek R2, will be released on March 17th. Previous rumors circulating on X claimed DeepSeek R2 would launch in mid-March, potentially offering breakthroughs in programming capabilities, multilingual reasoning, and cost-effectiveness. However, DeepSeek's official corporate account clarified in user groups: "Rumor debunked: The R2 release is false." This definitively puts an end to the speculation.

According to recent posts on X, DeepSeek's next-generation AI model, DeepSeek R2, is potentially launching on March 17th. This news has generated significant industry buzz, with many believing it could pose a strong challenge to existing AI giants such as Anthropic's Claude Sonnet 3.7. A post by X user tanvitabs early this morning suggests DeepSeek R2 boasts breakthroughs in several key areas, including improved...