Search AI Products and News

DeepSeek AI, a prominent Chinese artificial intelligence research lab, following its powerful open-source language model DeepSeek-R1, has achieved another significant breakthrough in the field of Large Language Models (LLMs). Recently, DeepSeek AI officially launched an innovative technology called Self-Principled Critique Tuning (SPCT), aimed at building more general-purpose and scalable AI reward models.

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.

The Alibaba Qwen team recently published a paper titled 'Lessons Learned from the Development of Process Reward Models in Mathematical Reasoning' and introduced two new models in the Qwen2.5-Math-PRM series, featuring 7B and 72B parameters respectively. These models break through the limitations of the existing PRM framework in mathematical reasoning, significantly improving the accuracy and generalization ability of reasoning models through innovative techniques. Mathematical reasoning has long been a major challenge for large language models (LLMs), especially regarding errors in intermediate reasoning steps.

Kunlun Wanwei Technology Co., Ltd. recently announced that its two new reward models, Skywork-Reward-Gemma-2-27B and Skywork-Reward-Llama-3.1-8B, have performed exceptionally well on the internationally recognized reward model evaluation benchmark RewardBench. Notably, the Skywork-Reward-Gemma-2-27B model achieved the top position and received high recognition from the RewardBench officials.

The Google DeepMind team, in collaboration with academic institutions, has developed an innovative method called Generative Reward Model (GenRM) aimed at improving the accuracy and reliability of generative AI in reasoning tasks. GenRM incorporates a validation process into text generation tasks, allowing the model to simultaneously generate and evaluate potential solutions while supporting Chain of Thought (CoT), enhancing the comprehensiveness of the validation process. Compared to traditional methods, GenRM has shown significant advantages in multiple tests, with accuracy improvements ranging from 16% to 64%, particularly in...