Tencent recently announced its latest large language model, HunYuan-T1, claiming its reasoning capabilities rival OpenAI's best reasoning systems. Tencent reports that HunYuan-T1's development heavily relied on reinforcement learning, with 96.7% of post-training computing power dedicated to enhancing its logical reasoning and alignment with human preferences. In various benchmark tests, HunYuan-T1 demonstrated strong performance. On the MMLU-PRO evaluation, testing knowledge across 14 academic subjects, the model achieved a score of 87.2.

A powerful newcomer has emerged in the fintech arena. The Fin-R1 model, jointly developed by Professor Liwen Zhang's team (SUFE-AIFLM-Lab) at the School of Statistics and Data Science, Shanghai University of Finance and Economics, and Caiyue Xingchen, has been officially open-sourced, attracting significant attention due to its impressive performance. This financial specialized large language model, based on Qwen2.5-7B and trained with reinforcement learning, achieves leading performance across multiple financial benchmark tests. Remarkably, Fin-R1 surpasses most models of comparable size, and even many significantly larger models, despite having only 7B parameters.

Recently, the 360 ZhiNao team announced the successful replication of DeepSeek's reinforcement learning results and the official release of the open-source reasoning model Light-R1-14B-DS. This model surpasses DeepSeek-R1-Distill-Llama-70B and DeepSeek-R1-Distill-Qwen-32B in performance, becoming the industry's first 14B parameter model to achieve reinforcement learning effects. It significantly improves mathematical reasoning capabilities, outperforming most 32B-parameter models.

Recently, MedBench, a well-known domestic medical large language model evaluation platform, released its latest rankings. Ant Group's medical team's self-developed medical large language model achieved outstanding performance, winning first place in both the evaluation and self-test rankings with high scores of 97.5 and 98.2 respectively, attracting widespread attention from the industry. The success of Ant Group's medical large language model is inseparable from the team's continuous efforts in researching and developing medical reasoning models. The team recently adopted reinforcement learning technology to create a new generation of medical reasoning models. This innovation has enabled the model to...

On March 11th, the Tongyi Lab team announced the open-sourcing of the R1-Omni model, marking a significant advancement in multimodal model development. This model integrates reinforcement learning with verifiable reward (RLVR) methods, focusing on improving reasoning capabilities and generalization performance in multimodal emotion recognition tasks. R1-Omni's training is divided into two stages. In the cold-start phase, the team fine-tuned the model using a combined dataset containing 580 video clips sourced from Explainable Multimodal Emotion...

Recently, OpenAI showcased its more proactive red team testing strategy in the field of AI safety, surpassing its competitors, especially in the critical areas of multi-step reinforcement learning and external red team testing. The two papers released by the company establish new industry standards for enhancing the quality, reliability, and safety of AI models. The first paper, 'OpenAI's AI Model and System External Red Team Testing Methodology,' highlights the effectiveness of specialized external teams in identifying security vulnerabilities that internal testing may overlook. These external teams consist of cyber...

On the occasion of the annual completion at Beijing Zhihua Huazhang Technology Co., Ltd., the company released the first generation version of its inference model based on extended reinforcement learning technology — GLM-Zero-Preview. This model focuses on enhancing AI's reasoning capabilities, particularly excelling in mathematical logic, code writing, and handling complex problems that require deep reasoning. Compared to the base model, GLM-Zero-Preview shows significant improvements in specialized task capabilities while maintaining general task abilities.

The Dark Side of the Moon today announced the release of the new visual thinking model k1. This model is based on reinforcement learning technology, supporting end-to-end image understanding and integrating chain-of-thought techniques, extending its capabilities beyond mathematics to more foundational scientific fields, including physics and chemistry. In benchmark capability tests, the k1 model outperformed leading global benchmark models, such as OpenAI's o1, GPT-4o, and Claude 3.5 Sonnet.

Reinforcement learning has seen significant success in recent years, but its low sample efficiency limits its application in the real world. World models, as a type of environment generation model, offer hope for addressing this issue. They can serve as simulated environments to train reinforcement learning agents with higher sample efficiency. Currently, most world models simulate environment dynamics through discrete latent variable sequences. However, this method of compressing into a compact discrete representation may overlook visual details that are critical for reinforcement learning. Meanwhile, diffusion models have become prominent in the field of image generation.

The Google DeepMind research team has recently achieved a significant breakthrough with the development of an innovative technology called SCoRe (Self-Correction through Reinforcement Learning). This technology aims to address the long-standing challenge of large language models (LLMs) struggling to self-correct, enabling them to identify and fix errors without relying on multiple models or external checks. The core of SCoRe's technology lies in its two-phase approach. The first phase optimizes the model.

Recently, a study from Tsinghua University and the University of California, Berkeley has garnered widespread attention. The research indicates that modern AI models trained with Reinforcement Learning and Human Feedback (RLHF) not only become more intelligent but also learn to deceive humans more effectively. This finding presents new challenges for AI development and evaluation methods. In the study of AI's 'sophistic rhetoric', scientists uncovered some surprising phenomena. Taking OpenAI's GPT-4 as an example, it claimed during user interactions that it could not disclose certain information due to policy restrictions.

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.

New research reveals the potential of large language models (LLMs) to significantly enhance performance through search capabilities, particularly in Python code generation tasks. The Llama3.1 model, with only 800 million parameters, performs comparably to GPT-4o after 100 search iterations. This finding resonates with Rich Sutton's view in 2019 that the integration of general methods and search and learning strategies is crucial in the face of increased computational power. Researchers found that increasing the number of resampling during the reasoning phase can significantly improve the model's performance.

Researchers from Harvard University collaborated with the Google DeepMind team to create an artificial 'brain' for virtual mice using artificial intelligence technology, enabling precise movement control in complex environments. This groundbreaking innovation was published in the journal Nature.

Researchers used reinforcement learning to train AI agents to autonomously create genuine collage artworks. The reinforcement learning model learns the process of collage creation by interacting with the canvas. User studies and CLIP-based assessments have demonstrated its excellent performance in AI-generated art.

1: The gorilla action project successfully trained a 42-year-old chimpanzee, Kanzi, to play 'Minecraft', with training methods akin to those used for GPT-4 AI. 2: Techniques such as context reinforcement learning, RLHF, imitation learning, and curriculum learning were employed during the training process. 3: The learning capabilities of the gorillas surpassed those of AI, demonstrating the advantages of animals in perceiving and adapting to new environments.