Search AI Products and News

A recent paper jointly published by Tsinghua University and Shanghai Jiao Tong University challenges the widely held belief that pure reinforcement learning (RL) enhances large model reasoning capabilities. The research found that models incorporating reinforcement learning performed worse than their original counterparts in certain tasks.

At the recently concluded 18th China Electronic Information Annual Conference, the China Electronics Society announced the winners of the 2024 Science and Technology Awards. Among them, the project "Key Technologies and Applications of Secure and Trusted Dynamic Behavior in the Internet with High-Efficiency Collaboration," jointly developed by Tsinghua University, Beijing Zhongguancun Laboratory, and Ant Group, won the first prize for scientific and technological progress. This achievement not only demonstrates the enormous potential of cutting-edge technology in the field of secure and trustworthy computing, but also provides an effective solution to address the increasingly complex network environment. With the proliferation of the internet, malicious traffic attacks and covert...

Amidst the growing prevalence of artificial intelligence, the collaboration between DeepSeek and Tsinghua University has garnered significant industry attention. DeepSeek, a Chinese startup, is renowned for its breakthroughs in low-cost inference models. This collaboration aims to further reduce the training costs of AI models, thereby enhancing operational efficiency. DeepSeek recently launched a new low-cost inference model that has generated considerable market excitement. To further optimize this model, DeepSeek's research team...

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.

Recently, Professor Zhai Jidong's team from the Institute of High Performance Computing at Tsinghua University, in collaboration with Tsinghua-affiliated innovative enterprise, Qingcheng Extreme Intelligence, announced the open-sourcing of Chitu, a groundbreaking large model inference engine. This innovative technology marks another significant breakthrough for China in the AI field, particularly in inference engine development. A core highlight of the Chitu engine is its ability to natively run FP8 precision models on non-Nvidia Hopper architecture GPUs and various domestic chips.

On the morning of February 17, Tencent Holdings' stock price surged in the Hong Kong market, rising over 7.5% at one point, leading to an increase in market value of approximately 320 billion Hong Kong dollars (about 298.1 billion yuan). Related concept stocks also showed strong performance, with Weimob Group (02013.HK) up 15%, Yuanshen Group (00772.HK) up 11.97%, and Youzan (08083.HK) up 10.16%. The significant stock price increase may be related to the latest developments in WeChat Search. On February 16, Tencent announced that WeChat Search has integrated the mixed large model to provide AI search functionality.

Recently, research teams from Tsinghua University, Fudan University, and Stanford University jointly released an Agent development framework called 'Eko'. This framework aims to help developers quickly build production-ready 'virtual employees' using simple code and natural language. The Eko framework can take over a user's computer and browser, performing various tedious tasks on behalf of humans. With Eko, users can achieve automated data collection, testing, and file management, among other functionalities. For instance, users can set Eko to automatically gather information from Yahoo Finance.

With the development of digital art, automated image processing technology has received increasing attention. Recently, a research team from Tsinghua University and Tencent's ARC Lab proposed a novel image sequence coloring model called ColorFlow. This model aims to address the challenge of maintaining consistency in character and object identity while coloring black and white image sequences, fulfilling the practical needs of industries such as comics and animation. ColorFlow is a three-stage diffusion base framework that leverages contextual information by utilizing a reference image pool for black and white.

In recent years, large language models represented by ChatGPT have sparked a new wave in the AI field. These powerful language models can not only understand human instructions but also formulate plans, explore environments, and utilize tools to solve complex tasks, demonstrating immense potential in areas such as robotics, personal assistants, and process automation. However, most existing AI entity systems are passive, requiring explicit human instructions to perform tasks. To schedule a meeting, one still has to manually input the time, location, and even list out all the participants, which is quite cumbersome.

Tencent Docs has recently launched version 3.9.0, featuring a comprehensive upgrade of the AI Document Assistant to enhance user experience and work efficiency. The new version deeply integrates the three core office functions of 'reading, writing, and searching', aiming to provide a one-stop solution for users' work challenges. In terms of 'reading', the AI Document Assistant's deep reading capability has significantly improved, supporting intelligent close reading for niche scenarios. Users can upload research reports, lengthy documents, and more, and the AI assistant will quickly extract key information. The upgraded assistant supports the upload of up to 50 files at once.

Kimi Technology Co., Ltd. and Tsinghua University's MADSys laboratory have jointly released an open-source project called Mooncake, aimed at collaboratively building a large model inference architecture centered around KVCache. In June 2024, both parties released the design plan for the Mooncake inference system based on Kimi, which significantly enhances inference throughput through PD separation and a storage-computation architecture, garnering widespread attention in the industry.