Search AI Products and News

Google announced the global launch of its latest video generation model, Veo3. This long-anticipated release has generated great excitement among users, as Veo3 is now available to Gemini users in over 159 countries, offering a new video creation experience. The key feature of the Veo3 video generation model is its ability to generate videos up to eight seconds long based on simple text prompts. According to Google, this technology is designed for creative users, especially those on social media who increasingly demand short-form content.

Recently, a research team from Renmin University, Shanghai Artificial Intelligence Laboratory, University College London, and Dalian University of Technology revealed an important finding in the reasoning process of large models: when the model is thinking, the 'thinking words' it uses actually reflect a significant increase in its internal information. This research result provides a new perspective for better understanding the reasoning mechanisms of artificial intelligence through methods of information theory. You may have seen large models output some language that seems human-like when answering questions, such as "Hmm..." or "Let me think...".

In the field of artificial intelligence, reward models are a critical component for aligning large language models (LLMs) with human feedback, but existing models face the issue of "reward hacking." These models often focus on superficial features, such as the length or format of responses, rather than identifying genuine quality metrics, such as factual accuracy and relevance. The root cause lies in standard training objectives failing to distinguish between spurious associations and true causal drivers present in the training data. This failure leads to fragile reward models (RMs), which generate misaligned policies.

On July 4, 2025, Kunlun Xiwang continued to open-source the second-generation reward model Skywork-Reward-V2 series. This series includes 8 reward models based on different foundation models, with parameter sizes ranging from 600 million to 8 billion. Upon its release, it won all seven major reward model evaluation rankings, becoming a focus in the open-source reward model field. Reward models play a key role in the reinforcement learning from human feedback (RLHF) process. To build the next generation of reward models, Kunlun Xiwang has constructed a dataset containing 40 million

According to KPMG China's recent report, "The First 50 Health Tech Companies," China accounts for more than 70% of the global release volume of medical large models. This data not only demonstrates China's rapid development in the field of intelligent healthcare, but also reflects the wide application of large language models in the healthcare industry. The report points out that about 65% of the currently released medical large models are large language models. These models can process and generate natural language, playing a significant supporting role in the analysis of medical data, patient communication, and scientific research.

In the long-standing copyright infringement lawsuit filed by The New York Times against OpenAI, the case has made significant progress. According to Ars Technica, the federal judge presiding over the case has authorized The New York Times and its co-plaintiffs, The New York Daily News and the Investigative Reporting Center, to access OpenAI's user logs, including deleted content, to accurately determine the scope of the infringement. The New York Times believes that ChatGPT users may delete their history after bypassing the paywall, and therefore it is necessary to conduct large-scale data collection.

In the new energy vehicle market, Xiaopeng Automotive has once again drawn attention. On July 3rd, the Xiaopeng G7 Ultra was officially launched, becoming the first intelligent vehicle equipped with the local-end "VLA+VLM" large model. This innovative technology marks an important step forward for Xiaopeng in the field of intelligent driving. The Xiaopeng G7 Ultra is equipped with the VLA (active thinking and rapid decision-making capability) large model, making the driving experience more intelligent. In daily driving, the G7 Ultra can flexibly handle various complex driving scenarios, such as in traffic.

A recent report titled "Health Tech 50 - The First Edition" released by KPMG China reveals that China has taken a leading position in the field of medical large models globally. The report indicates that the number of medical large models launched in China accounts for more than 70% of the global total, far surpassing other countries and regions. In terms of model categories, large language models (LLMs) are the most numerous, accounting for nearly 65%. Moreover, the report also highlights the strong growth momentum of the intelligent medical devices market in China. It is expected that by 2025, the scale of the intelligent medical devices market in China will reach 24.23 billion yuan, and it will continue to grow.

Large Language Models (LLMs) have achieved significant progress in complex reasoning tasks by combining task prompts with large-scale reinforcement learning (RL), as demonstrated by models like Deepseek-R1-Zero, which directly apply reinforcement learning to base models, showcasing strong reasoning capabilities. However, this success is difficult to replicate across different base model families, especially within the Llama series. This raises a core question: what factors lead to inconsistent performance of different base models during reinforcement learning? How does reinforcement learning perform in

In a recent major announcement, the global e-commerce and cloud computing giant Amazon revealed two important milestones in its robotics and artificial intelligence (AI) fields: the launch of a new AI foundation large model called Deep Fleet, and the successful deployment of more than one million robots. The launch of the Deep Fleet model aims to enhance the intelligence and efficiency of Amazon's largest industrial mobile robot fleet. The application of this model is expected to improve the operational efficiency of the robot fleet by 10%, thereby accelerating...

OpenAI recently announced that, despite initial testing, it will not adopt Google's TPU chips on a large scale. TPU (Tensor Processing Unit) is a custom ASIC chip developed by Google for machine learning tasks, designed to accelerate the training and inference of neural networks. TPU uses a dataflow-driven architecture, enabling efficient matrix multiplication pipeline computing and reducing memory access latency. Image source note: The image is AI-generated, provided by the licensing service Midjourney. OpenAI stated that it will continue

According to NetEase Technology, JD.com's layout in the field of embodied intelligence is accelerating rapidly. The embodied intelligence brand JoyInside under JD.com has reached cooperation with more than ten leading robot companies, becoming the core engine for JD.com to seize the smart robot market. According to insiders, JoyInside is supported by JD's large model technology, focusing on providing smart interaction capabilities between robots and consumers. Its product strategy focuses on scenario-based applications such as one person, one dog, and one toy. Since its launch, the brand has successfully attracted leading enterprises from multiple niche fields to join.

Recently, Hon Hai Precision Industrial Co., Ltd. (commonly known as Foxconn) submitted a trademark registration application for "FoxBrain" to the Trademark Office of the National Intellectual Property Administration. This AI inference large model is not only Foxconn's first attempt but also the first AI model of this type in Taiwan. According to public information, the international classification of this trademark is scientific instruments, and it is currently in the "waiting for substantive examination" status. "FoxBrain" is an AI inference large model launched by the Hon Hai Research Institute, covering data analysis