Search AI Products and News

A 73-year-old US woman was scammed by AI voice cloning. A caller posing as her daughter claimed she caused a car accident while texting, injured a pregnant woman, and needed $15,000 bail. In panic, she withdrew cash and handed it to a courier, only learning of the fraud after reaching her real daughter.....

Google I/O unveiled major AI creation tool upgrades, centered on the new Gemini Omni multimodal model. It seamlessly integrates text, image, audio, and video for efficient cross-modal understanding and processing, significantly lowering the creation barrier.....

On June 24, OpenAI and Broadcom unveiled Jalapeño, their first custom AI inference chip. Designed for large model inference, this ASIC marks a deep push into hardware architecture to reduce reliance on a single computing supplier, with remarkable development speed.....

Alphabet, the parent company of Google, announced plans to raise $8 billion by selling stocks to support AI infrastructure and global computing power expansion, including a $1 billion targeted stock sale to Berkshire Hathaway. This move reflects technology giants' fierce competition in the AI arms race to upgrade computing power and respond to strong demand for AI solutions from businesses and consumers.

OpenAI announced the retirement of two classic large models from ChatGPT: GPT-4.5, known as a liberal arts genius, will be decommissioned on June 27, and the hardcore reasoning model o3 will bid farewell on August 26. This adjustment has caused industry shock and left paid legacy users feeling reluctant.....

After Google's I/O Conference in 2026 announced the comprehensive AI transformation of its search engine, users became dissatisfied due to the lack of an 'one-click disable' option for AI features. Privacy search platform DuckDuckGo detected a significant shift in traffic, becoming the preferred alternative for users. Data shows a sharp increase in its installation volume between May 20 and 25, reflecting a trend of users voting with their feet.

ZhiXiang Future released HiDream-O1-Image-Pro, an image model based on Unified Transformer architecture with over 200 billion parameters, achieving multiple SOTA records at its Beijing Open Day. It also completed its second funding round within half a month, backed by top investors like Shenzhen Capital Group and Jinpu Investment, highlighting capital market recognition of native full-modal technology.....

At its first Open Day, Zhixiang Future launched the HiDream-O1-Image-Pro, an image model based on the 'Unified Transformer (UiT)' architecture with over 200 billion parameters. It unifies pixels, text, and task conditions, achieving top-tier industry performance in text-to-image, text rendering, and image editing tasks.....

Google and Samsung jointly launched two smart glasses at the 2026 I/O conference, in collaboration with Gentle Monster and Warby Parker. The products are positioned as an extension of smartphones, deeply integrated with Gemini AI, to free users' hands. Users can use them in scenarios such as commuting, walking, and shopping without frequently taking out their phones.

At Google I/O, Google and Volvo announced a partnership to integrate the AI assistant Gemini into Volvo's electric SUV EX60's external cameras, accelerating visual and motion perception capabilities. This marks a substantive breakthrough in deep integration of AI large models with smart car hardware, leveraging Volvo's native Google embedded in-vehicle system.....

Google I/O 2026 will take place on May 19-20, featuring the next-gen Gemini 3.2 or 3.5 model with performance comparable to GPT-5.5, focusing on on-device optimization and context handling. Additionally, Google will unveil the 'Jinju' Android XR smart glasses in partnership with Samsung, marking a full restart of its smart hardware lineup.....

Google I/O conference focuses on the next-gen AI model Gemini 4.0, aiming to compete with OpenAI's GPT-5.5. Google must maintain its lead amid accelerated innovation. The event will showcase ecosystem upgrades, with Gemini 4.0 expected to enhance logical reasoning, a key highlight in the AI race.....

OpenAI veteran Joanne Jang, a key early member involved in GPT-4 and DALL-E 2, has resigned. Her departure may impact product strategy and IPO plans.....

The Wuxi High-tech Zone has created an 'Double O' community AI manga drama space, promoting the 'One Person, One Army' content production model. Young people can complete the entire process of scriptwriting, voice acting, and editing with just a computer, demonstrating the reshaping of the creative industry by technology. This layout aims to seize the new trend in AI manga drama and help Wuxi gain an advantage in the digital economy.

OpenAI executive Joanne Jang, key to GPT-4o and ChatGPT's development, has resigned after 4.5 years, potentially impacting the company's IPO.....

Alibaba's Tongyi Lab introduces the FIPO algorithm, which overcomes traditional reinforcement learning bottlenecks in complex logical reasoning. Using the Future-KL mechanism, it accurately identifies key reasoning steps, effectively addressing model stagnation in tasks like mathematics, thereby enhancing both accuracy and efficiency.....

Alibaba's Tongyi Lab introduces FIPO, a novel algorithm with a 'Future-KL' mechanism that addresses the 'reasoning length stagnation' in pure reinforcement learning for long-text reasoning, enhancing complex logic alignment training.....

N.O.M.A.D. is an offline server system designed as a 'doomsday toolbox' to provide essential civilization information and technical tools in extreme scenarios where the internet is down but power remains, running on Debian or Ubuntu.....

Westlake Robotics Company released the humanoid robot Titan o1, featuring the world's first action generalization large model GAE Avatar System. This robot has millisecond-level imitation capabilities, allowing real-time replication of human movements and enabling 'one person controlling multiple cyborg avatars', showcasing a new breakthrough in the field of robotics.

The Alibaba Qwen 3.5 series small model breaks the conventional belief that parameter count determines intelligence. Among them, Qwen 3.5-4B with only 4 billion parameters performs equally well or even slightly better than GPT-4o, which has over 100 billion parameters, in third-party tests. This marks an important breakthrough in local deployment and efficiency optimization for domestic large models, ushering in a new era of 'winning with small size'.