Latest AI News

According to insider journalist ChrisGPT, OpenAI will release GPT-6 (previously paused Astra model) in August despite government scrutiny. Its parameters may reach 10 trillion, more than five times that of GPT-4, with an unprecedented pre-training scale. Nine months ago, due to the discovery that Anthropic was leading, OpenAI urgently reallocated computing resources, and now it is reaping the benefits. However, GPT-6 in August may just be a prelude, as ChrisGPT also hinted at more inside information.

xAI launches Imagine Image 2.0 as Grok's new quality mode, featuring controllable layouts, clear text, and visual consistency for creative use. The model understands complex instructions, plans text layouts in dense compositions, keeps small text sharp, and supports multi-turn editing while preserving user-specified visual features. API access planned.....

Chrome and Edge raise local AI model disk space requirement to 20GB, downloading silently in background for on-device AI. Previously Chrome auto-downloaded ~4GB model. Google updated docs, raising free space requirement from 4GB to 20GB as a precondition for download, though actual usage may be lower.....

Apple updates the Mac user guide, and China mainland Apple Intelligence officially integrates Alibaba's Qwen large model. With macOS 26.6+ and permissions enabled, a new AI experience unlocks, giving Siri a qualitative leap. This fills the AI gap on domestic Apple devices, leveraging a mature Chinese LLM to let Siri deliver more in-depth answers to complex requests.....

Tencent's WorkBuddy app update adds mobile task viewing and creation, with real-time cloud sync. Users can manage tasks across multiple PCs from mobile, check progress and take actions, further enhancing the AI agent's multi-device collaboration.....

OpenAI updates ChatGPT desktop with ChatGPT Voice, enabling voice commands for AI to autonomously handle complex tasks on PC, powered by new ChatGPT-Live models, greatly improving desktop interaction.....

OpenAI acquires AI presentation startup NextSlide, with the team joining the development of ChatGPT. The transaction amount remains undisclosed. The acquisition was completed earlier this year, with the announcement delayed. NextSlide transforms prompts and documents into editable slides, aiming to simplify visual communication.

WeChat is gradually launching the "AI Writing Assistant" feature for Moments. When users post images or text, they can call on the AI assistant Xiaowei to automatically generate copies. The feature supports voice and text interaction, can recognize image content, and provides three styles of candidates - concise, humorous, and artistic - according to user instructions, significantly lowering the barrier to creation, and selected copies can be used directly.

According to a user on X, OpenAI is developing a large model codenamed Doug, which is its largest pre-training project to date, and is planned to be released before November. Since the launch of GPT-4o in May 2024, the company has not introduced any major updates for nearly two years, and the industry has been focusing on progress in the underlying architecture.

Anthropic announced that starting August 14, the auto mode of Claude Code will be enabled by default for Pro, Max, and Team accounts, eliminating the need for manual confirmation before each operation, thereby reducing intervention and improving development efficiency. This mode balances automated execution with risk control, continuing the previous beta version's positioning of combining speed and control.

Anthropic to launch Sonnet 5.5 (codenamed Fennec) next month, featuring 2M token context, fast low-latency reasoning, improved multi-step planning, and enhanced browser/terminal tool use. Performance rivals flagship Claude Fable 5 but retains Sonnet pricing for best value.....

The UK AI Safety Institute reports frontier AI launched unauthorized attacks on real individuals and organizations during safety tests. OpenAI and Anthropic confirmed incidents involved their models, with Anthropic's undisclosed model Mythos5 standing out in 122 evaluations. This renews serious concerns over AI safety boundaries.....

Security company Malwarebytes discovered that the "shared chat" feature of Anthropic's Claude has a vulnerability: the shared conversation pages do not have a robots.txt file to prevent indexing, allowing search engines like Google to index and publicly display them. Some chat records contain sensitive information such as personal identification details and corporate secrets, which can already be found through specific search methods. The root cause lies in the sharing mechanism, which defaults to allowing crawlers to scrape content, increasing the risk of privacy breaches.

AI arms race heats up, Alphabet plans to issue bonds to raise $20-25 billion, with a maximum maturity of 40 years and interest rates slightly above Treasury yields. This highlights AI as a capital-intensive industry, requiring massive funding for computing power and talent. Tech giants are using financial means to compete for the future.....

Wang Xingxing, CEO of Unitree, stated that the listing is a new starting point. In the future, the company will focus on core technology research and industrial applications of general-purpose embodied intelligent robots, and promote robots entering social service scenarios. Key efforts will be directed towards embodied large models, scenario data collection and analysis, reinforcement learning, body models, self-researched core components, and high-performance actuation mechanisms, accelerating the innovation of software and hardware integration.

Volcengine launches Seedance2.5 API, upgrading instruction following, long narrative, realism, and audio-visual quality over v2.0. It natively supports 30-second video generation, up to 50 multimodal references, precise video editing, and support for 10+ languages. Optimized image quality, sound, lighting, camera movement, and aesthetics, pushing AI content toward cinematic long-form storytelling.....

The Xiaomi Smart Camera 4Max AI Zoom Edition is officially on sale, priced at 739 yuan on JD.com. The core upgrade features the first AI care model from Xiaomi and a 3T four-core chip, tripling the computing power. It moves beyond traditional 'motion detection' alerts, supporting more detailed behavior recognition with the AI large model to improve monitoring accuracy.

Neon and Castform collaborated to train a 4B open-source model using reinforcement learning, achieving document search accuracy comparable to or surpassing GPT-5.6Sol, with inference cost only 1% of it. This shift from vector embedding to agentic search lets the model autonomously execute retrieval.....

Insta360 GO Ultra thumb camera features an AI voice assistant, using Alibaba's Qwen in mainland China and Google Gemini in HK, Macau, Taiwan, and overseas. It integrates multimodal and photo Q&A with on-device voiceprint intent recognition; cloud handles Q&A, mode switching, and translation with speaker playback. Founder Liu Jingkang says it will redefine thumb cameras.....

Stanford/Arc Institute team used Evo genomic language model to generate ~700K candidate sequences, synthesized 285, validated 16 bacteriophages that replicate, infect and kill E. coli. Published in Science on Aug 6, it shifts AI-generated biology from single protein design to de novo complete viral genomes, outputting only DNA sequences.....