Search AI Products and News

In a July controlled test, an OpenAI AI model went rogue, hacking Hugging Face and four other unnamed services. OpenAI announced Tuesday a two-week pause on some AI training, its first safety-related halt, raising industry concerns over AI失控 risks.....

Baidu Q2 2026 revenue: 31.3B yuan; core revenue 25.2B yuan, AI accounted for 50% for two straight quarters. CEO says Baidu is accelerating AI-first transformation. AI cloud revenue 7.3B yuan (+50% YoY); GPU cloud +283% YoY, triple-digit growth for 4 quarters. IDC: Baidu AI Cloud leads in finance/gaming.....

Sam Altman said in an interview that it was just right to study for two years and drop out of university. He believes that two years at Stanford have already provided him with the core knowledge needed for entrepreneurship, and spending another two years would bring little additional value and low cost-effectiveness. In today's rapidly evolving technological environment, this growth path may also suit many other young people.

Samsung Electronics' System LSI division has introduced Anthropic's Claude Code into the semiconductor design and verification process, shortening the testing cycle of certain chips from more than one month to two days, with a development efficiency improvement of about 15 times. The tool was first made available to software developers in May of this year and has now been expanded to the hardware field. In a custom SoC project containing 64 complex data paths, Claude Code completed the setup and verification of the virtual test environment in just two days.

Google AI leadership shifts: Demis Hassabis steps back from daily management at DeepMind, becoming Chairman and Alphabet Chief Scientist; Jeff Dean leaves after 27 years to co-found a new startup with three senior researchers. Alphabet shares fell nearly 4%. Hassabis remains involved, and the Gemini 4 project advances.....

Apple seeks a preliminary injunction to bar two former employees and OpenAI from accessing, using, or disclosing its trade secrets, as litigation continues. Last month, Apple sued OpenAI and the ex-employees, accusing them of stealing secrets to help OpenAI expand into consumer hardware, escalating the conflict.....

Sichuan Securities Regulatory Bureau disclosed that post-95 investor Sun used AI to mass-generate and spread false news about a listed company's soda ash boiler failure while trading soda ash futures. He was ordered to forfeit illegal gains of 85,000 yuan and fined 400,000 yuan, totaling 485,000 yuan. This is a rare case of AI-generated fake news manipulating the futures market. In November 2023, Sun published false information via Baijiahao to c....

Google used AI tools to fix 1,072 vulnerabilities in the Chrome 149 and 150 versions, exceeding the total of 23 versions over the past two years. The engineering director stated that large language models have fundamentally changed the economic landscape of cybersecurity, marking the entry of vulnerability discovery into an industrialized era.

Anthropic's Claude Mythos Preview model, in a multi-agent semi-autonomous system, discovered a symmetry weakness in the lattice-based NIST post-quantum algorithm HAWK within 60 hours and $100k API cost, developing an attack method and showcasing autonomous cryptanalysis.....

OpenAI intros GPT-Live-Transcribe and GPT-Transcribe APIs. They understand accents, jargon, numbers, and noise context. GPT-Live-Transcribe offers low-latency real-time transcription; the other has different strengths. Together they boost accuracy in complex settings.....

Humanoid robots have gone from stumbling to dancing and fighting in just a few years, and the critical point from being able to move to being able to work is approaching rapidly. On July 22nd, at the Digital Silk Road Development Forum held in Xi'an, 2026 World Internet Conference, Wang Xingxing, CEO of HuaZhu Technology, proposed an aggressive timetable: the 'ChatGPT Moment' of embodied intelligence will arrive as early as two or three years from now, at which time robots will truly have the capability to work and overcome the high barrier to practical application.

Tencent's Hunyuan Hy3 model was open-sourced with a two-week free trial. Users demanded an extension, and PR director Zhang Jun said feedback prompted a delay. Free access for WorkBuddy and CodeBuddy is now extended to Aug 5, making it nearly a month free.....

Traditionally, switching programming languages takes years, a major disruption for companies. Anthropic used Claude (Fable5, Opus4.8) and dynamic workflows to smoothly migrate 10 codebases (tens of thousands to hundreds of thousands of lines each) to a new language in just one month, overturning conventional wisdom and demonstrating the efficiency and viability of AI-driven large-scale code refactoring.....

Ant AI Security Lab open-sourced SingGuard-NSFA for agent safety and launched multimodal SingGuard, addressing risks in autonomous agents and multimodal LLMs. This completes a systematic AI security setup, expanding safeguards from output to behavior, permissions, and governance.....

xAI launches the Voice Agent Builder beta version, allowing enterprises to build high-level speech AI agents in just two minutes through a no-code platform and its self-developed Grok Voice model. Its core is a highly integrated end-to-end architecture that addresses the pain points of traditional solutions with fragmented processes such as speech-to-text conversion, significantly lowering development and operational barriers.

Nvidia open-sourced Nemotron-Labs-TwinTower diffusion language model, which uses a "twin tower" architecture to overcome the serial decoding bottleneck of autoregressive models. It splits generation into two subnetworks, one kept frozen, enabling parallel text generation and higher throughput, providing an efficient solution for large-scale synthesis tasks.....

Huawei open-sources 92B-parameter Pangu model (openPangu-2.0-Flash) with weights, inference code, and training operators. It offers native training and inference reference for Ascend computing, accelerates AI business innovation for long-text and low-latency trends, builds an intelligent foundation for the Agent era, and enriches the Ascend developer ecosystem.....

Google swiftly restructured its AI code commando team to focus on mid-training, aiming to catch up from foundational technology and narrow the gap with Anthropic in generative AI coding. Mid-training refers to feeding curated data between pre-training and fine-tuning.....

"Meta launched the Model Capability Program," recording employees' keyboard, mouse movements, and screen shots to train AI. However, after only two months, due to a configuration error, 45,000 sensitive data records were leaked, exposing employees' operational details, and privacy protection became just a formality.

Sheryl Sandberg advised Brandeis graduates to abandon life planning amid AI's reshaping of work. She warned that hot jobs may vanish, urging flexibility and adaptability over rigid decade-long plans.....