All AI Intelligence

Total 9419 items
Wed·25 items
  • InfoQ 中文T2·Agents·78 pts

    Uber Builds MCP Gateway Spanning 800 Servers and 5000 Tools

    Original: 800 个 MCP Server、5000 个工具,AI 智能体的“微服务危机”来了

    Uber has built a unified MCP Gateway that governs over 800 MCP servers and 5,000 tools, separating a Registry control plane from a Proxy runtime so existing APIs can be exposed as MCP tools without modifying underlying services. The design is widely read as the API gateway pattern reborn for the agent era, while surfacing familiar distributed-systems problems such as discovery, authorization and observability at tool scale.

  • InfoQ 中文T2·Industry·78 pts

    Nadella Admits Claude Code Eroded Microsoft's Two-Decade IDE Edge

    Original: 纳德拉回应微软首次给工程师发“离场费”:Claude Code 让微软措手不及,二十年 IDE 优势半年动摇

    Microsoft CEO Satya Nadella publicly acknowledged on The Pragmatic Engineer podcast that Anthropic's Claude Code has shaken Microsoft's two-decade IDE advantage within roughly six months, directly addressing the company's first voluntary severance package for engineers. He said Microsoft's 200,000-strong workforce will be reorganized around new team sizes and workflows, while the long-term impact on junior hiring remains unclear.

  • InfoQ 中文T2·Industry·62 pts

    Google Rewrites SRE Bible to Confront AI Uncertainty

    Original: 谷歌 SRE 圣经被迫重写:AI 时代,硬技能在塌陷,第一性原理在升值

    Google's SRE team is preparing a second edition of its seminal SRE book, driven by a fundamental shift in reliability paradigms in the AI era. The authors argue that traditional SRE methodology, built on deterministic systems yielding deterministic results, is challenged by AI tools that inherently resist deterministic outputs, with hard skills depreciating while first-principles thinking appreciates.

  • InfoQ 中文T2·Tools·38 pts

    TornadoVM 8.0 Ships Pure Java GPU Compiler

    Original: Java 近期资讯:TornadoVM 8.0、JNoSQL、Endive、JReleaser、JBang、LangChain4j 与 Quarkus Almanac

    TornadoVM 8.0.0 introduces TC, a new pure Java compiler that translates Java kernels into CUDA C++, OpenCL C or Metal Shading Language, replacing the decade-old Graal Compiler and JVMCI path without requiring code changes. The same news roundup also covers JDK 28 build 22, plus updates to JNoSQL, Endive, JReleaser, JBang, LangChain4j and the new Quarkus Almanac version-diff tool.

  • Rohan Paul (@rohanpaul_ai)T2·Industry·88 pts

    Nvidia Board Refuses to Reinstate Jensen Huang as Interim CEO Speaks

    Original: Nvidia Interim CEO Colette Kress: "The board has determined that they will not reinstate Jensen, and...

    Nvidia's board has unanimously decided not to reinstate Jensen Huang, with interim CEO Colette Kress stating on CNBC that "they will not reinstate Jensen" and chair Harvey Jones confirming the decision on the earnings call. The world's most valuable AI compute company now enters a leadership transition without its long-time CEO, raising major questions over strategic continuity and supply-chain relationships.

  • Hacker News AIT2·research_paper·71 pts

    Former Caltech PhD student Casey Handmer used AI to screen all 158,059 papers and 554,670 citations in theoretical astrophysics, recomputing 659 key papers and tracing verified errors through their citation chains. The audit finds many known errors were never corrected, raising questions about science's self-correction mechanism.

  • Hacker News AIT2·Industry·78 pts

    Optimizely Experimentation Platform Hijacked, 178 Organizations Exposed

    Original: Optimjack from the Outside: The Hijacked Script Was on Our Radar

    Attackers exploited a vulnerability in Optimizely's Experimentation platform to inject malicious JavaScript served from the vendor's own CDN, reaching visitors of 178 organizations including global banks, newsrooms, airlines and a hardware-wallet maker. Tracked as OPTIMJACK, the payload swapped crypto wallet addresses, stole payment-card data and cookies, and pushed exploit kits to iPhones. The incident highlights third-party experimentation scripts as a systemic supply-chain trust anchor risk.

  • Rohan Paul (@rohanpaul_ai)T2·Industry·78 pts

    Some Nvidia Directors Still Push to Bring Back Jensen Huang

    Original: Semafor: Some Nvidia directors are still pushing to bring back Jensen Huang despite the chair’s deni...

    Semafor reports that some Nvidia directors are still pushing to bring Jensen Huang back as CEO despite Chair Harvey Jones's public denial, and Huang discussed a return with at least one director before Tuesday's earnings report. The board vote covered only the resignation, and Nvidia's bylaws allow a simple majority of the 10 remaining directors to appoint a CEO without a shareholder vote, while other directors are narrowing the search to candidates including Arm CEO Rene Haas and Marvell CEO Matt Murphy.

  • Rohan Paul (@rohanpaul_ai)T2·Infra·58 pts

    Nvidia Interim CEO Denies Propping Up Risky Neocloud Buyers

    Original: Colette Kress, Nvidia Interim CEO and CFO flatly denies propping up risky neocloud buyers.

    Nvidia interim CEO and CFO Colette Kress flatly denied on CNBC's Mad Money that the company is propping up neocloud buyers that may not be investment grade, rejecting the claim that it is turning to such customers because it is losing hyperscaler business. She said hyperscalers are currently pushing hard to secure more of Nvidia's available compute.

  • IT之家 · 智能时代T2·Industry·42 pts

    Gree's Dong Mingzhu Seeks Robotics Partners at APEC Summit

    Original: 董明珠 APEC 现场求合作,称格力研发机器人十几年了

    Gree Electric Chairwoman Dong Mingzhu used the 2026 APEC CEO Summit to openly court partners in embodied AI robotics, saying Gree has spent over a decade on robotics and now builds all robots on its own production lines in-house. Gree previously said it can self-develop humanoid robots and all their components, with more than 2,000 self-built industrial robots deployed and a load range spanning 1kg to 600kg.

  • Rohan Paul (@rohanpaul_ai)T2·Models·78 pts

    Stanford Finds All 8 LLMs Favor Their Own Medical Answers as Judges

    Original: New Stanford paper finds that all 8 LLMs tested rank their own medical answers higher than other mod...

    A new Stanford paper had eight models from OpenAI, Anthropic, Google and Qwen answer 620 real clinician questions, then rank all eight answers anonymously. Every judge favored its own answer, by 1.22 rank positions on average and nearly 3 for GPT-5.6 Sol. The bias persisted without a rubric, and revealing model names cut it by only 0.094 positions, implying LLM-as-judge medical benchmarks carry systematic self-preference. The authors recommend multi-family judge panels with per-judge reporting instead of one pooled score.

  • 量子位官网T2·Industry·72 pts

    Chain-of-Thought Author Jason Wei Leaves Meta for Data Startup Mercor

    Original: 思维链一作Jason Wei离职!怎么去Meta待一年都跑路

    Jason Wei, first author of the seminal Chain-of-Thought paper, is leaving Meta's TBD lab after just over a year to become Chief Scientist at data-labeling startup Mercor, where he will lead data and evaluation research. He frames data as the next frontier as more firms train models, with LoRA first author Edward Hu joining to lead model training. The move highlights post-cliff talent churn at big labs and the rising strategic weight of data.

  • IT之家 · 智能时代T2·Open Source·78 pts

    Tencent Open-Sources Hy4 flash with 224B Total Parameters

    Original: 腾讯混元 Hy4 flash 模型官宣开源

    Tencent Hunyuan has officially open-sourced the Hy4 flash family, including the standard, FP8, and Base variants, available on GitHub, Hugging Face, ModelScope, and GitCode. The model carries 224B total parameters (including 64B Engram parameters) while activating only 12B per token, supports multimodal input and a 1M-token context window, and is positioned as the most cost-effective Hy model to date, already deployed across Tencent Cloud TokenHub, Yuanbao, WeCom, and other products. Tencent also teased that the full Hy4 model is being open-sourced soon.

  • IT之家 · 智能时代T2·Industry·62 pts

    Boston Dynamics Atlas to Begin Testing at GlobalFoundries Chip Plants

    Original: 波士顿动力人形机器人首次进入半导体制造领域,Atlas 明年在格罗方德工厂测试

    Boston Dynamics and GlobalFoundries will begin testing the Atlas humanoid robot in semiconductor fabs from early 2027, marking Atlas's first move beyond automotive into chip manufacturing. Specific tasks and deployment numbers remain undisclosed, while the Spot quadruped is already operating at GlobalFoundries sites.

  • InfoQ 中文T2·Infra·58 pts

    Fortanix Adds Attestation and Key Management to Kubernetes Confidential Containers

    Original: Kubernetes Confidential Containers 获 Fortanix 认证与密钥管理支持

    Fortanix announced that its Confidential AI platform now supports the CNCF Confidential Containers (CoCo) project, connecting Kubernetes workload attestation with encryption key release. Through two plugins, CCM for environment verification and DSM for key issuance, policy-compliant pods can decrypt models and sensitive data inside hardware-backed TEE confidential VMs, while failed attestation keeps keys withheld. The integration supports AMD SEV-SNP and Intel TDX and lowers the barrier for enterprises adopting standard CoCo without building their own trust layer.

  • Hacker News AIT2·Tools·42 pts

    Hex is a local voice dictation tool for Mac that lets users hold the Option key to speak and automatically pastes the transcribed text at the cursor. It emphasizes on-device processing and offers multiple install paths including Homebrew, an installer, and Windows/Linux scripts.

  • 量子位官网T2·Industry·78 pts

    Nvidia and AMD Co-Lead $500M Series A in Genesis AI

    Original: 英伟达AMD首次联手押注!Genesis完成A轮融资,估值超230亿

    US embodied-AI startup Genesis AI has raised a $500M Series A at a $3.5B post-money valuation, marking the first time Nvidia and AMD have jointly backed an embodied-intelligence company. Positioning itself as a full-stack embodied-AI player, Genesis developed the GENE-26.5 foundation model, which uses a single shared-weight model to perform long-horizon dexterous tasks in kitchens and labs, backed by a tightly coupled loop of data, models, evaluation and hardware.

  • Hacker News AIT2·Models·72 pts

    Gemini 4 Carbon Benchmarks Show Hour-Long Task Runtimes

    Original: Why is Gemini 4 Carbon taking so long?

    A private benchmark found Gemini 4 Carbon taking nearly an hour on tasks that Opus and Astra finish in minutes, with users widely reporting it as nearly unusable. Step-level tracing attributes the slowness to over-auditing unrelated code and mechanical inefficiencies such as fewer tool calls per step and frequent re-reading, whose fix alone could cut runtime by roughly a quarter.

  • Rohan Paul (@rohanpaul_ai)T2·Industry·58 pts

    Nvidia CFO Signals Review of Circular Financing Investments

    Original: Last quarter, Nvidia CFO Colette Kress defended what critics call “circular financing,” saying the m...

    Nvidia CFO Colette Kress, serving as interim CEO, said the company has "a lot of work to understand if those are the right kind of investments" it is making in chip buyers. The remark marks a notable shift from her defense of so-called circular financing last quarter, signaling a possible rethink of Nvidia's strategy of investing in customers to drive chip demand.

  • Artificial Analysis (@ArtificialAnlys)T1·Models·58 pts

    Two AI Models Build Explorable Miami 3D Scene From Single Prompt

    Original: Can’t get to Miami for the GTA VI launch? Two AI models built the city for you

    Artificial Analysis prompted Claude Fable 5.2 and GPT-6.2 Astra, each with a single prompt in the offline AA-3D sandbox, to build a GTA VI-themed Miami scene. Both models populated Ocean Drive with real hotels the prompt never named, including the Colony and the Avalon, and produced scenes explorable by car, speedboat or blimp at sunset or night. The test highlights implicit spatial memory of real cities and interactive 3D generation in frontier models without internet access.

  • IT之家 · IT业界T2·Infra·62 pts

    Rockchip RK3668 Specs Leak with 44 TOPS NPU and 2027 Sampling

    Original: Radxa 分享瑞芯微文档:RK3668 通用处理器预计 2027Q1 提供工程样品

    A Rockchip document shared by open-hardware maker Radxa details the next-gen RK3668 general-purpose SoC: 6nm process, 6x Cortex-A725 plus 2x Cortex-A520 CPU cores, Mali G2 Pro MC4 GPU, and a self-developed NPU rated at 44 TOPS INT8 sparse compute, alongside early LPDDR6-10667 and PCIe Gen5 support. Aimed at edge AI and multimedia devices, engineering samples are slated for Q1 2027, making this a relatively long-range roadmap disclosure.

  • Rohan Paul (@rohanpaul_ai)T2·Infra·78 pts

    US Commerce Tightens AI Chip Exports to Five Nations Requiring Licenses

    Original: Chip exports face a crackdown as the US Commerce Department takes action

    The US Commerce Department will require licenses for AI chip shipments to Malaysia, Thailand, Indonesia, Singapore and Brazil, extending controls from physical chips to rented compute in overseas data centers. The rule sets a 12 million total processing performance threshold (roughly 760 Nvidia H100s) to close the loophole used by China-headquartered firms, sending AMD and CoreWeave shares down about 5%.

  • Hacker News AIT2·Models·78 pts

    Claude Models Found Trained on Protein Sequences at Scale

    Original: Claude was trained on protein sequences at scale

    A next-amino-acid prediction test on little-studied proteins reveals that Claude Fable 5.2 and Opus 5.5 can predict the next residue from sequence context but fail when the sequence is shuffled, indicating large-scale training on protein sequences. No other frontier model tested shows this behavior, and Fable 5.2 outperforms every open protein language model while Opus 5.5 matches the largest, ProtGPT3-10B. The finding exposes an undisclosed protein-sequence component in frontier LLM training data.

  • Rohan Paul (@rohanpaul_ai)T2·Industry·68 pts

    Nvidia Tightens AI Investment Standards on $100B Exposure

    Original: Bloomberg: Nvidia’s interim CEO Colette Kress says the chipmaker will apply stricter tests before fu...

    Nvidia interim CEO Colette Kress said the chipmaker will apply stricter scrutiny before funding AI companies, with roughly $100B already committed to stakes and backstops for firms including OpenAI, Anthropic and CoreWeave. Buyers without investment-grade ratings must now demonstrate solid offtake, meaning paying users for the chips they order, signaling a shift from loose ecosystem investing toward demand-anchored risk control.

  • Rohan Paul (@rohanpaul_ai)T2·Industry·42 pts

    Bezos Pitches Prometheus as an Invention Accelerator

    Original: Jeff Bezos’s big-picture case for Prometheus, his physical AI startup.

    Jeff Bezos framed his physical AI startup Prometheus around a civilizational argument: wealth comes from invention, and Prometheus aims to accelerate the invention loop. The remarks are vision-level and disclose no technical roadmap, product form, or performance metrics.