AI 今日新闻 · 2026-05-07

OpenAI, in collaboration with AMD, NVIDIA, and others, introduces MRC—a new networking protocol that enhances performance and reliability for large-scale AI training, now open-sourced via OCP to advance industry standards.

Anthropic co-founders reveal that Claude's rate limits stem from Q1 2026 usage growing at an 80x annualized rate, far exceeding their 10x compute planning. The company is responding with massive compute deals like the one with SpaceX.

Genesis AI unveiled GENE-26.5, its first general-purpose robot foundation model, capable of complex tasks like cracking eggs, solving Rubik's cubes, and playing piano—all autonomously with minimal real-world fine-tuning data.

On May 5, 2026, incorrect DNSSEC signatures from the .de registry caused global resolution failures; validating resolvers like 1.1.1.1 returned SERVFAIL, impacting millions of domains.

Qwen's new desktop voice input enables global activation, mixed Chinese-English recognition, and AI-powered content generation and task execution, enabling truly hands-free, efficient office work.

NVIDIA advocates for cost per token as the core economic metric for AI infrastructure, replacing traditional measures like compute cost or FLOPS per dollar, emphasizing full-stack optimization to reduce inference costs and enhance business value.

Berkshire's new CEO Greg Abel systematically outlines his management philosophy: adhering to a ten-year margin of safety, decentralized yet accountable leadership, opposing breakup, and emphasizing the unique advantages of an efficient conglomerate in capital allocation and risk resilience.

In a deep dialogue, Taleb systematically critiques mainstream forecasting models and behavioral economics, revealing the unpredictability of fat-tailed events in Extremistan and proposing a survival framework based on convexity, precautionary principles, and ergodicity.

Doist launched Ramble, using Gemini Enterprise Agent Platform to turn unstructured spoken input into structured task lists with low latency and high accuracy.

Google achieved 6x faster migration from TensorFlow to JAX using a specialized multi-agent AI system, solving key challenges like context loss and build failures in large-scale codebase transitions.

OpenAI, together with AMD, Intel, NVIDIA and others, has open-sourced MRC, a new networking protocol that improves reliability and efficiency in large-scale AI training clusters.

This podcast episode dives into the reasons behind the US's stagnation in nuclear energy, compares China and France's success stories, and analyzes the future of nuclear technology.

The article explores the author's realization that the boundaries between 'vibe coding' and 'agentic engineering' are blurring with AI-assisted programming, raising concerns about code quality and responsibility.

Remote work relies on far more than Wi-Fi — it depends on a layered stack of connectivity, cloud platforms, identity systems, and security architectures that collectively determine productivity.

QuRT is the real-time operating system running on Qualcomm's Hexagon DSP, designed for microsecond-level deterministic scheduling to handle low-latency tasks like audio processing, sensor fusion, and AI inference in smartphones.

Anthropic announced a 300MW/$5B/year compute deal with SpaceXai for Colossus I at its dev event, with 8000% annualized ARR growth and three new Claude Managed Agents features.

Due to the retirement of Ingress-NGINX, Stack Overflow evaluated alternatives and selected a Gateway API-compliant controller, validating migration through real-world use cases and performance benchmarks.

Frontier enterprises are compounding their AI advantage through deeper, broader adoption—especially in agentic workflows and complex task execution—outpacing peers significantly.

Uber partners with OpenAI to power AI assistants and voice features that help drivers optimize earnings in real time and enable riders to book rides more smoothly using large language models.

GPT-1 introduced a two-stage approach combining unsupervised generative pre-training with task-specific fine-tuning, significantly advancing language understanding and laying the foundation for large language models.

Google Cloud introduces a new IAM security framework for the AI agent era, featuring Agent Identity, Agent Gateway, and runtime defense to strengthen identity verification, access control, and zero-trust governance for agents.

Research reveals poor retrieval quality is the primary cause of high-fluency hallucinations in RAG systems—more convincing, confident, and wrong—while scaling models fails to fix the root issue.

Based on 297k Hugging Face users, this analysis reveals top AI hardware configurations, showing NVIDIA, Apple, and Intel dominance in their respective categories and the importance of VRAM for AI workloads.

The new Grafana Assistant integration for Database Observability leverages AI and real-time observability data to help users quickly diagnose database performance issues without manual context assembly.

Tomofun significantly reduced inference costs for vision-language models in pet behavior detection using AWS Inferentia2 chips, while maintaining high accuracy and throughput for large-scale real-time monitoring.

The vLLM upgrade from V0 to V1 focuses on backend inference correctness, fixing critical issues like logprob semantics, runtime defaults, and inflight weight updates to ensure reliable results in reinforcement learning training.

iFlytek's Vision Agent leverages multi-agent architecture to generate high-quality PPTs, moving beyond template collage to deliver commercially viable outputs across diverse scenarios, marking the transition of AI PPT into a 2.0 practical phase.

The article explores how to validate AI agent behavior in scenarios without a single correct answer, emphasizing goal achievement and path合理性 evaluation.

Boston Dynamics unveiled its new Atlas robot with advanced capabilities, but monthly production is only 4 units. Amid C-suite departures and IPO preparations, it faces pressure from Hyundai and rising competition.

Unisound launches U1-InsureMed, a vertical LLM for medical insurance, integrating billions of clinical records, enabling intelligent audit and risk control in social and commercial insurance, deployed in Jiangsu province with ~20% cost-control improvement.

In agile development, the real challenge for designers isn't speed—it's learning to design smaller, self-contained features that deliver immediate value, not complete systems.

Wuwen Xinqiong, a neutral AI infrastructure provider, supports the token explosion of domestic large models, with daily token calls up 20x in two years and nearly $2.2B in funding, becoming a core hub in the AGI era.

Databricks proposes re-architecting distributed systems for serverless environments by decoupling compute, storage, and metadata to improve performance and reliability.

Breuninger leveraged Google Cloud's AI-powered virtual try-on to let users preview clothing on themselves, significantly improving conversion and reducing returns.

Elastic showcased key advancements at Google Cloud Next 2026, including security integration, model deployment, and performance gains in the agentic era.

PDF parsing remains a critical bottleneck for AI automation of knowledge work; current OCR and vision-language models perform poorly on complex layouts and tables, requiring specialized tooling to improve data extraction quality.

Similarity alone isn't enough for business needs. Milvus Boost Ranker layers business rules on top to surface the right results first, not just the closest ones.

The Gemini API's File Search now supports multimodal retrieval. Use `gemini-embedding-2` to build a unified RAG system for PDFs and images with a single call. Storage and query-time embeddings are free; you only pay for indexing and inference.

Elasticsearch 9.4 introduces new capabilities supporting the Dell AI Data Platform powered by NVIDIA, enhancing vector search, context engineering, and native AI workflows to advance enterprise AI applications.

OpenAI, in partnership with AMD, NVIDIA, and others, has released MRC, an open networking protocol designed to solve data synchronization reliability and efficiency challenges in large-scale AI training clusters, significantly reducing GPU idle time.

This article introduces the motivation, challenges, and solutions behind Superlinked's development of inference infrastructure for small models.

Anthropic announced several updates to the Claude platform at the Code w/ Claude 2026 event, including increased API rate limits via SpaceX collaboration, multi-agent coordination, and self-improvement mechanisms.

This article compiles 10 technical news items from Hacker News on May 4, 2026, covering controversies around VS Code's AI co-author feature, the release of the AV2 decoder dav2d, Mercedes-Benz's return to physical buttons, and more.

Microsoft is expanding Azure datacenter regions across Europe to meet rising demand for cloud and AI, delivering sovereign, sustainable, and compliant infrastructure that empowers public and private sector innovation.

Through a dialogue between Lee Sedol and Faker on AI, the article reflects on top human players' emotional, artistic, and competitive relationship with artificial intelligence a decade after AlphaGo.

This episode critically examines AI's role in meditation and mental wellness, challenges the legitimacy of most AI companionship products, and explores how technology can authentically support healing through a case study of an intelligent meditation cushion startup.

Singular Bank in Spain built Singularity, an AI assistant powered by ChatGPT and Codex, enabling real-time portfolio analysis and automated meeting prep, saving bankers 60–90 minutes daily.

TokenSpeed is a new open-source LLM inference engine optimized for agentic workloads, featuring advanced KV caching, an efficient scheduler, and a modular kernel architecture with multi-silicon support.

OpenAI discusses the demand for new supercomputer networks driven by AI development, covering computing architecture, scalability challenges, and future directions.

Anthropic partners with SpaceX to significantly increase compute capacity, raising usage limits for Claude to support more enterprise AI applications.

Google introduces new AI Mode and AI Overviews features that use generative AI to help users explore web content more efficiently, supporting multi-turn conversations, deep aggregation, and source tracing.

Google Flow Music partners with Believe to deliver AI-powered music creation tools to independent artists, producers, and songwriters, enhancing creativity and streamlining distribution.

Vercel introduces a new feature that automatically adds Git committers to your team, streamlining collaboration and reducing manual member management overhead.

Anthropic announces a compute partnership with SpaceX and increases usage caps for Claude Code and API; rolling limits are doubled and peak throttling is removed for Pro and Max users.

Perplexity has launched its in-house inference engine ROSE, enabling efficient serving from embedding models to trillion-parameter LLMs, with CuTeDSL integration for faster GPU kernel customization.

AI can generate a PR in minutes, but reviewing its correctness, safety, and readiness still takes longer. Devin Review and Quick Review in Windsurf 2.0 offer deep PR analysis and fast local bug detection.

Google DeepMind partners with the developers of EVE Online to use its complex, player-driven universe as a safe sandbox for testing AI agents in memory, continual learning, and long-term planning.

Google is recognized by Gartner as a Leader in 2026, leveraging Mandiant, VirusTotal, and Gemini AI to deliver high-fidelity threat intelligence with 98% accuracy.

LlamaIndex launched LlamaParse Mobile, an iOS and Android app built with Expo + React Native, powered by the LlamaParse TypeScript SDK, enabling text extraction from photos in three simple steps.

The post quotes Boris Cherny on the future of async agents, higher-order prompts, and Claude self-prompting, stressing verification—but lacks depth and context.