Signal Archive
资讯归档
按发布时间倒序浏览全部已发布资讯。
Archive
全部资讯
共 4428 条 · 每页 20 条

LLM 0.32 增加推理轨迹与 OpenAI Responses 支持
LLM 0.32 是一个重要版本,新增了可见的推理轨迹、服务端提供方工具、重新设计的内容寻址 SQLite 日志、新模型,以及由 OpenAI Responses API 支持的新功能。Simon Willison 还同步发布了更新后的 llm-anthropic 插件。
llm-anthropic 0.26 支持 Claude 5 和服务端工具
llm-anthropic 0.26 于 2026 年 8 月 4 日发布,新增对 Claude 5 系列模型的支持,包括 claude-fable-5、claude-sonnet-5 和 claude-opus-5。它还引入了 WebSearch、WebFetch、CodeExecution 和 AnthropicMCP 等服务端工具,并与 LLM 0.32 的新推理和思考机制对齐。

AMD’s datacenter business is booming while gaming takes a backseat
AMD’s AI-driven data center revenue surged to $6.7 billion in Q2 2026, offsetting a 31% decline in gaming revenue and making data center operations 58% of total sales.

Texas halts data center connections to power grid amid overwhelming demand
Texas has paused new data center grid connections while regulators audit projects amid concerns that rapid AI-driven development is overwhelming the state’s power infrastructure and affecting communities.

Open-weight AI models are catching up to the frontier. The safety gap remains. | TechCrunch
A SaferAI evaluation suggests China’s open-weight GLM-5.2 is nearing leading models in cyber and biological capabilities but lacks comparable refusal behavior and safety controls.

Anthropic signs $10B deal with AI cloud startup Volta | TechCrunch
Anthropic reportedly signed a six-year, $10 billion cloud-compute deal with Volta, backed by Bitdeer and Nvidia infrastructure, for a 133-megawatt AI data center in Norway.

Nvidia doesn't mess around: A week after open AI industry group formed, it's already showing progress | TechCrunch
Nvidia-led Open Secure AI Alliance has quickly formed a working group proposing open guidelines for confidential AI security incident reporting and collaborative analysis.

PipeNetwork/minimax-h3-mlx
PipeNetwork's minimax-h3-mlx package makes the MiniMax-H3 multimodal video-generation model runnable on Apple Silicon through MLX.
Third-party cyber evaluations involving OpenAI models
OpenAI describes recent third-party cybersecurity evaluation incidents and announces strengthened safeguards for testing its models.

‘Not healthy’ LLM use is more common than you think
The article argues that unhealthy reliance on LLMs may be more widespread than public debate acknowledges, using Hank Green’s AI controversy to examine psychological, ethical, and credibility concerns.

This year's Pulitzer Prizes saw a record number of winners disclose AI use
A record eight Pulitzer recipients and finalists disclosed using AI tools for tasks such as searching documents, translation, and classification, reflecting growing acceptance of carefully supervised newsroom AI use.

Google moves billions in Anthropic chip risk off its balance sheet
Google is using a complex, off-balance-sheet financing arrangement to provide Anthropic with up to $35 billion in TPU infrastructure while spreading substantial financial and infrastructure risk across partners and investors.

Texas halts new data centers as governor calls for audits | TechCrunch
Texas is requiring audits for all new data center projects after ERCOT's connection queue more than doubled to 474 gigawatts, with about 90% attributed to data centers.

Texas says data centers must pass an audit before connecting to the grid
Texas is requiring audits of new data center proposals before they can connect to the state grid, adding scrutiny around power, water use, incentives, and community impacts.

Elon Musk spends half his time talking robots and AI on Tesla earnings calls | TechCrunch
TechCrunch and Hudson Labs analyze seven years of Tesla earnings calls to show Elon Musk increasingly emphasizing AI, Optimus robots, and robotaxis despite cars remaining the company’s main business and revenue source.

Apple says more ex-employees may have taken confidential data to OpenAI | TechCrunch
Apple is escalating its trade secrets lawsuit against OpenAI and related parties, alleging more former employees may have been involved in taking confidential data tied to AI device development.

Run CI/CD for millions of repos — on your platform, on Cloudflare
Cloudflare announces a CI/CD workflow system that lets developers store, build, test, and deploy code on Cloudflare at scale, using Workflows, the CI SDK, and Artifacts.

Announcing Cloudflare Wallets: The programmable wallet for the agentic Internet
Cloudflare announces Wallets, a programmable wallet and unique handle system designed to help AI agents sign up for and pay for APIs more autonomously.

How we built a software factory to drive Astro’s GitHub issue count to zero
Cloudflare describes an automated AI-driven issue-triage pipeline for Astro that aims to reduce maintainer burden and has reportedly driven the repository’s GitHub issue count to zero.

Introducing: Cloudflare Agents
Cloudflare Agents is a new platform offering for deploying and managing hosted AI agents, starting with agent tracing and observability.