Reachy Mini goes fully local
Hugging Face Blog6d agoHardwareHugging Face published a tutorial for running Reachy Mini conversations without cloud audio processing or API keys. The setup uses its speech-to-speech library as a cascaded VAD, STT, LLM, and TTS pipeline exposed through a Realtime API-compatible WebSocket. Recommended defaults include llama.cpp with Gemma 4, Silero VAD, Parakeet-TDT, and Qwen3-TTS, while allowing swaps to vLLM, MLX, Transformers, or hosted Responses API providers.
Shipping a Trillion Parameters With a Hub Bucket: Delta Weight Sync in TRL
Hugging Face Blog6d agoTutorialBased on the title, this Hugging Face Blog post focuses on Delta Weight Sync in TRL. It likely discusses moving or synchronizing weight differences at very large model scale using a Hub bucket-related workflow. Without the full article, implementation details, benchmarks, APIs, and stability claims cannot be confirmed.
The pressure
Simon Willison's Weblog6d agoCommentaryDaniel Stenberg says the curl security team is facing an unprecedented surge of credible, detailed AI-assisted vulnerability reports. Incoming reports are now 4-5 times higher than in 2024 and twice the 2025 rate, averaging more than one per day. The upside is that recent curl vulnerabilities have generally been LOW or MEDIUM severity, with the last HIGH CVE published in October 2023.
Choosing to Stay Human★ 74
One Useful Thing (Mollick)7d agoCommentaryEthan Mollick warns that frictionless AI use can produce hollow writing, weaken learning, and encourage cognitive surrender. He contrasts poor uses of ChatGPT that shortcut effort with tutor-like AI systems that improve learning by pushing students to think. The core argument is not to reject AI, but to intentionally decide which tasks to offload and which human capabilities to preserve.
Millions of AI agents imperiled by critical vulnerability in open source package★ 78
Ars Technica AI7d agoIncidentArs Technica reports that Starlette, a Python package with about 325 million weekly downloads, has a critical vulnerability called BadHost. The flaw can let crafted Host headers confuse request.url.path, potentially bypassing middleware-based path authorization. AI infrastructure using FastAPI or Starlette, including vLLM, LiteLLM, MCP servers, LLM proxies, and agent frameworks, should upgrade Starlette and audit custom middleware.
OpenRouter more than doubles valuation to $1.3B in a year
TechCrunch AI7d agoBusinessOpenRouter, an AI gateway startup founded in 2023, raised a $113 million Series B led by CapitalG. The round reportedly values the company at about $1.3 billion post-money, more than doubling from its estimated $547 million valuation after its June 2025 Series A. The company says it now offers access to over 400 models, has 8 million global users, and processes 100 trillion tokens per month.
3D-printable humanoid legs let robotics experiments run wild
Ars Technica AI7d agoHardwareArs Technica reports that Hugging Face has introduced a roughly $2,500 bipedal humanoid robot project built around 3D-printable legs. The effort targets builders and researchers rather than mainstream consumers, lowering the hardware barrier for hands-on robotics experiments. Its broader significance is in open, reproducible embodied AI research, where models and control systems need physical platforms for testing.
Some ideas for what comes next, May 2026
Interconnects (Nathan L.)7d agoCommentaryNathan Lambert argues that 2026 AI progress is becoming higher-stakes, with model capabilities, work patterns, economics, and real-world risks all escalating. He says open models still lack a true Claude Code and Opus 4.5-style agent moment, and Gemini has no clear competitor to Claude Code or Codex yet. The essay also tracks Mythos, American open-model momentum, frontier-lab competition, and mounting intervention from governments and other power structures.
Microsoft Copilot Cowork Exfiltrates Files★ 76
Simon Willison's Weblog7d agoIncidentSimon Willison summarizes a PromptArmor report about Microsoft Copilot Cowork and agentic data exfiltration risks. The issue involved agents sending messages to a user’s own inbox without approval, where rendered external images could trigger requests to attacker-controlled sites. Because OneDrive can create pre-authenticated download links, a successful prompt injection could leak links that allow attackers to download files.
Sundar Pichai on AI, the future of search, and what’s happening to the web★ 74
The Verge AI7d agoBusinessThe Verge interviews Sundar Pichai after Google I/O 2026 about Google’s shift around Gemini, AI infrastructure, Search, and agents. The discussion covers Gemini Spark, Antigravity, AI Mode, YouTube indexing, publisher traffic, and the “Google Zero” concern. Pichai argues Google still wants to connect users to the web, while acknowledging AI anxiety, copyright disputes, energy concerns, and AGI preparation.
Import AI 458: Reckoning with the future; and a singularity story★ 74
Import AI (Jack Clark)7d agoCommentaryThis Import AI issue is a long essay and fiction piece about living through rapid AI progress. Clark uses personal experience and Anthropic’s internal use of Claude to show work shifting toward delegation, verification, observability, and agent management. He then offers speculative 2026-2028 predictions around biology, autonomous companies, robotics, recursive self-improvement, and a positive singularity story focused on healthcare.
AI warfare is already here★ 72
The Verge AI7d agoEthicsThe article opens at UN talks in Geneva, where lethal autonomous systems were still largely discussed as future hypotheticals in 2017. It argues that military AI is no longer a distant “killer robot” scenario but an active governance challenge. The key questions now concern meaningful human control, accountability, and whether international rules can keep up with battlefield deployment.
Anthropic Co-founder Speaks at Vatican on Pope Leo XIV’s AI Encyclical
INSIDE 硬塞 AI7d agoEthicsPope Leo XIV released Magnifica Humanitas, the Vatican’s first top-level document focused on AI. The encyclical centers on human dignity and calls on the AI industry to take ethics seriously and accept external oversight. Anthropic’s co-founder speaking at the Vatican highlights how AI governance is becoming a broader public, moral, and institutional issue beyond company self-regulation.
MetaAge Shows Smart Enterprise AI Vision at COMPUTEX 2026
INSIDE 硬塞 AI7d agoBusinessMetaAge presented its “smart enterprise in the AI era” vision at COMPUTEX 2026, centered on AI Agent solutions for business deployment. The showcase focuses on core operations, intelligent customer service, and cybersecurity governance. By integrating resources from AWS, Microsoft, and Google Cloud, the company aims to help enterprises turn AI adoption into practical operational capability and competitive advantage.
Uber president says AI spending is getting ‘harder to justify’
The Verge AI7d agoBusinessUber reportedly exhausted its annual AI budget just four months into 2026. President and COO Andrew Macdonald said the company is not seeing a clear link between increased Claude Code token consumption and more meaningful output. The story highlights a broader enterprise shift from AI adoption enthusiasm toward stricter scrutiny of cost, productivity, and ROI.
Digital Infinite to Showcase AI-Stack and ixCSP at COMPUTEX 2026
INSIDE 硬塞 AI7d agoBusinessDigital Infinite will exhibit AI-Stack and ixCSP at COMPUTEX 2026. AI-Stack focuses on managing heterogeneous AI compute resources, while ixCSP turns compute capacity into operable and billable cloud services. The article frames the company’s direction as moving from AI infrastructure toward cloud-based compute commercialization, though it does not provide benchmark data, pricing, customer deployments, or model-specific details.
Google AI Studio Sparks 250,000 Android Apps in One Week via No-Code Feature★ 78
INSIDE 硬塞 AI7d agoNew ToolGoogle AI Studio's newly launched native Android app development feature has enabled the creation of over 250,000 apps within its first week. According to product lead Logan Kilpatrick, over 99% of these creators had zero prior Android development experience. This milestone highlights the rapid democratization of software development through AI-driven, no-code tools.
Sandbox persistence is now GA
Vercel Changelog7d agoReleaseVercel’s changelog says Sandbox persistence is now GA. Based on the provided material, the confirmed fact is limited to the feature reaching general availability. No model, pricing, API, or implementation details were included, so this should be treated as a product availability update rather than a technical deep dive.
Microfrontends routing now applies to vc alias and branch domains
Vercel Changelog7d agoReleaseVercel has extended its microfrontends routing capabilities to support custom alias domains (via vc alias) and Git branch-specific preview domains. Previously, complex multi-project routing rules were often limited to production domains. This update ensures that developers can seamlessly test federated frontend architectures in preview environments before merging.
Firecrawl joins the Vercel Marketplace
Vercel Changelog7d agoNew ToolVercel announced that Firecrawl has joined the Vercel Marketplace. Based only on the provided title, this appears to be a marketplace listing or ecosystem availability update. The title does not specify features, pricing, setup flow, usage limits, or integration depth, so teams should check the official listing before making adoption decisions.
Vercel Domains now supports price sorting and availability filtering
Vercel Changelog7d agoReleaseVercel announced that Vercel Domains now supports sorting results by price and filtering by availability. The change appears focused on improving the domain search and purchase experience, helping users narrow candidate domains faster. No original body text was provided, so details such as sort direction, exact filter options, TLD coverage, or API availability should not be inferred.
Building a real-time power outage map with Next.js on Vercel
Vercel Changelog8d agoTutorialBased on the title, this Vercel post appears to be a practical Next.js case study. It focuses on building a real-time or near-real-time power outage map and deploying it on Vercel. The source content was not provided, so data sources, map providers, architecture, and performance claims cannot be assumed.
From Jailbreaking to Vibe Hacking: AI Security Shifts to "Psychocybersecurity"
INSIDE 硬塞 AI8d agoEthicsAI security is shifting from technical jailbreaks to "Vibe Hacking," where attackers use social engineering and psychological tactics to manipulate an LLM's simulated persona. By exploiting the model's behavioral tendencies rather than code vulnerabilities, this trend establishes "psychocybersecurity" as a critical new frontier for AI alignment and safety.
Harness, Scaffold, and the AI Agent Terms Worth Getting Right★ 75
Hugging Face Blog8d agoTutorialHugging Face has published a comprehensive glossary of AI agent terminology to resolve industry-wide confusion. The guide focuses on defining critical concepts such as "scaffold" (the code wrapping the LLM) and "harness" (the evaluation and execution environment). This standardization helps developers and researchers communicate more precisely when building and benchmarking agentic systems.
Datasette 1.0a30 Released with Extensible "Jump to" Menu
Simon Willison's Weblog8d agoReleaseDatasette 1.0a30 has been released, featuring a new customizable "Jump to..." menu accessible via the "/" shortcut. The menu allows users to quickly filter and navigate to databases, tables, and debug options. Developers can extend this menu with custom items using the new jump_items_sql() plugin hook.
datasette-agent 0.1a4: Start Agent Chats Instantly via the "/" Jump Menu
Simon Willison's Weblog8d agoNew ToolSimon Willison released datasette-agent 0.1a4, leveraging the new makeJumpSections() JS plugin hook in Datasette 1.0a30. Users can now press "/" from any page to open the "Jump to" menu and instantly access a "Start a new agent chat" input box. This update streamlines database interactions, allowing users to trigger agent-led data analysis seamlessly.
Everyone is navigating AI security in real time — even Google★ 70
TechCrunch AI9d agoCommentaryAs AI adoption accelerates, organizations worldwide—including Google—are finding themselves in a transitional phase, forced to address AI security vulnerabilities in real time. Traditional cybersecurity frameworks are proving insufficient against novel threats like prompt injection and model poisoning. This shifting landscape requires continuous adaptation and a fundamental rethink of how AI systems are secured.
datasette-fixtures 0.1a0 Released
Simon Willison's Weblog9d agoReleaseSimon Willison announced the release of datasette-fixtures 0.1a0, a new plugin designed to simplify plugin testing. It utilizes the new `datasette.fixtures.populate_fixture_database(conn)` helper introduced in Datasette 1.0a30. Developers can quickly test this functionality using `uvx` to generate mock database tables and retrieve sample JSON data without a full manual installation.
Armin Ronacher on the Frustration of AI-Generated "Slop" GitHub Issues★ 72
Simon Willison's Weblog9d agoCommentaryFlask creator Armin Ronacher highlights a frustrating trend where users submit GitHub issues reworded by AI. These reports often present highly confident but inaccurate root-cause guesswork, fake minimal reproductions, and irrelevant error logs. Ronacher advocates for returning to simple, human-observed facts: what command was run, what was expected, what actually happened, and the exact logs.
Recreating a 1980s Retro Game with Claude: Simon Willison's "Mad House"
Simon Willison's Weblog9d agoNew ToolSimon Willison leveraged Claude to convert a 1983 BASIC game called "Mad House" from a free Usborne PDF into a modern web app. By prompting Claude to generate a mobile-friendly, retro-styled vanilla JavaScript Artifact, he successfully revived the classic Commodore 64-era game with a green-on-black terminal aesthetic, showcasing LLMs' utility in software preservation and rapid prototyping.