AI News & Insights
What changed, why it matters for your work, and what you can try this week.·Page 1 of 2
Meta has released Muse from Meta, an official Chrome extension. Instead of only using a remote browser on Meta's side, Muse now sits in the side panel and works in your own Chrome: filling in forms, comparing prices and chasing refunds on sites you're already logged into, and asking before it spends money. What it can see, how to start, how to stop it anytime, and the safe setup we recommend.
Hardware security researcher Kate has published work showing that NVIDIA's CMP 170HX mining card uses the same chip as the data center A100. She used voltage fault injection to bypass the fuse limits and unlock most cards to 80GB of memory and more compute units, with half-precision speed above the retail A100. The method is very hard to do and can destroy the card.
An open-source Agent Skill that lets AI agents like Claude Code and Codex write instrumental music note by note, render it with real instrument samples, and export MP3, MIDI, stems and an editable project. Includes seven example tracks, setup requirements, commercial limits and a ready-to-use prompt.
Google has released Gemini 4 Hydrogen to replace Gemini 3.8 Flash. Its overall score is slightly above Argon, it runs about 2.5 times faster, and it has a free tier in AI Studio and the Gemini API, with paid use at $1.50 per million input tokens. We cover the benchmarks, pricing, where to use it, how to migrate from older models, and a prompt you can give your AI agent to switch models for you.
Song Xiatian Dazzle (宋夏天Dazzle) released the map Skill Three Scope Map 3.0, which lets your AI agent generate runnable 3D globe and administrative-region map projects. Click from the globe into China, its provinces, cities, and districts or counties. Switch between 3D, 2D, and 2D layered over 3D views, recolor everything in one step, and use the new native React support. Note that it uses a custom license with platform restrictions, not an open-source license.
A rumor says GPT-6.2 Astra launches Thursday, but the graphic itself is labeled as an unconfirmed concept. We checked OpenAI's news page and Codex's open-source code: there is no GPT-6.2, and Daybreak is a cybersecurity access program, not a new model. Don't change your setup over a rumor.
Google's Gemini Decisions API and the new Gemini Decision 1 model handle closed questions: yes or no, pick one, or score on a scale. Up to 1000 questions per request, about 0.2 seconds per response, and free output. Useful for choosing tools for an AI agent, routing requests, and running safety checks. Includes a prompt you can hand to your AI agent to try it.
Meta's smallest and cheapest Muse model, Muse Ember, is built for repetitive work: tool calling, small coding tasks, routing, and extraction. It costs $0.10 per million input tokens and offers a 1M-token context window. Covers pricing and specs, how it compares with Claude Haiku 5.5 and GPT-6 Luna, and where it fits in an AI agent's workflow.
Gemini 4 Carbon is now open to everyone through the paid API and Google AI Ultra. It's built for long-running agent work and cybersecurity. Here's how it splits work with Argon, when the price doubles, and what Antigravity's /swarm does.
ERNIE 5.5 Flash Preview is live on Baidu's Qianfan platform and OpenRouter, with open weights promised by November 25. What it's good at, where it lags on benchmarks, and how to try it for free.
MiniMax released H3-Image, an open-weights image model that makes 4K pictures in about 10 seconds and accepts up to 16 reference images. Three ways to use it, what local hosting demands, and a prompt that lets your AI agent generate images for you.
Google's A2UI 1.0 lets an agent say what belongs on screen while your app paints it with its own native parts: forms, charts and confirmation cards. Jira and Alation already use it, and it runs in Claude and Gemini through MCP Apps. Includes a prompt that walks your AI agent through the quickstart.
Hermes Agent v0.23.0 adds a web UI (Webapp): the full Desktop app workspace in your browser, served by your own gateway. Sessions, group chats, plugins and files sync with the Desktop app. Off by default, three commands to turn on. Set up sign-in before opening it to the local network or internet. Includes a prompt to hand to your AI agent.
OpenAI launches Businesses in ChatGPT in beta. Verified businesses in Business Hub can see how often ChatGPT mentions them, track clicks and topics, and edit their own prices and stock. US e-commerce and online service businesses can add a Business Agent.
ByteDance's Seedream 5.5 can edit the same image for 100 rounds while unchanged areas stay pixel-level identical. It makes more realistic photos, writes small Chinese and English text more accurately, and exports transparent PNGs. Includes two prompt templates.
Anthropic's Claude Fable 5.2 is up to 30% faster, costs $8 per million input tokens and scores 49.1% on FrontierCode 2.0, though Opus 5.5 remains the everyday pick. Includes a table for picking the right model.
The open-source Poki project has released a preview iPhone app. An AI agent reads the screen, taps, and types on the phone locally. It takes commands from the app, iMessage, Telegram, or a terminal, and it can plug into Claude Code and Codex as an MCP server. Payments, deletions, and messages need your approval in the Dynamic Island first.
Meta has released Astryx, a React design system it used internally for eight years and that powers over 13,000 apps, on GitHub under the MIT license, with themes, templates, and a hosted MCP and command-line tool for AI coding assistants. It is currently in Beta and requires React 19.
The open-source project Horizon collects content from Hacker News, Reddit, X, GitHub, RSS and other sources, has AI score, deduplicate, and add background and community discussion, then generates a bilingual Chinese-English daily digest that can be published as a website or email, or pushed to Feishu or Slack. Includes a source list, what you need to prepare, and notes.
Someone has put together 10 reverse-engineering Skills and integration solutions for Agents: comprehensive analysis, scenario routing, Android interfaces, product analysis, lightweight command line, CTF, IDA assistance, JS signatures, and an all-in-one lab. We have added all of them to our site's Skill pages, with selection advice, license differences, and authorization reminders.