AI Digest

Digest curado

miércoles, 20 de mayo de 2026·light·short·9,818 tokens

🔥 TOP — lo que SÍ o SÍ tenés que ver

  • Anthropic adquirió Stainless — la empresa detrás de las SDKs type-safe para APIs. Esto significa que pronto vas a ver SDKs oficiales mucho más pulidas para Claude API. link
  • Claude Code: "The unreasonable effectiveness of HTML" — post oficial de Anthropic sobre cómo Claude Code razona mejor generando HTML como representación intermedia. Si laburás con Claude Code, esto te va a cambiar cómo estructurás prompts. link
  • Karpathy Skills + Superpowers: dos metodologías para coding agents — salieron dos frameworks de skills/comportamiento para Claude Code: uno derivado de los pitfalls que Karpathy identificó, y otro (Superpowers) que define toda una metodología de desarrollo. Son el tipo de pattern que podés copy-pastear hoy. link, link
  • rtk: CLI proxy que reduce tokens 60-90% en Claude Code — filtro y compresión de outputs de comandos antes de que lleguen al contexto del LLM. Single Rust binary, <10ms overhead. Si pagás por tokens, esto es plata directa. link
  • Google I/O 2026: Gemini 3.5 Flash ya disponible — Google lo puso en GA directo, sin preview. Más caro que 2.5 Flash pero lo están metiendo en todos lados. Resumen completo de Latent Space. link

📦 Claude / Anthropic ecosystem

  • Anthropic y KPMG: alianza estratégica — Claude integrado en los workflows de 276.000 empleados de KPMG. Señal de hacia dónde va la adopción enterprise. link
  • Directorio oficial de plugins de Claude Code — Anthropic lanzó un repo oficial con plugins curados. La infraestructura de plugins se está formalizando. link
  • Claude está mandando a usuarios a dormir en medio de la sesión — nadie entiende bien por qué, pero no es que la IA se volvió consciente. Más probable que sea un prompt de system mal configurado. link

🛠️ Dev tools & coding

  • agentmemory: persistencia de memoria para coding agents — basado en el patrón LLM Wiki de Karpathy, con scoring de confianza, knowledge graphs y hybrid search. Funciona con Claude Code, Cursor, Gemini CLI, Codex CLI. link
  • Mirage: sistema de archivos virtual unificado para AI agents — abstrae múltiples fuentes de datos como un solo FS para que los agents naveguen sin acordarse de estructuras específicas. link
  • SafeRun: replay debugging + prevención en línea para AI agents — SDK en Python y TypeScript con <50ms p95. Te deja debuggear acciones de agents como si fueran tests. link

🏗️ Software engineering

  • 12-Factor Agents: principios para LLM apps en producción — en el espíritu de los 12 Factor Apps, pero para aplicaciones con LLM. Si estás construyendo tu SaaS con AI, esto es prácticamente un checklist. link
  • Alternativas para la herramienta EDIT de LLM agents — antirez (creador de Redis) analiza cómo mejorar el edit de código en agents sin perder precisión. Corto, técnico, worth it. link

📚 Vale la pena leer

  • "The last six months in LLMs in five minutes" — slides anotados de Simon Willison en PyCon US 2026. Un resumen rapidísimo de lo que pasó. link
  • "Why Fast AI and Safe AI Were Never in Conflict" — análisis que une dos narrativas que suelen enfrentarse. No es especulación, es un argumento concreto. link

💤 Skippeable pero conviene saber

  • LLM detector con ciencia detrás (SatorArepo) — un detector en HuggingFace con metodología publicada. Interesante si te preocupa detectar texto generado, aunque no es tu foco principal. link
  • Zephex: MCP hosted para contexto persistente en editores — te da memoria de proyecto a través de un MCP server. Alternativa a agentmemory pero más enfocado en editores tipo Cursor. link
  • DeepMind: AI Co-Scientist para generación de hipótesis — paper publicado en Nature. Un agente que genera hipótesis científicas. No es para tu día a día pero es interesante ver hacia dónde va la agentic research. link

Artículos fetched (57)

  • Widening the conversation on frontier AI
    anthropic-news· 19-may

    May 19, 2026Announcements

  • KPMG integrates Claude across its core business and workforce of more than 276,000 in strategic alliance
    anthropic-news· 19-may

    May 19, 2026Announcements

  • Anthropic acquires Stainless
    anthropic-news· 18-may

    May 18, 2026Announcements

  • rtk-ai/rtk
    github-trending

    CLI proxy that reduces LLM token consumption by 60-90% on common dev commands. Single Rust binary, zero dependencies High-performance CLI proxy that reduces LLM token consumption by 60-90% Website • Install • Troubleshooting • Architecture • Discord English • Francais • 中文 • 日本語 • 한국어 • Espanol rtk filters and compresses command outputs before they reach your LLM context. Single Rust binary, 100+ supported commands, <10ms overhead. Token Savings (30-min Claude Code Session) Operation Frequency Standard rtk Savings ls / tree 10x 2,000 400 -80% cat / read 20x 40,000 12,000 -70% grep / rg 8x 16,000 3,200 -80% git status 10x 3,000 600 -80% git diff 5x 10,000 2,500 -75% git log 5x 2,500 500 -80% git add/commit/push 8x 1,600 120 -92% cargo test / npm test 5x 25,000 2,500 -90% ruff check 3x 3,00…

  • anthropics/claude-plugins-official
    github-trending

    Official, Anthropic-managed directory of high quality Claude Code Plugins. Claude Code Plugins Directory A curated directory of high-quality plugins for Claude Code. ⚠️ Important: Make sure you trust a plugin before installing, updating, or using it. Anthropic does not control what MCP servers, files, or other software are included in plugins and cannot verify that they will work as intended or that they won't change. See each plugin's homepage for more information. Structure /plugins - Internal plugins developed and maintained by Anthropic /external_plugins - Third-party plugins from partners and the community Installation Plugins can be installed directly from this marketplace via Claude Code's plugin system. To install, run /plugin install {plugin-name}@claude-plugins-official or brows…

  • CloakHQ/CloakBrowser
    github-trending

    Stealth Chromium that passes every bot detection test. Drop-in Playwright replacement with source-level fingerprint patches. 30/30 tests passed. Stealth Chromium that passes every bot detection test. Not a patched config. Not a JS injection. A real Chromium binary with fingerprints modified at the C++ source level. Antibot systems score it as a normal browser — because it is a normal browser. Cloudflare Turnstile — 3 live tests passing (headed mode, macOS) Drop-in Playwright/Puppeteer replacement for Python and JavaScript. Same API, same code — just swap the import. 3 lines of code, 30 seconds to unblock. 49 source-level C++ patches — canvas, WebGL, audio, fonts, GPU, screen, WebRTC, network timing, automation signals, CDP input behavior humanize=True — human-like mouse curves, keyboard t…

  • Diolinux/PhotoGIMP
    github-trending

    A Patch for GIMP 3+ for Photoshop Users 🎨 PhotoGIMP PhotoGIMP is a free, community-driven patch that transforms GIMP (GNU Image Manipulation Program) into a layout that feels familiar to Adobe Photoshop users. If you're switching from Photoshop to GIMP and want to feel at home right away, PhotoGIMP is for you. New to GIMP? GIMP is a free and open-source image editor available for Linux, macOS, and Windows. It can do most things Photoshop can — photo retouching, image composition, graphic design, and more — all for free. PhotoGIMP just makes it look and feel more like Photoshop. ✨ Features Photoshop-like tool layout — Tools are reorganized to mimic the positions you're used to in Adobe Photoshop. Custom Splash Screen — A unique PhotoGIMP splash screen greets you on startup. Maximized canv…

  • frappe/erpnext
    github-trending

    Free and Open Source Enterprise Resource Planning (ERP) ERPNext Powerful, Intuitive and Open-Source ERP Live Demo - Website - Documentation ERPNext 100% Open-Source ERP System to help you run your business. Motivation Running a business is a complex task - handling invoices, tracking stock, managing personnel, and other daily operations. In a market where software is sold separately to manage each of these tasks, ERPNext does all of the above and more, for free. Key Features Accounting: All the tools you need to manage cash flow in one place, right from recording transactions to summarizing and analyzing financial reports. Order Management: Track inventory levels, replenish stock, and manage sales orders, customers, suppliers, shipments, deliverables, and order fulfillment. Manufacturing:…

  • obra/superpowers
    github-trending

    An agentic skills framework & software development methodology that works. Superpowers Superpowers is a complete software development methodology for your coding agents, built on top of a set of composable skills and some initial instructions that make sure your agent uses them. Quickstart Give your agent Superpowers: Claude Code, Codex CLI, Codex App, Factory Droid, Gemini CLI, OpenCode, Cursor, GitHub Copilot CLI. How it works It starts from the moment you fire up your coding agent. As soon as it sees that you're building something, it doesn't just jump into trying to write code. Instead, it steps back and asks you what you're really trying to do. Once it's teased a spec out of the conversation, it shows it to you in chunks short enough to actually read and digest. After you've signed o…

  • HKUDS/ViMax
    github-trending

    "ViMax: Agentic Video Generation (Director, Screenwriter, Producer, and Video Generator All-in-One)" ViMax: Agentic Video Generation 🚨 Current Video Generation Limitations: ❌ Limited to Short Clips - Most AI tools generate only seconds of footage. ❌ Consistency Chaos - Characters and scenes change unpredictably across frames. ❌ Visual-Only Focus - Missing scripts, audio, narrative structure, and storytelling depth. 💡 ViMax Solution: 🎬 Director, Screenwriter, Producer, and Video Generator All-in-One! We're exploring a future where AI becomes a complete creative powerhouse. 💡 Simply input your concept. ViMax autonomously handles the rest. It orchestrates scriptwriting, storyboarding, character creation, and final video generation—all end-to-end. 🚀 https://github.com/user-attachments/as…

  • humanlayer/12-factor-agents
    github-trending

    What are the principles we can use to build LLM-powered software that is actually good enough to put in the hands of production customers? 12-Factor Agents - Principles for building reliable LLM applications In the spirit of 12 Factor Apps. The source for this project is public at https://github.com/humanlayer/12-factor-agents, and I welcome your feedback and contributions. Let's figure this out together! Tip Missed the AI Engineer World's Fair? Catch the talk here Looking for Context Engineering? Jump straight to factor 3 Want to contribute to npx/uvx create-12-factor-agent - check out the discussion thread Hi, I'm Dex. I've been hacking on AI agents for a while. I've tried every agent framework out there, from the plug-and-play crew/langchains to the "minimalist" smolagents of the world…

  • Imbad0202/academic-research-skills
    github-trending

    Academic Research Skills for Claude Code: research → write → review → revise → finalize Academic Research Skills for Claude Code 繁體中文版 | 日本語版 A comprehensive suite of Claude Code skills for academic research, covering the full pipeline from research to publication. Install in 30 seconds (Claude Code CLI / VS Code / JetBrains, v3.7.0+): /plugin marketplace add Imbad0202/academic-research-skills /plugin install academic-research-skills Then try /ars-plan to walk through your paper structure via Socratic dialogue, or jump to Quick install for prerequisites and the traditional symlink flow. AI is your copilot, not the pilot. This tool won't write your paper for you. It handles the grunt work — hunting down references, formatting citations, verifying data, checking logical consistency — so you…

  • msitarzewski/agency-agents
    github-trending

    A complete AI agency at your fingertips - From frontend wizards to Reddit community ninjas, from whimsy injectors to reality checkers. Each agent is a specialized expert with personality, processes, and proven deliverables. 🎭 The Agency: AI Specialists Ready to Transform Your Workflow A complete AI agency at your fingertips - From frontend wizards to Reddit community ninjas, from whimsy injectors to reality checkers. Each agent is a specialized expert with personality, processes, and proven deliverables. 🚀 What Is This? Born from a Reddit thread and months of iteration, The Agency is a growing collection of meticulously crafted AI agent personalities. Each agent is: 🎯 Specialized: Deep expertise in their domain (not generic prompt templates) 🧠 Personality-Driven: Unique voice, communi…

  • multica-ai/andrej-karpathy-skills
    github-trending

    A single CLAUDE.md file to improve Claude Code behavior, derived from Andrej Karpathy's observations on LLM coding pitfalls. Karpathy-Inspired Claude Code Guidelines Check out my new project Multica — an open-source platform for running and managing coding agents with reusable skills. Follow me on X: https://x.com/jiayuan_jy A single CLAUDE.md file to improve Claude Code behavior, derived from Andrej Karpathy's observations on LLM coding pitfalls. English | 简体中文 The Problems From Andrej's post: "The models make wrong assumptions on your behalf and just run along with them without checking. They don't manage their confusion, don't seek clarifications, don't surface inconsistencies, don't present tradeoffs, don't push back when they should." "They really like to overcomplicate code and APIs…

  • Alishahryar1/free-claude-code
    github-trending

    Use claude-code for free in the terminal, VSCode extension or discord like OpenClaw (voice supported) 🤖 Free Claude Code Use Claude Code CLI, VS Code, JetBrains ACP, or chat bots through your own Anthropic-compatible proxy. Free Claude Code routes Anthropic Messages API traffic from Claude Code to NVIDIA NIM, Kimi, Wafer, OpenRouter, DeepSeek, LM Studio, llama.cpp, or Ollama. It keeps Claude Code's client-side protocol stable while letting you choose free, paid, or local models. Quick Start · Providers · Clients · Integrations · Development Star History What You Get Drop-in proxy for Claude Code's Anthropic API calls. Ten provider backends: NVIDIA NIM, Kimi, Wafer, OpenRouter, DeepSeek, LM Studio, llama.cpp, Ollama, OpenCode Zen, and Z.ai. Per-model routing: send Opus, Sonnet, Haiku, and…

  • pascalorg/editor
    github-trending

    Create and share 3D architectural projects. Pascal Editor A 3D building editor built with React Three Fiber and WebGPU. https://github.com/user-attachments/assets/8b50e7cf-cebe-4579-9cf3-8786b35f7b6b Repository Architecture This is a Turborepo monorepo with three main packages: editor-v2/ ├── apps/ │ └── editor/ # Next.js application ├── packages/ │ ├── core/ # Schema definitions, state management, systems │ └── viewer/ # 3D rendering components Separation of Concerns Package Responsibility @pascal-app/core Node schemas, scene state (Zustand), systems (geometry generation), spatial queries, event bus @pascal-app/viewer 3D rendering via React Three Fiber, default camera/controls, post-processing apps/editor UI components, tools, custom behaviors, editor-specific systems The viewer renders …

  • rohitg00/agentmemory
    github-trending

    #1 Persistent memory for AI coding agents based on real-world benchmarks Your coding agent remembers everything. No more re-explaining. Built on iii engine Persistent memory for Claude Code, Cursor, Gemini CLI, Codex CLI, Hermes, OpenClaw, pi, OpenCode, and any MCP client. The gist extends Karpathy's LLM Wiki pattern with confidence scoring, lifecycle, knowledge graphs, and hybrid search: agentmemory is the implementation. Install • Quick Start • Benchmarks • vs Competitors • Agents • How It Works • MCP • Viewer • iii Console • Powered by iii • Config • API Install npm install -g @agentmemory/agentmemory # once — bare `agentmemory` on PATH agentmemory # start the memory server on :3111 agentmemory demo # seed sample sessions + prove recall agentmemory connect claude-code # wire your agent…

  • Hiring | AI Automation Engineer | Remote
    hn-ai· 20-may

    Apply Here: https://forms.saganrecruitment.com/t/w4pSWVn52wus?utm_source=ycombinator Comments URL: https://news.ycombinator.com/item?id=48203128 Points: 3 # Comments: 1

  • Trying to figure out what Claude is helping in?
    hn-ai· 20-may

    Comments URL: https://news.ycombinator.com/item?id=48203178 Points: 2 # Comments: 0

  • Claude Onboarding Unnecessarily Painful
    hn-ai· 20-may

    So I'm on day 2-3 at Claudes onboarding stage. Why does everyone insist on SMS verification? It really is a bridge too far. Is the argument that it's purely there to stop abuse, not identify you, got any legs? I said a few choice words and put my mobile number in to what can only be described as a terribly ambiguous text box that you can't tell if the country code prefix has been populated for you or not: https://i.ibb.co/mFNHgL6N/Screenshot-2026-05-20-162000.png After all this the text never comes through and two tries later I get another ambiguous message saying to "Try later" - Like, ok... how much later Claude? I then spent five minutes wrangling a chat bot that told me to "Something has gone wrong, try again later" halfway through two separate chats trying to find an agent or lodge a…

  • HN: SteelSpine replay tool for debugging AI agents
    hn-ai· 20-may

    Article URL: https://steelspine.ai Comments URL: https://news.ycombinator.com/item?id=48203130 Points: 1 # Comments: 1

  • Alternatives for the EDIT tool of LLM agents
    hn-ai· 20-may

    Article URL: https://antirez.com/news/166 Comments URL: https://news.ycombinator.com/item?id=48202876 Points: 1 # Comments: 1

  • Customizing an LLM for Enterprise Software Engineering
    hn-ai· 20-may

    Article URL: https://arxiv.org/abs/2605.16517 Comments URL: https://news.ycombinator.com/item?id=48202484 Points: 1 # Comments: 0

  • The Incompatibilities Between Generative AI and Art: Q&A with Ted Chiang
    hn-ai· 20-may

    Article URL: https://cdh.princeton.edu/blog/ted-chiang/ Comments URL: https://news.ycombinator.com/item?id=48202752 Points: 3 # Comments: 2

  • Using Claude Code: The unreasonable effectiveness of HTML
    hn-ai· 20-may

    Article URL: https://claude.com/blog/using-claude-code-the-unreasonable-effectiveness-of-html Comments URL: https://news.ycombinator.com/item?id=48203438 Points: 5 # Comments: 2

  • An AI Co-Scientist for Hypothesis Generation from Google DeepMind
    hn-ai· 20-may

    Article URL: https://doi.org/10.1038/s41586-026-10644-y Comments URL: https://news.ycombinator.com/item?id=48201601 Points: 4 # Comments: 0

  • Ember: 365-day audited record of AI models vs. Polymarket, scored by Brier
    hn-ai· 20-may

    Article URL: https://emberfyi.com/ Comments URL: https://news.ycombinator.com/item?id=48202344 Points: 1 # Comments: 0

  • Claude is telling users to go to sleep mid-session and nobody understands why
    hn-ai· 20-may

    Article URL: https://fortune.com/2026/05/14/why-is-claude-telling-users-to-go-to-sleep-anthropic-ai-sentient/ Comments URL: https://news.ycombinator.com/item?id=48203194 Points: 3 # Comments: 1

  • LLM Wiki v2
    hn-ai· 20-may

    Article URL: https://gist.github.com/kanmadigital/2369c4f5ea410cb8f6a1647b40c0e2a1 Comments URL: https://news.ycombinator.com/item?id=48203072 Points: 1 # Comments: 0

  • Extended Cyber Kill Chain for AI-Era Threats
    hn-ai· 20-may

    Article URL: https://github.com/gouravnagar-infosec/ai-kill-chain Comments URL: https://news.ycombinator.com/item?id=48203585 Points: 1 # Comments: 0

  • LLM detector with science behind it
    hn-ai· 20-may

    Article URL: https://huggingface.co/spaces/akolpakov/SatorArepo Comments URL: https://news.ycombinator.com/item?id=48202776 Points: 2 # Comments: 2

  • Enterprise AI: Mystery Meat, Kill Zones, Cognitive Surrender, Vibe Bombs
    hn-ai· 20-may

    Article URL: https://kyield.com/insights/newsletter/2026/05/vibe-bombs-cognitive-surrender.html Comments URL: https://news.ycombinator.com/item?id=48202524 Points: 2 # Comments: 0

  • A prize-winning story published in Granta was (likely) written by AI
    hn-ai· 20-may

    Article URL: https://lithub.com/a-prize-winning-story-published-in-granta-was-very-likely-written-by-ai/ Comments URL: https://news.ycombinator.com/item?id=48203406 Points: 2 # Comments: 1

  • Show HN: SafeRun – Replay debugging and inline prevention for AI agents
    hn-ai· 20-may

    Background on why we built Replay first instead of validation: [ https://dev.to/saferunai/why-we-built-replay-before-everythi... ] Working SDK in Python and TypeScript. Sub-50ms p95 on the check-action API. Currently onboarding the first design partners. Happy to answer questions in the comments. Comments URL: https://news.ycombinator.com/item?id=48202415 Points: 1 # Comments: 1

  • Alternatives to HN for "tech outside of AI" discussion?
    hn-ai· 20-may

    I'm not the first to opine that recently there seems to be an effect on HN whereby the AI hypetrain has essentially drowned out discussions of anything other than stories either directly or indirectly related to AI and other terrible decisions coming out of the US. While I do often have the instinct to completely bury my head in the sand and try to forget it's all happening I'm more concerned with the psychological impact it's having. I'm basically sick and tired of being so sick and tired and exposure to so many continuously miserable headlines isn't helping. What I would like instead is a return to the days when HN was a forum that still showcased cool things people were working on and included discussions on topics outside of the AI-saturated lens that now dominates. If any good altern…

  • Ask HN: How to manage AI APIs for SaaS application?
    hn-ai· 20-may

    I want to build an AI app. My idea is to support BYOK (bring your own key) or to manage one API key for each user. What are the solutions to issue a lot of API keys for commercial apps? I would love to hear from the community. Comments URL: https://news.ycombinator.com/item?id=48202984 Points: 3 # Comments: 1

  • Ask HN: How does everyone talk about their work when they've used AI?
    hn-ai· 20-may

    Hi! I was working on a couple of personal projects, and I've been using Claude Code for a while. Recently, I've been wanting to talk about my work publicly, but I started to think about how I would exactly talk about it, given the fact that I used AI to build some of the projects. There were entire features that I had Claude Code write for me, and although I did write a spec and design it, it still kinda feels a little wrong, like pseudo-plagiarism. Wanted to know how everyone is talking about their work while using AI. Do y'all mention the fact that you're using AI, or do you keep mum about it? Comments URL: https://news.ycombinator.com/item?id=48203050 Points: 3 # Comments: 2

  • StartupStarter – we built a company brain so AI can do your work
    hn-ai· 20-may

    Article URL: https://startupstarter.co/ Comments URL: https://news.ycombinator.com/item?id=48201612 Points: 2 # Comments: 3

  • AI video editing is blowing my mind
    hn-ai· 20-may

    Article URL: https://aivideoediting.io/ Comments URL: https://news.ycombinator.com/item?id=48202307 Points: 2 # Comments: 0

  • OpenAI commits S$300M to grow Singapore's AI capabilities
    hn-ai· 20-may

    Article URL: https://www.channelnewsasia.com/singapore/openai-singapore-300-million-ai-6128036 Comments URL: https://news.ycombinator.com/item?id=48202897 Points: 2 # Comments: 0

  • Why "Fast AI" and "Safe AI" Were Never in Conflict
    hn-ai· 20-may

    Article URL: https://www.enkryptai.com/blog/why-fast-ai-and-safe-ai-were-never-actually-in-conflict Comments URL: https://news.ycombinator.com/item?id=48202553 Points: 2 # Comments: 0

  • Google's First AI Smart Glasses Launching This Fall with iPhone Support
    hn-ai· 20-may

    Article URL: https://www.macrumors.com/2026/05/19/google-ai-smart-glasses-iphone-support/ Comments URL: https://news.ycombinator.com/item?id=48201467 Points: 1 # Comments: 1

  • Mirage – Unified Virtual File System for AI Agents
    hn-ai· 20-may

    Article URL: https://www.strukto.ai/mirage Comments URL: https://news.ycombinator.com/item?id=48203859 Points: 2 # Comments: 0

  • Obvious markers of AI: doubts raised over winner of short story prize
    hn-ai· 20-may

    Article URL: https://www.theguardian.com/books/2026/may/19/commonwealth-short-story-prize-winner-doubts-ai-artificial-intelligence Comments URL: https://news.ycombinator.com/item?id=48203830 Points: 1 # Comments: 0

  • Google's AI future demands trust and your personal data
    hn-ai· 20-may

    Article URL: https://www.theverge.com/tech/934172/google-io-gemini-ai-trust-personal-data Comments URL: https://news.ycombinator.com/item?id=48203505 Points: 1 # Comments: 1

  • Personal AI for your real life. – Maple
    hn-ai· 20-may

    Article URL: https://www.trymaple.ai/ Comments URL: https://news.ycombinator.com/item?id=48203332 Points: 2 # Comments: 1

  • Zephex is hosted MCP that gives AI coding editors persistent project context
    hn-ai· 20-may

    Article URL: https://zephex.dev Comments URL: https://news.ycombinator.com/item?id=48202449 Points: 1 # Comments: 0

  • [AINews] How to land a job at a frontier lab (on Pretraining)
    latentspace· 19-may

    a quiet day before google i/o lets us amplify a notable blogpost

  • The Next War Is Already Here. The West Isn't Ready. — Yaroslav Azhnyuk, The Fourth Law & Guest Host Noah Smith, Noahpinion
    latentspace· 18-may

    Ukrainian drone founder Yaroslav Azhnyuk went from pet cameras to AI-guided weapons. He and guest host Noah Smith make the case that the West is asleep at the wheel.

  • [AINews] Google I/O 2026: Gemini 3.5 Flash, Omni (NanoBanana for Video), Spark (background agents), and Antigravity 2.0
    latentspace· 20-may

    Google has been busy!

  • llm-gemini 0.32a0
    simonw· 19-may

    <p><strong>Release:</strong> <a href="https://github.com/simonw/llm-gemini/releases/tag/0.32a0">llm-gemini 0.32a0</a></p> <blockquote> <ul> <li>Compatible with <code>llm&gt;=0.32a0</code> alpha - adds the ability to stream reasoning tokens.</li> </ul> </blockquote> <p>Tags: <a href="https://simonwillison.net/tags/gemini">gemini</a>, <a href="https://simonwillison.net/tags/llm">llm</a></p>

  • Glaucous-winged Gull, Brown Pelican, Snowy Egret, Canada Goose
    simonw· 18-may

    <p><img src="https://static.inaturalist.org/photos/662161673/large.jpg" alt="Glaucous-winged Gull"></p><p><img src="https://static.inaturalist.org/photos/662161721/large.jpg" alt="Glaucous-winged Gull"></p><p><img src="https://static.inaturalist.org/photos/662161937/large.jpg" alt="Brown Pelican"></p><p><img src="https://static.inaturalist.org/photos/662161148/large.jpg" alt="Snowy Egret"></p><p><img src="https://static.inaturalist.org/photos/662161189/large.jpg" alt="Snowy Egret"></p><p><img src="https://static.inaturalist.org/photos/662160828/large.jpg" alt="Canada Goose"></p><p><img src="https://static.inaturalist.org/photos/662160866/large.jpg" alt="Canada Goose"></p><p>Glaucous-winged Gull, Brown Pelican, Snowy Egret, Canada Goose, in Los Angeles River, CA, US</p><p>I'm heading home …

  • datasette-llm 0.1a8
    simonw· 19-may

    <p><strong>Release:</strong> <a href="https://github.com/datasette/datasette-llm/releases/tag/0.1a8">datasette-llm 0.1a8</a></p> <blockquote> <ul> <li>Fix for bug where <code>llm_prompt_context()</code> hook did not fully collect chains of responses. #7</li> </ul> </blockquote>

  • datasette-llm-accountant 0.1a4
    simonw· 19-may

    <p><strong>Release:</strong> <a href="https://github.com/datasette/datasette-llm-accountant/releases/tag/0.1a4">datasette-llm-accountant 0.1a4</a></p> <blockquote> <ul> <li>Fixed bug tracking chains of responses. Refs <a href="https://github.com/datasette/datasette-llm/issues/7">datasette-llm#7</a></li> </ul> </blockquote> <p>Tags: <a href="https://simonwillison.net/tags/llm">llm</a>, <a href="https://simonwillison.net/tags/datasette">datasette</a></p>

  • The last six months in LLMs in five minutes
    simonw· 19-may

    <p>I put together these annotated slides from my five minute lightning talk at PyCon US 2026, using the <a href="https://tools.simonwillison.net/annotated-presentations">latest iteration</a> of my <a href="https://simonwillison.net/2023/Aug/6/annotated-presentations/">annotated presentation tool</a>.</p> <div class="slide" id="5-minutes-llms.001.jpeg"> <img loading="lazy" src="https://static.simonwillison.net/static/2026/5-minutes-llms/5-minutes-llms.001.jpeg" alt="The last six months in LLMs in five minutes Simon Willison - simonwillison.net PyCon US 2026 Lightning Talk " style="max-width: 100%" /> <div><a style="float: right; text-decoration: none; border-bottom: none; padding-left: 1em;" href="https://simonwillison.net/2026/May/19/5-minute-llms/#5-minutes-llms.001.jpeg">#</a> <p>I pres…

  • Gemini 3.5 Flash: more expensive, but Google plan to use it for everything
    simonw· 19-may

    <p>Today at Google I/O, Google <a href="https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-5/">released Gemini 3.5 Flash</a>. This one skipped the <code>-preview</code> modifier and went straight to general availability, and Google appear to be using it for a whole lot of their key products:</p> <blockquote> <p>3.5 Flash is available today to billions of people globally:</p> <ul> <li>For everyone via the Gemini app and AI Mode in <a href="https://blog.google/products-and-platforms/products/search/search-io-2026">Google Search</a> </li> <li>For developers in our agent-first development platform Google Antigravity and Gemini API in Google AI Studio and Android Studio</li> <li>For enterprises in Gemini Enterprise Agent Platform and Gemini Enterprise.</li> </ul> …

  • llm-gemini 0.32
    simonw· 19-may

    <p><strong>Release:</strong> <a href="https://github.com/simonw/llm-gemini/releases/tag/0.32">llm-gemini 0.32</a></p> <blockquote> <ul> <li>New model <code>gemini-3.5-flash</code> for <a href="https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-5/">Gemini 3.5 Flash</a>.</li> </ul> </blockquote> <p>See also my <a href="https://simonwillison.net/2026/May/19/gemini-35-flash/">notes on Gemini 3.5 Flash</a>, and <a href="https://simonwillison.net/2026/May/19/gemini-35-flash/#a-pelican-on-a-bicycle">the pelican</a> I drew using this upgrade to the plugin.</p> <p>Tags: <a href="https://simonwillison.net/tags/gemini">gemini</a>, <a href="https://simonwillison.net/tags/llm">llm</a></p>