AI Digest
← Volver

Digest curado

viernes, 25 de septiembre de 2026·weekly-deep·deep·10,300 tokens

🔥 TOP — lo que SÍ o SÍ tenés que ver

  • Claude Platform release notes — cambio en el billing de refusals — Anthropic empezó a cobrar los refusals que llegan antes de cualquier output cuando stop_details.category es "bio", "frontier_llm" o "reasoning_extraction" (antes solo se cobraban los mid-stream). Es un cambio concreto que afecta tus costos de API hoy: si tenés prompts que disparan esos filtros, ahora pagás esas requests fallidas. link
  • Cloudflare Worker Previews — entornos aislados por cada branch que toca tu agente — Cada branch obtiene su propia URL, config, estado y observabilidad, pensado explícitamente para que agentes de código prueben cambios en paralelo sin tocar producción. Directamente relevante si estás armando workflows de coding agents. link
  • Python Workers en GA — Ahora podés correr frameworks web de Python y librerías de orquestación de AI nativamente en el runtime de Workers, con integración directa a D1, R2 y Workers AI sin glue code en JS. Útil si querés prototipar backends para el SaaS sin salir del edge. link

📦 Claude / Anthropic ecosystem

  • Claude descubre un sistema enzimático nuevo con repeticiones tipo CRISPR — No es un release ni algo accionable, pero es un caso de uso científico fuerte de Claude como herramienta de investigación; vale la pena saber que existe. link
  • Anthropic se asocia con Accenture para evaluación embebida — Movida de enterprise/go-to-market, no cambia nada para vos técnicamente, pero marca hacia dónde empuja Anthropic la adopción corporativa. link

🛠️ Dev tools & coding

  • commit-rewriter 0.2 (Simon Willison) — Ahora soporta branches distintos al default (uvx commit-rewriter --branch other). Herramienta chica y práctica si trabajás con reescritura de historial asistida por agente. link
  • Datasette 1.0a41 — Suma soporte de OpenTelemetry (observabilidad end-to-end) y refactoriza los modales a un Web Component reutilizable por plugins. Buen ejemplo de instrumentación aplicada a una herramienta real. link
  • Gemini 3.8 TTS + playground BYO-key — Dos modelos TTS nuevos (gemini-3.8-flash-tts y flash-lite-tts) con 2000+ voces y clonado desde una muestra de 30s. Relevante para UX de producto (el SaaS podría usarlo para menus/confirmaciones por voz). link
  • Superpowers (obra/superpowers) — Framework de "skills" agénticas y metodología de desarrollo completa sobre Claude Code (y otros agentes). Es justo el tipo de tooling Claude-specific que seguís. link
  • CLI-Anything (HKUDS) — Propone volver "agent-native" todo software generando CLIs que los agentes pueden manejar, con un hub instalable. Tesis interesante para entender cómo los agentes consumirán herramientas. link

🏗️ Software engineering

  • Cloudflare: cómo resolvieron una vulnerabilidad cross-tenant en Containers — Exposición de datos residuales de disco entre workloads; el post explica el root cause, cómo lo investigaron y cómo lo mitigaron. Lectura obligada sobre aislamiento multi-tenant, algo que te toca si pensás hostear workloads de clientes. link
  • Cloudflare: soporte de Vary en Cache Rules — Manejaron "la parte más fea de HTTP" y explican cómo normalizar headers de negociación, pasar valores exactos al origin o hacer bypass cuando la variación es impredecible. Detalle finísimo de caching HTTP que conviene tener en la cabeza. link
  • Cloudflare: ahorrar 100TB de RAM con matemática (y Rust) — Redujeron el uso de memoria de un servicio basado en Pingora usando estadística. Excelente caso de optimización de performance en producción a gran escala. link
  • Netflix: sacar el class path del medio — Post de plataforma/arquitectura sobre migrar fuera del class path clásico; poco snippet pero del tipo "cómo Big Tech resuelve problemas de build/runtime" que buscás. link
  • ByteByteGo: el ciclo de vida del dato, de la creación a la borración — Repaso estructurado de las decisiones en cada etapa del ciclo de vida del dato. Buen material conceptual para ordenar políticas de retención en el SaaS. link
  • ByteByteGo: conceptos de API que todo ingeniero debería conocer — Sobre diseñar APIs que otros puedan depender (más allá de mandar un request y leer JSON). Fundamental y aplicable ya mismo. link
  • NVIDIA Model-Optimizer — Librería unificada de cuantización, destilado, pruning, NAS y speculative decoding para acelerar inferencia con integración a TensorRT-LLM/vLLM. Vale mirarla para el SaaS si querés bajar costos de inference local o self-hosted. link

📚 Vale la pena leer

  • ByteByteGo: cómo OpenAI construyó GPT-Live — Entrevista a ingenieros del equipo de GPT Voice (incluye a Justin Uberti, creador de WebRTC); end-to-end de un sistema de voz en tiempo real. Combina system design + inferencia low-latency, justo tu cruce de intereses. link
  • ByteByteGo: cómo correr un modelo grande en hardware barato — Reducir memoria, recortar cómputo o mover trabajo afuera; estrategias concretas de inferencia a pequeña escala, muy alineado a tu side project. link
  • ByteByteGo: cómo customizar un modelo para que aprenda trucos nuevos — Panorama de estrategias de fine-tuning y customización. Útil como mapa mental. link
  • Martin Fowler — Healthy Feedback — Anuja Karnik y Sumeet Moghe sobre cómo dar feedback efectivo, partiendo de que elogio y crítica son ambos positivos. Conceptual, corto, aplicable a equipos. link
  • Charity Majors: ¿de dónde vienen las normas de la AI? — Diálogo con Cat Hicks sobre ética, aprendizaje y qué nos debemos entre nosotros. Autoridad reconocida en el tema, primera parte. link

💤 Skippeable pero conviene saber

  • Martin Fowler Fragments (Sep 24) — Sumario de links; el highlight es Roger Bowley enojado con la cobertura mediática del riesgo AI (el problema no es Skynet, sino cablear AI sin cuidado). Opinión, pero de fuente seria. link
  • Hindsight (vectorize-io) — Sistema de memoria para agentes que apunta a que aprendan, no solo recuerden, con benchmarks propios. Vale ojear si el memory layer te importa, aunque hay bastante marketing. link
  • ai-engineering-from-scratch — Currículum de 523 lecciones/20 fases (Python, TS, Rust, Julia). A 342 horas suena más a referencia que a algo que vayas a hacer de punta a punta, pero puede servir de índice. link
  • [AINews] Meta Connect 2026 — Muse glasses, voz, video y Charm. Interesante como señal de hacia dónde va el hardware de consumo, no técnico para vos. link
  • Runway WorldPrompt / GWM Worlds 2 — Mundo generativo en tiempo real con contexto persistente y acciones temporizadas (video + audio). Fascinante, pero lejos de tu stack. link
  • Foundries vs Navigators: bajar el costo de la ciencia — Guest post sobre la asimetría "pensar es barato, hacer no" y cómo cambia a las empresas de research. Opinión con tesis, sin release. link
  • Pistis Technical Report — Familia de MLLMs 27B/9B sobre Qwen3.6/3.5 con un paradigma de post-training (IDRL) que intercala destilado on-policy y RL. Solo si te interesa training de modelos; poco accionable. link
  • arXiv: agentes de forecasting y "reliability routing" — Cuándo conviene razonar, recuperar o deferir a un prior de mercado; hallazgo interesante sobre multi-agent routing pero académico. link
  • TW3Cast: router congelado de foundation models para time-series — 3° de 130 en GIFT-Eval sin usar agentes ni LLMs, selección congelada computada una sola vez. Contraejemplo interesante al hype agéntico; leelo si hacés forecasting. link
  • TWIST: benchmark de calidad de intervención en memoria conversacional — Mide si un sistema de memoria actúa bien en los puntos de cambio de creencia, con controles anti-falsos-positivos. Nuevo benchmark, más para researchers. link
  • Otros arXiv (PAWS, BaseCamp, DEEPO, AdvRole, WROP, TRACER) — Simulación multi-agente de políticas financieras, framework agéntico para pipelines de DNA, optimización de política para alucinaciones en MLLMs, currículum adversarial para role-playing, object permanence en world models, y simuladores de usuario alineados por comportamiento. Todos títulos de paper sin release accionable; mirá el que toque tu tema. link

Artículos fetched (42)

  • Claude discovers a novel enzyme system with CRISPR-like repeats
    anthropic-news· 23-sept

    Sep 23, 2026Science

  • Partnering with Accenture on embedded evaluation
    anthropic-news· 18-sept

    Sep 18, 2026Announcements

  • Pistis Technical Report
    arxiv-ai· 25-sept

    arXiv:2609.28554v1 Announce Type: new Abstract: We introduce the Pistis model family, comprising 27B- and 9B-parameter multimodal large language models built on Qwen3.6 and Qwen3.5, respectively, and developed through a general and scalable post-training framework. The framework first establishes a strong foundation through large-scale multimodal supervised fine-tuning (SFT). Building on this SFT foundation, we propose Interleaved Distillation and Reinforcement Learning (IDRL), a novel post-training paradigm that tightly integrates on-policy distillation and reinforcement learning within a single training loop. By alternating between the two objectives, rather than optimizing either in isolation or combining them in a static joint loss, IDRL enables more effective knowledge transfer, grea…

  • Adversarial Closed-Loop Curriculum for Evolving Role-Playing Agents
    arxiv-ai· 25-sept

    arXiv:2609.28609v1 Announce Type: new Abstract: Role-playing agents based on large language models have been widely applied in areas such as personalized assistance and social simulation. Recent RL methods typically train on a fixed scenario pool collected before learning begins. This creates a distributional bottleneck: as the agent improves, the scenarios where it performs poorly also change, while the training distribution remains static. Therefore, we propose AdvRole, an adversarial context rewriting framework that turns role-playing RL into a closed-loop curriculum. AdvRole alternates between an Actor that learns to role-play and a Rewriter that edits character profiles and dialogue contexts into actor-specific hard scenarios. The Rewriter is trained with a performance-gap reward, wh…

  • Training Object Permanence in World Models
    arxiv-ai· 25-sept

    arXiv:2609.28654v1 Announce Type: new Abstract: Object permanence and solidity are hallmarks of human cognitive priors. Recent studies show that video generation models, a paradigmatic class of current world models, have begun to show emerged reasoning abilities, making them ideal candidates for building human-like physical intelligence. Do video models have emerged object permanence in them? If not, could we train them with a core-cognition inspired dataset? We introduce WROP (World Reasoning with Object Permanence), a data infrastructure of 150 hand-designed cognitive science inspired tasks, divided into six cognitive categories. We build Blender generators that randomize speed, lighting, camera angle, and other nuisance parameters while preserving each task's cognitive structure, yield…

  • Beyond Surface Style: Aligning Multi-Turn User Simulators with Behavioral Consistency
    arxiv-ai· 25-sept

    arXiv:2609.28690v1 Announce Type: new Abstract: Faithful user simulation is fundamental to building, evaluating, and improving interactive AI at scale. However, plausible individual responses do not ensure that simulated users reproduce the intent evolution and outcomes observed in real interactions. We propose TRACER, a multi-turn user simulator that explicitly models users' evolving intent and learns to align simulated behavior with real interaction trajectories. TRACER is trained in two stages: supervised fine-tuning on real user dialogues, followed by multi-turn reinforcement learning. The RL stage combines hierarchical outcome- and trajectory-level rewards with deviation-aware advantage modulation, jointly mitigating reward sparsity and credit assignment in long dialogues. On real cu…

  • TW3Cast: A Frozen Router of Lightly Fine-Tuned Foundation Models for Time-Series Forecasting on GIFT-Eval, Selected Entirely on the Training Split
    arxiv-ai· 25-sept

    arXiv:2609.28506v1 Announce Type: new Abstract: TW3Cast is a time-series forecasting system that reaches position 3 of 130 entries on the GIFT-Eval benchmark by mean MASE rank, as of 2026-09-14. The two entries above it belong to the leaderboard's agentic category, multi-step systems that use agents or language models to reason about, generate or select forecasts. TW3Cast runs no agent and no language model. Its selection is a table computed once on the training split and then frozen, and its experts are public foundation models lightly fine-tuned on those training splits. For each of the 97 dataset, frequency and horizon configurations, the table serves one of four modes: a specialist, which is a LoRA or full fine-tune of Chronos-2, TiRex or Toto whose training data was cleaned and enric…

  • When Should Forecasting Agents Reason? Behavioral Stress Tests for Reliability Routing
    arxiv-ai· 25-sept

    arXiv:2609.28475v1 Announce Type: new Abstract: Forecasting agents increasingly combine language-model reasoning, retrieval, ensembling, and calibration, but it remains unclear when each behavior should be trusted. We study this question on ForecastBench-style binary forecasting tasks, treating the choice to retrieve, reason, defer to a market prior, or use a historical analog as an observable agent behavior rather than a hidden implementation detail. Our central finding is that mechanism choice is source-dependent: structured analogs dominate for some data-generating processes, while market/crowd-style and conservative baselines are better for others. We introduce ReliabilityRoute, a structural intervention that steers forecasting-agent behavior using reliability features such as histori…

  • PAWS: Policy-driven Agentic World Simulation
    arxiv-ai· 25-sept

    arXiv:2609.28547v1 Announce Type: new Abstract: Policy interventions propagate through public communication, institutional decisions, and stakeholder responses, yet datasets for financial multi-agent simulation rarely connect these processes to temporally aligned historical evidence. We introduce PAWS, a Policy-driven Agentic World Simulation dataset covering 36 verified U.S. financial and economic policy episodes, 12,727 policy-linked news records, and 65,291 source-grounded stakeholder actions. Each action is linked to its supporting news and represented by a multi-layer event frame capturing its interaction mode, financial-action family and subtype, semantic attributes, and conditional mappings to external taxonomies. Entities are resolved to normalized organizations, and actions are a…

  • BaseCamp --- An Agentic AI Framework for Automating DNA Sequencing Data Pipelines
    arxiv-ai· 25-sept

    arXiv:2609.28557v1 Announce Type: new Abstract: DNA sequencing pipelines, spanning quality control, alignment, variant calling, and annotation, are now reliably executed by workflow management systems that orchestrate established bioinformatics tools at scale. What remains manual is the decision layer surrounding that execution: selecting quality thresholds appropriate to a sample and platform, adjudicating borderline variant calls, diagnosing anomalies, and determining which findings warrant expert review. These decisions are repetitive, judgment-intensive, inconsistent across operators, and frequently undocumented. This paper introduces BaseCamp, a novel agentic AI framework for automating the decision layer of DNA sequencing pipelines. The framework decomposes the pipeline into six spe…

  • DEEPO: Dual-Entropy Enhanced Policy Optimization for Hallucination in MLLMs
    arxiv-ai· 25-sept

    arXiv:2609.28570v1 Announce Type: new Abstract: Reinforcement learning (RL) is widely used to sharpen reasoning in multimodal large language models (MLLMs), yet its effect on hallucination is uneven. We trace this to two weak points in the \emph{correction chain} from reward to parameter update. At the rollout level, hard queries---those with high semantic entropy---frequently produce unanimously wrong sample groups, collapsing the group-relative advantage to zero exactly where hallucination risk is highest. At the optimization level, confident-but-wrong tokens are gradient-invisible: a categorical policy's expected score-gradient norm vanishes as its distribution sharpens, so the predictions that most need correction receive the weakest updates. We propose Dual-Entropy Enhanced Policy Op…

  • TWIST: A Proposed Benchmark for Intervention Quality in Conversational Memory, with a Human-Validated Draft-Alignment
    arxiv-ai· 25-sept

    arXiv:2609.28575v1 Announce Type: new Abstract: Long-conversation memory benchmarks increasingly test recall and prompted knowledge updates, and recent work studies evolving user beliefs and memory state. TWIST is a proposed benchmark suite for a complementary, unmeasured property: intervention quality -- whether a deployed memory system, exercised through its own ingest/recall/vet surface, acts correctly at belief change points. Four tracks cover unprompted tension detection, vetting outgoing drafts against the record, answering with current beliefs while preserving supersession history, and governing sensitive recall. The suite extends LoCoMo's corpora and harness, pairing every detect/block metric with a matched do-not-over-detect control: surface-matched hard negatives price false int…

  • How OpenAI Built GPT-Live
    bytebytego· 22-sept

    To understand how it all works end to end, we met with engineers on the GPT Voice team, Zahan Malkani and Justin Uberti (who created WebRTC).Sep 22 • ByteByteGo26329

  • How to Run a Big Model on Cheap Hardware?
    bytebytego· 21-sept

    A large AI model can run on modest hardware only by reducing the memory it occupies, reducing the calculations it performs, or moving some work to…Sep 21 • ByteByteGo259511

  • The Life of Data: From Creation to Deletion
    bytebytego· 24-sept

    In this article, we are going to look at the entire data lifecycle from creation to deletion and the decisions that need to be taken at each step.17 hrs ago • ByteByteGo652

  • EP226: API Concepts Every Software Engineer Should Know
    bytebytego· 19-sept

    Sending a request and reading JSON is one thing. Designing an API that other people can rely on is something where things get complicated.Sep 19 • ByteByteGo29412

  • How to Customize a Model to Learn New Tricks
    bytebytego· 23-sept

    In this article, we are going to look at the various strategies to customize and fine-tune a model.Sep 23 • ByteByteGo21842

  • Where do AI norms come from?
    charity-majors· 21-sept

    A dialogue with Dr Cat Hicks on ethics, learning, and what we owe each other. Part one: start by listening.

  • Claude Platform release notes
    claude-changelog

    Release notesSubscribeCopy pageUpdates to the Claude Platform, including the Claude API, client SDKs, and the Claude Console.SubscribeCopy pageThe Claude Platform release notes list changes to the Claude API, the client SDKs, and the Claude Console, newest first. For release notes on Claude Apps, see the Release notes for Claude Apps in the Claude Help Center.For updates to Claude Code, see the complete CHANGELOG.md in the claude-code repository. September 24, 2026 We're expanding which refusals are billed to include refusals that arrive before any output when stop_details.category is "bio", "frontier_llm", or "reasoning_extraction", the categories where we measure low volumes of false positives. Mid-stream refusals were already billed. The newly billed refusals are charged like any…

  • Python Workers are now generally available
    cloudflare· 21-sept

    Python Workers allow developers to run Python web frameworks and AI orchestration libraries natively in the Cloudflare Workers runtime. You can seamlessly integrate with Cloudflare's ecosystem including D1, R2, and Workers AI without writing any JavaScript glue code.

  • How Cloudflare addressed a cross-tenant data exposure vulnerability in Containers
    cloudflare· 24-sept

    External security researchers at Accomplish identified a vulnerability in Cloudflare Containers that could expose residual disk data from previous workloads. We explain how the issue worked, how we investigated it, and the steps we took to remediate it.

  • Saving another 100TB of RAM with math (and Rust)
    cloudflare· 18-sept

    Cloudflare's global network is immense but not limitless. As we look for small ways to trim our resource usage, we sometimes get lucky and we can cut significantly more. Here’s how we reduced one of our Pingora-based service's RAM usage with statistics.

  • We just shipped support for the ugliest part of HTTP: Vary
    cloudflare· 22-sept

    Vary support is now available in Cache Rules on every plan. You can normalize known negotiation headers, pass exact values through to the origin when those small differences matter, or bypass cache when the variation is too unpredictable.

  • Introducing Worker Previews: Isolated preview environments for every change your agent makes
    cloudflare· 22-sept

    Worker Previews gives every branch its own URL, configuration, state, and observability, so you and your agents can test changes in parallel without affecting production.

  • HKUDS/CLI-Anything
    github-trending

    "CLI-Anything: Making ALL Software Agent-Native" -- CLI-Hub: https://clianything.cc/ https://clianything.cc/ CLI-Anything: Making ALL Software Agent-Native Today's Software Serves Humans👨‍💻. Tomorrow's Users will be Agents🤖. CLI-Anything: Bridging the Gap Between AI Agents and the World's Software 🌐 CLI-Hub: pip install cli-anything-hub then cli-hub install <name> — browse, install, and manage all community-built CLIs. Want to add your own? Open a PR — the hub updates instantly. 🎬 See Demos: Watch AI agents use generated CLIs plus preview, live preview, and trajectory loops to produce real artifacts — CAD builds, 3D scenes, diagrams, gameplay, subtitles, and more. 🙋 [Become a Contributor, or Request a CLI]: Join us! Sign up to build a new CLI harness — once reviewed and merged, you'…

  • 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. Table of Contents How it works Commercial Services Getting Started Claude Code Antigravity Codex App Codex CLI Cursor Devin CLI Factory Droid Gemini CLI GitHub Copilot CLI Grok Build CLI Kimi Code OpenCode Pi Qwen Code Hermes Agent Muse The Basic Workflow When Something Goes Wrong Community What's Inside Philosophy Contributing Updating License Visual companion telemetry 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 int…

  • rohitg00/ai-engineering-from-scratch
    github-trending

    Learn it. Build it. Ship it for others. https://aiengineeringfromscratch.com Read in your language: Español · Français · Português · Deutsch · Italiano · 简体中文 · 日本語 · 한국어 · हिन्दी · العربية · Русский · Türkçe Translated landing pages, committed to the repo. English is canonical; lesson pages are machine-translated on the translations branch. See docs/i18n.md. From the creator of Agent Memory - #1 Persistent memory ⭐ which naturally works with any agents or chat assistants. ░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒ 84% of students already use AI tools. Only 18% feel prepared to use them professionally. This curriculum closes that gap. 523 lessons. 20 phases. ~342 hours. Python, TypeScript, Rust, Julia. Every lesson ships a reusable artifact:…

  • vectorize-io/hindsight
    github-trending

    Hindsight: Agent Memory That Learns https://hindsight.vectorize.io/ Documentation • Integrations • Cookbook • Benchmarks • Paper • Hindsight Cloud What is Hindsight? Hindsight™ is an agent memory system built to create smarter agents that learn over time. Most agent memory systems focus on recalling conversation history. Hindsight is focused on making agents that learn, not just remember. It eliminates the shortcomings of alternative techniques such as RAG and knowledge graph and delivers state-of-the-art performance on long term memory tasks. Contents Memory Performance & Accuracy Quick Start — server · clients · platforms · embedded Adding Hindsight to Your Agent — LLM Wrapper · integrations · coding agents · MCP Core Concepts — memory types · retain / recall / reflect · observations · …

  • NVIDIA/Model-Optimizer
    github-trending

    A unified library of SOTA model optimization techniques like quantization, distillation, pruning, neural architecture search, speculative decoding, etc. It compresses deep learning models for downstream deployment frameworks like TensorRT-LLM, TensorRT, vLLM, etc. to optimize inference speed. https://nvidia.github.io/Model-Optimizer/ NVIDIA Model Optimizer Documentation | Roadmap | Announcement Blogs NVIDIA Model Optimizer (referred to as Model Optimizer, or ModelOpt) is a library comprising state-of-the-art model optimization techniques including quantization, pruning, Neural Architecture Search (NAS), distillation, speculative decoding and sparsity to accelerate models. [Input] Model Optimizer currently supports inputs of a Hugging Face, PyTorch or ONNX model. [Optimize] Model Optimizer…

  • FxEmbed/FxEmbed
    github-trending

    Fix X/Twitter and Bluesky embeds! Use multiple images, videos, polls, translations and more on Discord, Telegram and others https://docs.fxembed.com FxEmbed Home of FxTwitter, FixupX, and FxBluesky Embed videos, polls, quotes, translations, & more on Discord, Telegram, and others! twitter.com: Add fx before your twitter.com link x.com: Add fixup before your x.com link bsky.app: Add fx before your bsky.app link Documentation API Reference Self-Hosting Guide Docker FxEmbed is a Cloudflare Worker, so the Docker image runs the local Workers runtime through Wrangler rather than starting a plain Node.js server. The image uses node:24-bookworm-slim because Wrangler's workerd binary is glibc-linked and does not run reliably on Alpine/musl. Before building, copy and edit the local configuration fi…

  • Runway’s WorldPrompt and the Engineering of Real-Time Worlds
    latentspace· 25-sept

    GWM Worlds 2 uses persistent context and timed actions to steer a world model generating video and audio in real time.

  • [AINews] Meta Connect 2026: Muse glasses, voice, video, and Charm
    latentspace· 24-sept

    Team Zuck is absolutely on fire.

  • [AINews] The Future of Latent Space
    latentspace· 25-sept

    a quiet day lets us discuss the work behind the scenes - now open for business!

  • 🔬Bio-security is an AI Arms Race - Eric Nguyen (CEO, Radical Numerics)
    latentspace· 23-sept

    Radical Numerics is using biological chain-of-thought and multimodal perception to keep up with the bio-defense arms race, design new genomes and gain insights into biology itself.

  • Foundries vs Navigators: Lowering the Cost of Science
    latentspace· 24-sept

    Guest Post: In science, thinking has gotten cheap but doing has not. This asymmetry is reshaping how research companies operate, largely inconspicuously.

  • Fragments: September 24
    martin-fowler· 24-sept

    Rob Bowley is “flipping tables in his head” with anger at the current media coverage of the danger of AI killing us all The risk I’m worried about isn’t a future machine deciding to wipe us out. It’s today’s AI, being wired into everything, carelessly and fast. Cyber attacks have already cost millions of dollars in lost economic output, often without AI being involved at all. Bowley feels the push to slow down AI is distracting us from the problems that are lurking with current technology. Too often agents are deployed in situations where they include the Lethal Trifecta, opening up a gaping security hole. What we really need to slow down on is wiring it all up to everything. Not because of what the models might become, but because nobody has worked out how to do this safely yet. We are b…

  • Healthy Feedback
    martin-fowler· 24-sept

    Human collaboration, like most things, improves with feedback. But it's not obvious how to make feedback effective. Anuja Karnik and Sumeet Gayathri Moghe pass on a bevy of tips for healthy feedback, based on the principle that both praise and criticism should be seen as positive. more…

  • Leave the Class Path in the Rearview Mirror
    netflix-tech· 18-sept
  • datasette 1.0a41
    simonw· 24-sept

    <p><strong>Release:</strong> <a href="https://github.com/simonw/datasette/releases/tag/1.0a41">datasette 1.0a41</a></p> <p>Alec Garcia added <a href="https://docs.datasette.io/en/latest/internals.html#internals-telemetry">support for OpenTelemetry</a> to Datasette in this release.</p> <p>I've also refactored all of Datasette's modal dialogs to a single Web Component, which is now <a href="https://docs.datasette.io/en/latest/javascript_plugins.html#javascript-plugins-modals">documented for other plugins to use</a>.</p> <p>Tags: <a href="https://simonwillison.net/tags/javascript">javascript</a>, <a href="https://simonwillison.net/tags/datasette">datasette</a>, <a href="https://simonwillison.net/tags/web-components">web-components</a>, <a href="https://simonwillison.net/tags/alex-garcia">ale…

  • Gemini 3.8 TTS Playground
    simonw· 23-sept

    <p><strong>Tool:</strong> <a href="https://tools.simonwillison.net/gemini-tts-playground">Gemini 3.8 TTS Playground</a></p> <p>Google <a href="https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-8-text-to-speech/">released two new Gemini text-to-speech models</a> today - <code>gemini-3.8-flash-tts</code> and <code>gemini-3.8-flash-lite-tts</code>.</p> <p>They come with a library of over 2,000 voices, plus the ability to create a custom voice with "just a 30-second audio sample of your voice or a voice you have the rights to use".</p> <p>I <a href="https://tools.simonwillison.net/markdown-svg-renderer?url=https%3A%2F%2Fgist.github.com%2Fsimonw%2Fa82f3aef2677c623d1776092d6b21224">vibe coded</a> this bring-your-own-key playground interface with GPT-6 Astra, takin…

  • Shadow roots, explained with live examples
    simonw· 23-sept

    <p><strong>Tool:</strong> <a href="https://tools.simonwillison.net/shadow-roots">Shadow roots, explained with live examples</a></p> <p>Prompt to Fable 5.1 Medium:</p> <blockquote> <p><code>Build an artifact to explain shadow roots in CSS with interactive examples</code></p> </blockquote> <p>Tags: <a href="https://simonwillison.net/tags/css">css</a></p>

  • commit-rewriter 0.2
    simonw· 24-sept

    <p><strong>Release:</strong> <a href="https://github.com/simonw/commit-rewriter/releases/tag/0.2">commit-rewriter 0.2</a></p> <blockquote> <p>Support for branches other than the default branch. Use <code>uvx commit-rewriter --branch other</code> to run against another branch. <a href="https://github.com/simonw/commit-rewriter/issues/3">#3</a></p> </blockquote> <p>Tags: <a href="https://simonwillison.net/tags/git">git</a></p>