Closing his talk at WeAreDevelopers World Congress North America, Simon Willison used three kakapo photos and a single prompt to have Claude Opus 5.5 generate Kākāpō Party, an HTML5 canvas pixel-art animation featuring at least 20 kakapos hopping to music with confetti and balloon effects, which he then recorded to video with Claude Code.
The Latent Space podcast spoke with OpenRouter co-founder and CEO Alex Atallah and AMP's Anjney Midha, tracing how OpenRouter grew from the early wave of open-weight models like Llama and Alpaca into a neutral routing layer serving over 10 million developers and handling over 100 trillion tokens per day, and why it was acquired by Stripe.
The US Court of Appeals for the DC Circuit ruled 2-1 to approve the Department of Defense's blacklisting of Anthropic technology, finding the government may restrict the company even absent malice for refusing to open some Claude features to the military. The opinion weighed the risk that restricted models could disrupt military operations against the risk that unrestricted models could hallucinate wrong strike targets, held the defense secretary did not overstep, and rejected Anthropic's request for a rehearing. The court had previously denied its emergency stay request in April.
Why it matters: The report gives the core reasoning of the ruling and the 2-1 vote, helping readers understand the boundaries of military supply-chain controls on AI companies.
According to The Information, Tesla's Optimus robots still need programming for specific tasks in controlled environments, and their hand tactile sensors are unreliable; the factory has developed replaceable glove-like sensor layers for this purpose.
A new working paper by CESifo economists Robert Fairlie and Jane Wu in Munich finds no evidence, in absolute or relative terms, of significant or widespread job displacement or hiring cutbacks among recent college graduates.
Proaction uses Codex, GPT-Live-1, and GPT-6 Astra to build and operate a modern fleet management business, dramatically boosting efficiency. With these tools, Proaction's sales grew 60% and it saved 75+ hours.
An official GitHub tutorial shows how to build custom workflows with canvases in the GitHub Copilot app: run the /create-canvas skill in an agent session, describe the workflow, user actions, and agent actions in natural language, and the agent generates boards, checklists, dashboards, and other interfaces that open in the side panel.
Why it matters: The tutorial covers the three elements of /create-canvas and its two-way collaboration mechanism, so readers can turn their own workflows into reusable canvases.
Simon Willison quoted John Gruber's article 'Muse Looks Cute, but Looks are Deceiving': Meta's Muse gives every user a full persistent Linux VM running on Meta's cloud, packaged as a cute mascot and easy to install and use, and is being called the first consumer-accessible agentic AI system.
AWS proposed an architecture based on Amazon EKS, EFA and DeepEP to accelerate large-scale reinforcement learning post-training of MoE models, delivering 40% higher throughput. The solution targets workloads like RLHF, PPO and GRPO, addressing resource balancing between rollout generation and policy training, high-throughput communication across hundreds of accelerators, and the dynamic all-to-all token routing overhead introduced by Expert Parallelism.
AWS showed how to use the open-source RL framework SkyRL on SageMaker HyperPod to train the Qwen3-VL-8B vision-language model on a visual maze navigation task. Starting from a VisGym SFT checkpoint, GRPO post-training raised the solve rate on a fixed set of 64 mazes from 43.75% to over 95%.
The AWS team detailed five techniques NarrateAI uses to achieve production-grade LLM quality assurance on Amazon Bedrock: adaptive pipeline routing, cross-account multi-model failover, real-time streaming evaluation, a composite evaluation framework, and data accuracy validation.
An AWS Solutions Architect demonstrates deploying Qwen3-TTS-12Hz-1.7B-Base to a real-time inference endpoint via Amazon SageMaker JumpStart, cloning a speaker's voice from a few seconds of reference audio without retraining the model.
Datacor integrated Amazon Quick Sight into its TrackAbout solution, giving industrial gas and welding distributors self-service rental analytics without submitting IT tickets or waiting days or even weeks for custom reports.
A US Department of Defense budget request proposes spending $30.3 million over five years on an AI polygraph project called Polygraph+ or Polygraph Next, led by the Defense Counterintelligence and Security Agency (DCSA). It focuses on AI and machine learning-based scoring algorithms plus contactless "remote sensing" technology, for employee vetting and insider threat detection.
Lowe’s uses OpenAI models to power Mylow and Mylow Companion, which have answered over 25 million questions. The tools help customers and employees move home improvement projects forward.
Latent Space's swyx announced a revamp of AINews, which has run for 3 years and has over 200,000 subscribers, planning to merge AINews v3 with the tens-of-thousands-member Latent Space Discord, migrate the site to Beehiiv and launch a new homepage.
Runway has released the research preview of GWM Worlds 2, which uses the WorldPrompt format to control characters, cameras, and environments in simulated worlds via prompts, generating 720p, 24 fps video and 48,000 Hz audio in real time.
Wayfair is using OpenAI models to improve e-commerce customer support and product catalog accuracy, automating ticket triage and enriching millions of product attributes at scale.
Simon Willison argues that the more he works with coding agents, the more convinced he is that they make software engineering harder. He says these agents can produce stunning results, but unlocking their full potential requires extreme discipline and knowledge.
Simon Willison released commit-rewriter 0.2 on September 24, 2026. The original post is a short beat, related to git, with no further feature details provided.
GitHub Copilot authors argue that chat is not the best interface for collaborating with AI, introducing the canvas feature in the GitHub Copilot app—a small full-stack app that communicates bidirectionally with agents, can call third-party APIs, and runs code locally.
Google Research unveiled its AI video co-director research, building a multi-agent orchestration layer based on Gemini and Veo with four frameworks — Co-Director, CANVAS, A²RD and VQQA — that reformulates long-form video generation as a global optimization and world-state tracking problem.
Antonio Morales of GitHub Security Lab has released Fuzzing Taskflow, an autonomous fuzzing pipeline for C/C++ projects built on the Taskflow Agent — you only specify a GitHub repository, and it automatically identifies entry points, analyzes the build system, writes harnesses, runs AFL++, and triages crashes.
Why it matters: The author explains the design of the automated fuzzing pipeline, showing how LLM agents can take over manual steps like coverage analysis and crash triage.
Google DeepMind has launched Gemini 3.8 Live with Live Avatar, combining near-real-time video generation with speech to give Gemini's conversational AI a dynamic visual presence, with accurate lip sync, natural expressions and smooth turn-taking.
Endura Therapeutics CEO Adrian Sanborn proposes two ways biotech can adapt to AI: Foundries, which use new technology to speed up experimental data output by an order of magnitude, and Navigators, which use AI to accelerate the knowledge work around experiments.
Liquid AI has released experimental DSpark draft models for the vision-language model LFM2.5-VL-3B, using speculative decoding to trade for faster inference without changing output quality — decoding speeds up to 3.13x on-device and 2.66x on H100, with end-to-end gains up to 2.62x and 2.27x respectively.
Why it matters: The article details a speculative decoding acceleration scheme for vision-language models with concrete speedups and memory overhead, plus integration guidance for llama.cpp, MLX-VLM, and SGLang.
NVIDIA, together with Google DeepMind, EMBL-EBI and other institutions, has released predicted 3D structures of protein complexes for more than 2,800 viruses via the AlphaFold Database, with about 30% of the protein interactions previously undocumented by the scientific community.
Remedy Entertainment’s CONTROL Resonant launched day one on GeForce NOW, letting players wield Dylan Faden’s paranormal abilities in the cloud against the Hiss and other threats. Ultimate members get RTX 5080-class performance, DLSS 4, ray tracing, and streaming up to 5K HDR, with no 100GB local install needed.
TrendForce has observed that some cloud service providers are adding QLC SSDs to SSD PODs, replacing the default TLC SSD plus HDD configuration. In AI inference, the Prefill and Decode stages repeatedly read and write KV Cache, and as context windows and batch sizes grow, KV Cache capacity needs multiply, prompting vendors to move caching from costly, capacity-limited HBM down to cheaper tiers.
AINews rounds up AI news from September 22-23, 2026, headlined by Meta Connect 2026, where the Muse personal agent gained voice, live video, and Mac computer use.
Perplexity launched a portable agentic PC powered by AMD Ryzen AI Max series processors. The page's body consists mainly of terms and conditions for downloading AMD brand assets, with no further technical details about the product.
The GitHub Copilot app team rebuilt the pull request view so it can smoothly open an open-source PR with 2,200 files, over one million changed lines, and more than 400 inline comments.
Why it matters: The author offers a first-hand breakdown of the dual-geometry architecture and automated measurement methods for rendering huge PRs, useful for anyone building interfaces with large-scale dynamic content.
Google today released two Gemini text-to-speech models, gemini-3.8-flash-tts and gemini-3.8-flash-lite-tts, offering over 2,000 voices and support for creating custom voices from a 30-second audio sample.
Interactive examples demonstrate shadow DOM's style encapsulation, inheritance, slots, parts, and JavaScript access, showing how shadow roots create isolated DOM trees with private stylesheets that interact with the page DOM in controlled ways. The explainer page was generated by Fable 5.1 Medium from a prompt.
A Microsoft Research study shows that running physical AI inference solely on robots' onboard GPUs limits performance, battery life and scalability, and offloading inference to edge or cloud GPUs can significantly improve it.
Google's Private AI Compute team announced adding private server-side persistent memory to the platform, giving AI assistants long-term memory across devices while maintaining privacy standards at the level of on-device processing.
OpenAI Academy marks its second anniversary. OpenAI says it will continue bringing AI skills to more communities.
On September 23, Google DeepMind released two text-to-speech models, Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS, which support customizing character voices via natural-language prompts in over 100 languages, cloning voices from 30-second audio samples, directing delivery and dialect changes line by line, and natively generating two-speaker scenarios.
Why it matters: The official announcement details both TTS models' capabilities, benchmark results, and availability, so readers can judge how voice creation workflows will change.
NVIDIA validation engineer Sakeena Fiza tests data center hardware before mass production; her team witnessed the first system-level power-on of the Rubin GPU. Validation spans the entire chain from tray to rack to AI factory—a single board contains tens of thousands of components and a rack can approach 500,000—with the goal of catching issues before customers do.
Radical Numerics co-founder and CEO Eric Nguyen argues that the same genomic language models (GLMs) that boost biological capabilities could also let defenses keep up with attacks, but defense is currently losing this bio-security arms race.