Ai Code
Topic archive • 39 matches
AI-generated: summaries written by AI from the linked sources. How we use AI
2026-11-09
Technology
Claude Code settings files are accidentally exposing secrets on GitHub, with passwords, API keys, and other credentials being shared after users accept a 'Do not ask again' prompt. The exposure affects thousands of secrets according to Ars Technica.
Security • Ars Technica AI
PermalinkByteDance Seed introduced Seedream 5.5, an image model built for professional production workflows that supports stable multi-pass editing over up to 100 iterations without color drift. It generates native RGBA images with transparent backgrounds and, according to ByteDance, outperforms GPT-Image-2.5 Sunburst in realism; a Pro version is planned.
AI Models • The Decoder
PermalinkTencent chief AI scientist Vincent Yao Shunyu warned that AI will soon be able to develop bioweapons and argued for a middle ground between open and closed models. For its upcoming open-weight model Hy4, Tencent is filtering dangerous hacking knowledge directly from the training data.
AI Safety • The Decoder
Permalink
Tips
Productivity
Simon Willison notes that agent-written PR descriptions from Claude Code or Codex are accurate and comprehensive but so long they read like slop, putting reviewers off. He wants human editorial judgement: a short, human-approved introduction that helps a reader understand the change at a high level.
PermalinkProductivity
Willison asked Claude (Opus 5.5 Medium in regular Claude Chat) to port a Three.js 3D project to Vanilla JS in the same thread where it had built the original, and reports it was capable of 3D modeling in Vanilla JS + WebGL without Three.js. The two versions were not identical.
Permalink
2026-11-10
Technology
Anthropic is expanding beyond enterprise and coding tools with its own AI-powered office suite. The move follows the hype cycle around Claude Code earlier this year and puts Anthropic in competition with productivity incumbents.
Product Launches • The Verge AI
PermalinkNvidia's GR00T 2 model is designed to teach robots new tasks without teleoperation by transferring human hand movements from video to robotic arms, trained on about 400,000 hours of footage. Success rates vary widely, from 10 to 80 percent depending on the task.
Research • The Decoder
PermalinkGoogle DeepMind researchers used the agentic AI framework AlphaProof Code to formally verify the Signal protocol's cryptographic guarantees in a four-week proof-of-concept. According to Wired, the researchers were able to formally verify that the widely used encryption protocol works as intended.
Research • Techmeme
PermalinkAnthropic's Fable 5.2 completes 41.3 percent of real freelance jobs at professional quality on the Remote Labor Index, more than double the next-best systems Opus 5.5 and GPT 6 Astra, according to The Decoder. A year ago the top score was 2.5 percent, though areas like product design and animation remain weak spots.
Model Releases • The Decoder
Permalink
Investment
Graphon • Series A
Graphon, which wants to resolve AI agent memory limitations by mapping connections across docs, videos, and images, raised a $35M Series A led by RRE Ventures. AI agents can write code, search databases, and carry out multistep tasks, but they can still lose track of past conversations.
Permalink
Tips
Prompt Tips
Anthropic documentation, quoted by Simon Willison, says Claude Fable 5.2 responds to prompts about who owns the code it works on. In Anthropic's testing on small coding tasks, when a prompt said nothing about ownership or framed the codebase as someone else's, the model tended to add to the existing structure even when that repeated existing code.
PermalinkProductivity
Modern Web Guidance is a set of skills that embed web platform expertise, best practices, and browser compatibility data into coding agents, stewarded by the Chrome, Edge, Firefox and MDN teams. The project says coding agents often default to older patterns because training data contains vast amounts of legacy code.
PermalinkPrompt Tips
Simon Willison tested Gemini 4 Carbon in Google's Antigravity IDE with a three.js animation prompt, then had to prompt "build that into a standalone HTML file" to get a version with no external dependencies.
PermalinkPrompt Tips
Simon Willison prompted Claude Fable 5.2 Medium to make a three.js animation of Yonkers as a mech using OpenStreetMap data; the model correctly deduced that "átomos" was a mobile keyboard typo for "stomps".
Permalink
2026-11-11
Technology
AWS released AWS Emulator, a free tool that simulates more than 200 of its services locally. AWS says AI agents wrote the tool and that its team has not written or reviewed any of the code, which the company is touting as a selling point.
Product Launch • Techmeme
PermalinkBased on a transcript excerpt: Matt Wolfe interviews Peter Yang, a former product manager who quit his job to become a full-time creator, about the personal AI system he uses to run his one-person business. Yang demonstrates how he connects tools like ChatGPT and Claude Code to create an AI 'chief of staff' that reads his email, calendar, and Slack twice a day, plus automated workflows for podcast research, video editing, and content creation. Wolfe also builds simplified versions of these systems so viewers can follow along and set them up.
Practical Tutorial • Matt Wolfe
PermalinkAnthropic says it used its in-house Mythos model to port Claude to AMD GPUs in three days with almost no human intervention, a switch it says would normally take months. Anthropic now runs Claude on four chip families and aims to cut costs and reduce vendor dependencies, claims that, if accurate, would erode Nvidia's CUDA moat.
Infrastructure • The Decoder
PermalinkHelsing unveiled Clarity, an AI platform that uses agents to distill sensor data into target recommendations for weapons systems, under human oversight and European control. The company says every decision will carry a human name, though reporting notes that automation speed and data fusion create risks that are hard to contain.
Defense AI • The Decoder
PermalinkAccording to a ChinaTalk investigation, US data brokers including Mercor, Surge AI, and AfterQuery are selling roughly half a billion dollars' worth of AI training data, including RL environments previously built for OpenAI or Anthropic, to Chinese AI labs. At least nine firms are involved, and export controls barely apply.
Policy • The Decoder
Permalink
Tips
Productivity
Google DeepMind used AI to generate 6.5 million lines of mathematical proofs to check Signal's encryption code in four weeks, and restored two previously fixed security bugs to test whether the system could catch them, according to The Neuron.
Permalink