6:08 PM
Today in brief.
Apple's fall AI bet: prove-it photos and always-on Watch
Apple's fall event leans into AI where the device lives: Reference Image to prove a photo is not slop, an always-listening Watch, an AI-designed foldable hinge, and a fatter A20 Pro Neural Engine.
🔧 Tech
Qwen 3.8 follows GPT-5.5 Pro reasoning prefills
A controlled gist shows Qwen3.8 A95B jumps +18.18 points toward GPT-5.5 Pro answers when seeded with just the first 1% of the teacher's reasoning.
🔬 Science
Growing proof autonomous cars save lives
IEEE Spectrum stacks IIHS and Waymo numbers showing robotaxis and strong ADAS cut crashes hard, with a global upside measured in hundreds of thousands of lives a year.
🔧 Tech
Geiger inventories every AI agent on your machine
Show HN drops Geiger: one read-only npx scan that lists every AI agent, harness, MCP server, and extension on a machine, plus what each can touch.
💼 Business
Anthropic's 2030 economy explorer vs 10k Americans
Anthropic's Economics team ships an interactive 2030 scenario model; plug in your capability bets and compare them to more than 10,000 surveyed Americans.
💼 Business
Suno v6 is the first label-backed AI music model
Suno ships v6, its first music model trained with record-industry help, including licensed material from Warner, BMG, and Believe.
🏛️ Policy
OECD PISA: students who use AI score worse
Verge covers OECD PISA findings that students who use AI generally score worse, unless they are trained to judge AI output. Then scores can rise.
🔬 Science
GPT-5.6 Sol plus Codex runs quantum lab experiments
OpenAI documents an MIT EQuS workflow where GPT-5.6 Sol, wired through Codex, runs and adapts superconducting-qubit measurements that used to need constant human babysitting.
🤖 AI
Perplexity Search lands inside Nous Hermes Agent
Arav puts Perplexity Search into the Hermes Agent harness with a 450B+ URL index, and Perplexity ships Q2D-Web, a public leaderboard for agentic RAG retrieval.
No stories in this category for this edition.