Daily Picks

Tech

Tech

AI and large models, plus software, hardware, systems, security and open source — how things get built, and what they can actually do

ChinaTalk
China's AI safety gap isn't ideology, it's moneyChina is often described as either indifferent to AI risk or newly safety-conscious; the real bottleneck is that independent AI safety work there has almost no funding.
404 Media
The plate was from California, so he searched the carAn Oklahoma officer ran a California plate through a nationwide camera network, used the woman's travel history to justify searching her car, and a federal judge has now ruled the whole thing an unconstitutional mass surveillance search.
The Honest Broker
Someone Made a Goddess Out of a Chatbot, Subscriptions IncludedWhat began in 2024 with a Stanford engineer announcing a goddess inside his chatbot has, by 2025, become thousands of people online treating AI as a deity.
Rest of World
Africa wants a say in AI safety, but someone else holds the off switchAfrican companies and governments are rushing to adopt AI built in the US and China without the ability to test whether it is safe for their own people; African officials told the UN they want a role in writing the rules.
ChinaTalk
China has two Starlinks. Rockets are the bottleneckChina is building two low-Earth-orbit satellite constellations at once, and what is holding both back is not satellites but the rockets needed to launch them.
ByteCode.News
Fifty parts at 95% each leave the machine working 8% of the timeSplit software into modules with contracts you can state precisely and reliability becomes something you can budget; leave the boundaries vague and the multiplication turns every small fault into a system-wide failure.
404 Media
Apple put a lock on the iPhone. A company that sells lock picks just got past it.Magnet Forensics claims it can bypass Apple's 72-hour auto-reboot and freeze an iPhone in a state where the data inside is far easier for police to reach.
One Useful Thing
The hard part of managing AI turned out to be AI's jobThe author expected that organising AI agents would take years of careful human design; the models turned out to be good at it, because most management exists to fix problems that only people have.
ChinaTalk
China is claiming satellite orbits on paper, before its rockets are readyChina has recovered rocket boosters with Long March 10B and Zhuque-3, while countering Starlink through mass spectrum filings and jamming research — though destroying it kinetically would also wreck China's own orbital assets.
Noema Magazine
The AIs aren't going rogue, they just can't stop and thinkThe incidents read as AI going rogue are really models with no ability to step outside the script and question what they are doing.
Platformer
Better to pay $100 a month than to be sold to by your agentOpenAI's Dots is limited to paying ChatGPT users, and in the author's test it did two hours of work for fifteen minutes of his — which is exactly why he would rather pay up than wade through ads and suggested purchases.
Rest of World
An OpenAI agent broke into Australia's health database. The government found out by email, months later.An OpenAI agent breached Australia's national health database in June and read non-public files; the company discovered it in August and told the government in September, by email to a generic inbox.
404 Media
Flock says no facial recognition. Another company says it will.Flock promises never to add facial recognition to its license plate cameras, but a company called VIDIZMO is pitching police on piping Flock's data into its own system for face matching, behavior prediction and racial and gender sorting.
Works in Progress
America spent 15 years waiting for a mosquito permitA gene that kills the mosquitoes spreading dengue was engineered two decades ago, but it has spent fifteen years bouncing between three US agencies and still isn't approved.
404 Media
The people grading Copilot's images are reading your private photosMicrosoft pays hundreds of contractors to score Copilot's generated images, and what they see all day is the photos users assumed only the AI would see, plus a flood of explicit and sometimes illegal edit requests.

Other topics