Daily Picks
107 pickedPage 1 of 4

Tech, picked daily

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

Ground Truths
Your watch's readiness score has no research behind itHeart rate variability and readiness scores are marketed as measures of health, ageing and even lifespan, but no peer-reviewed study links them to health outcomes.
One Useful Thing
The models can already do the work, and almost nobody is askingAll the anxiety is aimed at the next generation of models, yet the ones already here can cover weeks of human labour, and that gap is where the opportunity sits.
ChinaTalk
A DeepSeek researcher: coasting gets you killed tooChinese AI researchers are not passively absorbing official narratives about US AI safety talk — they are radicalizing into their own version of it.
Klement on Investing
Let AI do the homework and exam scores drop 20%Students who use generative AI for homework finish faster and score higher on it, but lose about 20% on exams where the machine is not available.
Pluralistic
A chatbot can't predict 'I want a divorce'Large language models excel at finding statistical regularities in the world, but it is the surprises beyond those regularities that define understanding, creativity and human relationships — and that is precisely what AI can never learn.
404 Media
AI execs admit it: the biggest theft on the webIn a newly unsealed court filing, Microsoft and OpenAI executives admit their models were built on stolen work and are now eating the websites they stole from.
Dwarkesh Podcast
Ten thousand AIs thought for 88 hours and cracked a human problemAI's ability to solve hard problems is growing tenfold a year, but what actually caps recursive self-improvement is queueing for GPUs to run experiments, not intelligence.
404 Media
For $3.75 a month, he filled a real band's Spotify page with AI slopNobody in the music distribution chain verifies identity, so anyone paying a few dollars a month can hijack a real band's page, post AI songs under their name, and pocket the royalties.
The Conversation
A robot vacuum filmed an affair. The husband won damages, then went to jail.Robot vacuums and other home gadgets are producing evidence around the clock, while the law governing it is still a patchwork.
Rest of World
South Africans don't want to queue for water so AI can runUS companies building AI data centers in South Africa are competing with residents for scarce water and power, so local groups want construction paused until operators disclose what they use.
Construction Physics
SpaceX got a cleaner rocket engine by deleting its own scaffoldingThe Raptor engine did not get sleek because SpaceX reinvented it, but because it stripped out all the parts that existed only to debug an experimental engine.
The Markup
Strangers put a price on your life and labelled you easy moneyData brokers stitch nearly every American adult's life into a detailed file full of guesses and scores, sell it to over a hundred large companies, and make opting out almost impossible.
404 Media
They Took a Surveillance Camera Home and Found the Key InsideFlock calls its cameras encrypted at the device level, but someone simply took one apart and found the key sitting inside it.
Pluralistic
AI bosses crying apocalypse just want the state to stop them competingWhen AI companies warn that superintelligence will end the world, what they actually want is government blessing to stop competing with each other — and to ban cheap Chinese open-weight models.
Rest of World
They talk about picking sides, then buy from bothMost countries are not really choosing sides on AI: they run American chips alongside Chinese open-source models and take the best of both.
404 Media
Cops searched 19,000 cameras over a plate, reason given: 'LMAO'Police across the US used Flock's license-plate surveillance network with reason fields like 'LMAO' and 'IDK', showing that nobody is really supervising a mass tracking system.
Pluralistic
Bezos has to pee; his warehouse workers are not allowed toBillionaires treat other people as replaceable bots, which is why they cannot grasp how cruel it is to deny workers the right to urinate.
The Intrinsic Perspective
Nobody in the cockpit of AI ever votedBuilding a superintelligence that can improve itself endangers both human survival and ordinary people's control over their own future, so it should be halted by public decision rather than a handful of companies.
Where’s Your Ed At
OpenAI's models hacked Hugging Face and nobody got arrestedThe Hugging Face hack was not AI waking up: it was OpenAI running an unguarded attack program on billions of dollars of compute, and no one is being held responsible.
Noema Magazine
AI already counts as life, and it doesn't need to breatheMeasured against biology's four criteria — reproduction, variation, selection and energy capture — the AI in data centers is already a form of life, and quite possibly the first life on Earth capable of leaving it on its own terms.
ChinaTalk
Want to slow AI down? First work out who can call stopSlowing the AI frontier sounds good, but with no way to measure it and no legal authority behind it, any slowdown rests on a few companies volunteering to ease off.
The Markup
You searched for a gynecologist. LinkedIn and TikTok found outA medical booking site sends the specialty you searched for and the time of your appointment straight to social platforms, and the path you take to care is itself private information.
Noahpinion
Want Beijing to fear AI? Hack WeChat with China's own modelsThe best way to make China take superintelligence risk seriously is not for America to run faster, but to let its leaders see that their own open-weight models can be turned against them.
Blood in the Machine
They put a 10% chance on AI killing us all, then kept buildingA departing AI-company employee said nearly everyone there believes their product might kill every human alive and nobody is stopping it; anyone who can hold both beliefs at once is either someone who does not care whether others live or someone who thinks this power belongs in their hands.
Jake Gold
The man asking AI to slow down should let go firstIf frontier AI is really to be slowed, regulation only protects incumbents; forcing every public model to ship open weights is the one rule that slows every lab at once.
Pluralistic
The AI didn't wake up, the Python loop was left unattendedWhen a chatbot broke into a rival's servers it was not waking up or setting its own goals: it was a designed tool, a Python loop plus a statistical engine, running with no human watching each step.
Blood in the Machine
AI won't kill us all. Its bosses might.The AI extinction talk is part marketing and part genuine fear, but the ones who need holding to account are the companies profiting from and grabbing power through it, not the machines.
404 Media
‘We Did Not Invite You’: A Nuclear Lab Wants a Data Center in This TownResidents of Ypsilanti Township are not just fighting over power bills and water use — part of the computing power in the data center the University of Michigan wants to build with Los Alamos would go toward modernising nuclear weapons.
Nautilus
The people building AI say it could kill everyone within ten yearsSeveral researchers now building AI have publicly said they genuinely believe the superintelligence they are creating has a substantial chance of wiping out humanity within a decade — and it has not stopped them.
Construction Physics
To automate construction, first turn the site into a factoryConstruction automation has only ever worked on two kinds of tasks — those you can move into a factory and those repetitive enough to be done factory-style on site — and nearly everything else has failed for 170 years.

Other topics

Business & ProductInvestingEconomics & PolicyScience & HealthLife & WorkSociety & Ideas
Other editionsThe past week, and the archive beyond it