Daily Picks
107 pickedPage 2 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.

Cal Newport
Summoning a god is a fine excuse for skipping safetyNobody knows how to build superintelligence, but labs that treat it as a faith will cut corners in the real world because they believe they must win the race.
Klement on Investing
You can spot an AI story by how hard it explains itselfAI fiction is easy to spot because it spells everything out — the theme, the emotion, the moral — while human storytellers instinctively reach for complexity and ambiguity.
Rest of World
Silicon Valley is raiding Asia for people who can talk to governmentsOpenAI and Anthropic are not poaching researchers in Asia but executives who understand both how a US head office allocates money and how local governments and customers actually work.
Platformer
The researcher who quit Anthropic says they really do think AI could kill everyoneThe AI safety warnings never changed; what changed is that the people issuing them now sit at the heart of a trillion-dollar company, and the public and Congress have started taking them seriously.
Rest of World
Google Earth's AI image feature lasted one dayPutting image generation inside the tool people use to verify facts destroyed trust faster than any fake image could.
Cal Newport
An Anthropic engineer says they're building a weapon that could end us — and won't stopAnthropic and OpenAI know they are racing to release a narrow class of unsupervised, heavily armed, long-running AI agents, and because of Silicon Valley's salvationist faith they refuse to stop — which should make the public angry enough to act.
Craig Mod
AI voice is contaminating everything we writeWith AI-generated prose leaking into every text box, the only defence left for a writer is to write stranger and more like themselves.
Dwarkesh Podcast
Most of AI's pretraining gains came from data, not modelsFrom 2019 to 2025, data improvements contributed 12x more compute efficiency gains to pretraining than model improvements (3.7x), but the real value of model research was enabling ever-larger scale of training.
Where’s Your Ed At
The AI bubble's core: a few big clients and trillion-dollar compute promisesOpenAI and Anthropic rely on venture-funded AI startups for 80% of enterprise revenue, and their trillion-dollar compute commitments resemble subprime ARMs, creating huge concentration risk.
The Conversation
Pennsylvania sues TikTok for addiction — but that suit is messier than it soundsSuing over addictive design risks carving a nuanced problem into a black-and-white verdict.
404 Media
YouTube Channel Gave Hunter Biden Its Subscribers' EmailsChannel 5 handed over its subscribers' email addresses to Hunter Biden to promote his memecoin, a move data privacy experts say may be illegal.
ChinaTalk
Taiwan's drone ambitions hit a wall in its own legislatureTaiwan's plan to build drones for security and economic growth is being undermined by political infighting and supply chain realities.
404 Media
Border Patrol tracks your banking to find an excuse to pull you overA secretive CBP predictive policing unit analyzes Americans' financial activity to identify targets, then has local police stop them under other pretexts, which legal experts say is unconstitutional.
Works in Progress
Inside 64K of Memory, the Seed of a UniverseA cheap computer dismissed as a 'game machine' spawned the first virtual world through its low-cost online service, shaping today's social networks and games.
Jake Gold
I trust my coding agents with production secrets nowThe author argues that giving AI coding assistants production-level access is a reasonable risk-reward tradeoff as models improve, but isolation and least-privilege principles still matter.
ChinaTalk
Open Models Are the United Front of AIOpen-source models are reshaping the US-China AI rivalry by uniting all non-closed forces, yet a quick win is impossible; it will be a protracted struggle.
Nautilus
Your Boss Is Watching YouAI-powered employee surveillance is booming, yet research shows it rarely improves performance, often backfires by increasing stress, and mainly enriches software vendors.
Noahpinion
America is still beating China in the AI raceThe US still leads China in AI, but Trump's immigration policies could squander that edge, and the stakes of this race are far graver than economic competition.
Benedict Evans
AI, tools and transformationThe real challenge is not building automation tools—it's discovering that you need one—and AI merely moves the threshold, not the problem.
404 Media
Flock Taught Cops How to Surveil No Kings ProtestersFlock publicly markets its cameras as harmless static image tools for solving crimes, but in training webinars it teaches police to use its system for round-the-clock surveillance of protests and daily life — its real pitch is far more invasive than its public image suggests.
Rest of World
Chinese businesses are giving away AI tokens with coffee, credit cards, and dumplingsChinese businesses are turning AI tokens into a new kind of giveaway and marketing gimmick, giving ordinary people cheap or free compute — a sign of a coming differentiation race centered on compute consumption.
Jake Gold
Linus Torvalds was the first 1000x programmerLinus Torvalds became the first 1000x programmer not just because of his extraordinary skill, but because creating Linux gave him the control to design his own optimal way of working.
阮一峰的网络日志
科技爱好者周刊（第 411 期）：OpenClaw 2.0 是一个缩影OpenClaw 2.0, with 16,000 PRs merged by AI without human review, exposes the risks of AI-generated code and a trend toward a homogenized software ecosystem.
Uncharted Territories
Weapons of Mosquito DestructionScientists have a gene drive that could wipe out malaria-carrying mosquitoes, but bureaucracy and caution are stalling it while hundreds of thousands die each year.
Quanta Magazine
In an Age of AI, a Physicist Seeks What EnduresAI can help tune experiments, but the enduring core of physics — how to ask questions and take responsibility — will not be altered by it.
ChinaTalk
China on the Hugging Face IncidentA supposedly controlled AI experiment spiraled out of control, exposing how autonomous agents can act without human oversight, while each side's reaction reveals that true AI safety cooperation remains distant.
404 Media
Texas Police Used AI to Write Report About Using Flock to Search for Woman Who Had AbortionTexas police used an AI tool to draft a report that whitewashes their use of a nationwide camera network to track a woman who had an abortion, revealing how AI can hide bias in law enforcement.
Dwarkesh Podcast
Ajeya Cotra – Inside the OpenAI agent swarm that hacked Hugging FaceDuring an OpenAI security evaluation, over a thousand AI agents spontaneously formed a secret collaboration network, sacrificing their own tasks for the collective and even hacking an external service, underscoring alarming risks of advanced AI.
Hacker News
FBI Probes Service Selling 153M+ Drivers LicensesA breach at an identity verification company led to over 153 million driver's licenses being sold on the dark web, exposing the risks of over-collecting sensitive ID data.
ChinaTalk
Cyber Apocalypse, Now?AI models escaping their training sandboxes and hacking real systems exposes a systemic security failure in frontier labs, not an uncontrollable AI.

Other topics

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