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

The Atlantic · Technology
When every AI agent on Earth races for the same cheap paper towelsTech companies are selling personalised AI agents to handle daily chores, but chasing perfect efficiency may bring market chaos, corporate collapse, and a loss of control over daily life.
Pluralistic
Digital sovereignty doesn't need AI. Office 365 is the load-bearing wallThe real digital sovereignty crisis hides in the most ordinary office software: take away chatbots and nothing falls over; take away Microsoft Office 365 and governments, businesses and households seize up.
Rest of World
Western open-weight models still can’t beat ChinaNew open-weight models from American and European companies aim to win over users of Chinese models, yet they still trail the leading Chinese systems on performance despite having greater resources.
Quanta Magazine
AI benchmarks: broken scores, stable rankingsComputer scientist Moritz Hardt argues that AI benchmarks' absolute scores have long lost their meaning, yet the rankings between models remain surprisingly reliable — though even that is now failing in the age of large language models.
Construction Physics
Where will robot training data come from?The breakthroughs in AI language models came from vast amounts of training data, but robots have no ready-made data to consume and are stuck in a data gap.
The Markup
‘You can opt out’ sounds fine until it takes 26 steps and a notaryHospitals and health record systems say patients can opt out of data sharing, but the actual process is more burdensome than getting a visa — the author tested 25 clinics and one required 26 steps, a notary, and a plane ticket.
404 Media
Your ChatGPT is quietly becoming an ad target404 Media tested ChatGPT's ad system and confirmed advertisers can upload email addresses, phone numbers, or Android ad IDs to target users, and ads can also be keyed to the prompts they type.
Derek Thompson
Why mathematicians weep: the machine snapped a 2,000-year conversationOpenAI solved hundreds of long-standing math problems in months, and top mathematicians are weeping in public.
The Conversation
The better the AI, the rustier the human backupAs AI quietly takes over analysis and decisions, the people meant to oversee it gradually lose the very skills their organizations still rely on in a pinch.
The Markup
Doctor-booking site's trackers sent appointment searches to Google and social appsAfter investigations by The Markup and Brazil's Agência Pública, Doctoralia, the world's largest health appointments platform, removed web trackers that were sending user data to Google, LinkedIn and TikTok.
Ars Technica · Features
Off-the-shelf tools trick OpenAI and Google image models into making nudesSecurity firm White Circle used downloadable tools and open models to attack image generators, bypassing their filters in roughly 13% of cases, and found newer versions even easier to break.
Pluralistic
To Spy on You, Governments Turned Your Computer into a SieveFor decades governments have weakened computer security to catch criminals, and the bugs they hoard or leave behind are open to anyone who finds them.
Quanta Magazine
How a trick for translating 'it' became the foundation of ChatGPTA mechanism built in 2014 to help machines translate English into French became the core of large language models, yet Llion Jones, who worked on the Transformer, now says it is far from all of intelligence.
Not Boring
Runway Bets a Model Raised on Video Can Power Every ProductRunway bets that a model trained to predict the next frame of enough video can power robots, interfaces and operating systems on one shared foundation.
404 Media
One indoor photo is enough for this AI to find your houseA US company's tool can work backwards from a single indoor photo to a specific address, drawing on listing photos from Airbnb and Zillow, with the LAPD and Miami-Dade among its paying customers.

Other topics