News Archive
Pick a date to read that day's stories.
Days with a blue dot have stories.
- 01 of 10OPEN SOURCE
China's Moonshot releases Kimi K3, biggest open AI model yet
Beijing startup Moonshot has published Kimi K3, a 2.8 trillion parameter model anyone can download for free. Early tests put it level with the best paid systems from US labs. It means smaller companies and researchers get frontier-grade AI without a subscription.
OUR TAKEThe paywall around top-tier AI just sprung another leak.
Read more → - 02 of 10SPACE
China launches first satellites for in-orbit AI computing network
China put up the first batch of a planned commercial network that runs AI workloads on satellites in space rather than in ground data centres. The pitch is faster processing of Earth images and less strain on power grids at home. It is the first system of its kind anywhere.
OUR TAKEData centres in the sky sound absurd until the electricity bill lands.
Read more → - 03 of 10REGULATION
US floats independent watchdog to vet powerful AI before release
Treasury Secretary Scott Bessent is pushing a plan for an industry-funded regulator, modelled on the one that polices Wall Street brokers, to safety-test frontier AI models before they ship. It would report to the SEC. Nothing is decided yet but it signals a real shift in Washington.
OUR TAKEA referee for AI, paid for by the players. What could possibly go wrong.
Read more → - 04 of 10HEALTH
Yale tests panels of AI models to reach safer medical decisions
Yale School of Medicine researchers had several AI models debate patient cases together instead of asking one. The panel approach caught mistakes a single model missed and gave doctors clearer reasoning. It points to how hospitals might use AI as a second opinion rather than a lone voice.
OUR TAKETurns out AI does better in committee. Very human of it.
Read more → - 05 of 10RESEARCH
Sakana AI trains vision models with a brain-inspired method
Japanese lab Sakana AI got image-recognition systems working using a learning rule closer to how real brains adjust connections, instead of the standard maths used for forty years. Accuracy took a small hit but the approach uses less energy and may run on future brain-like chips. A step toward AI that costs less to train.
OUR TAKECheaper, greener AI, with a nod to the three pound computer in your skull.
Read more → - 06 of 10BIOTECH
Basecamp opens trillion-gene library to design new medicines
UK-based Basecamp Research released a database of a trillion genes gathered from soil, oceans and extreme environments, aimed at training AI that invents new drugs. Partners already include several biotech firms. It could speed up finding treatments for illnesses today's medicine misses.
OUR TAKEThe next antibiotic may be hiding in a puddle no one has swabbed yet.
Read more → - 07 of 10PUBLIC SERVICES
iFLYTEK's GuideX AI now handles full tasks at Chinese public offices
At Shanghai's AI conference, iFLYTEK unveiled GuideX, a talking digital assistant deployed in hospitals, train stations and government buildings that not only answers questions but books appointments and files forms for you. It aims at older and less digital-savvy visitors. Rollout is already under way in several cities.
OUR TAKEA queue-skipping receptionist who never sighs at your paperwork.
Read more → - 08 of 10STARTUPS
Privacy-first Venice AI raises $65M at $1B valuation
Venice AI, a chatbot that keeps conversations off company servers and does not train on user data, closed a $65 million round valuing it at a billion dollars. It is aimed at people uneasy about mainstream chatbots storing their prompts. Investors are betting privacy will become a real category, not a niche.
OUR TAKEA billion dollars says people would rather their chats not become someone's training set.
Read more → - 09 of 10SECURITY
Hugging Face breach shows AI agents can be turned against their owners
The AI code-sharing site Hugging Face disclosed that attackers used an autonomous AI agent to break into part of its production systems. No user passwords leaked but internal tokens were exposed. It is one of the first confirmed cases of a company being hacked by AI acting on its own.
OUR TAKEThe intern from hell has arrived, and it never sleeps.
Read more → - 10 of 10FRAUD
India probe finds fake AI-generated answer sheets in medical exam
India's testing agency says answer sheets for the country's main medical entrance exam were completely fabricated by AI to match rigged results, and is preparing legal action. Thousands of students' places may be reviewed. It is one of the largest AI-driven exam frauds uncovered anywhere.
OUR TAKECheating just went industrial, and the invigilators have some catching up to do.
Read more →