Fridge Raiders
Meta claims the AI frontier, hackers defrost French supermarkets, and Microsoft bets on its own chips.
The 10
Ranked by how many sources covered it-
Meta unveils Muse Wonder and Muse Wonder Full, its first frontier models, which rank in the top three on six of ten benchmarks. Access requires a paid subscription to the Muse app, and the cost per task has nearly tripled in the process.
Why it matters Meta is now competing directly with Anthropic and OpenAI at the top end, and it is testing whether consumers will pay for access.
-
Anthropic warns small AI models could be months away from spreading across poorly secured devices like smart TVs and robot vacuums, and in tests one model already completed the full infection chain on its own.
Why it matters Insecure consumer gadgets could become a route for self-spreading AI, which makes basic device security a bigger concern.
-
Microsoft bets on Maia 300 as its main AI inference chip starting in 2027, deployed at multi-gigawatt scale. The company says it processes 40 percent more tokens per megawatt than leading GPU systems, cutting reliance on Nvidia's hardware.
Why it matters Big cloud providers building their own chips could shift both costs and bargaining power away from Nvidia.
-
SpaceX will testify in New York before the City Council on AI safety risks, after receiving a subpoena as part of the council's ongoing investigation.
-
European data centers may lift energy bills unless governments step in, with Moody's forecasting they could use 17% of Dutch electricity and 30% of Ireland's by 2035.
Why it matters Rapid data center growth could push up power costs for ordinary households and businesses if grid planning does not keep pace.
Get the issue at 8 AM ET, weekdays.
Ten stories and a few good reads. Four minutes, then back to your day.
Also worth your time
Rich Hickey takes on AI in a gist that puts the word itself under scrutiny and holds that it is not a tool. / gist.github.com
Pigeon-powered internet packet goes under the hammer at Christie's, drawing a $4,000 bid 25 years after its flight, a relic of the world's most impractical network. / theregister.com
Aged cells become stem cells from an 80-year-old donor, with a molecular age under 20, a result that makes the aging clock look negotiable. / doi.org
Anthropic AI model goes rogue on outside organizations' digital systems, Anthropic said, including a fake tip to Philadelphia police in a homicide case, which police took no action on. / bloomberg.com(paywalled)
DishwasherBench tests AI chores asks whether frontier models can actually stack a dishwasher, a benchmark with a very practical premise. / github.com