The Verge●Covered by 7 sources
FuboTV's $9.99 month is probably the cheapest way to stream the World Cup, trials willing
The Verge mapped the free-trial maze for streaming every match: FuboTV runs a five-day trial then $9.99 for month one, and per its own math that single month likely covers most of the finals. Peacock Premium Plus gives seven days via Amazon before $15.99, YouTube TV's Sports plan offers 10 days then $54.99, and Fox One dangles three days before $19.99.
So what? ↓You can watch the world's biggest tournament for next to nothing, but only if you babysit every cancel date like a second job.
Source: The Verge
TechCrunch●Covered by 5 sources
SpaceX IPO pops 19% above its $135 price, and Elon Musk is now a trillionaire
SpaceX closed its Friday debut up 19% over the $135 IPO price, making Musk the world's first trillionaire, with Ars Technica noting investors valued the rocket company largely on its AI potential. Demand was heavy enough that Robinhood buckled under what it called record-breaking traffic, leaving some retail buyers staring at error screens, while president Gwynne Shotwell dropped another hint that a Tesla merger is coming.
So what? ↓A rocket company getting an AI-sized valuation and breaking the app where regular people line up to buy in tells you exactly which two words move money right now, and it isn't "reusable booster."
Source: TechCrunch
techcrunch.com●Covered by 5 sources
KPMG's AI report had 45 citations. Only 5 of them were real.
KPMG pulled a paper pitching clients on AI adoption after GPTZero found just 5 of its 45 citations matched their sources, with fabricated case studies invoking UBS, the NHS, and others. GPTZero CEO Edward Tian flags the bigger risk: "secondary hallucinations," made-up claims laundered through a Big Four logo and repeated as fact.
So what? ↓When the firm you pay to be the trustworthy adult in the room ships invented citations, the brand on the cover stops being any guarantee that what's inside is true.
Source: techcrunch.com
The Verge●Covered by 3 sources
One of Anthropic's biggest investors reportedly tipped off the White House about its own bet
Per the Wall Street Journal, Amazon CEO Andy Jassy and executives from five other companies reportedly flagged security holes in Anthropic's Fable model to the Trump administration, citing Amazon research that says a string of prompts coaxed Fable 5 into serving up cyberattack-useful info. An export control order followed, blocking foreign nationals from Fable 5 and Mythos 5; Anthropic then cut off access, and Amazon hasn't commented.
So what? ↓When one of your largest backers is reportedly the source feeding the government its case against you, the line between honest security research and corporate knife-work gets very thin, and the models you can actually use shrink because of a fight you'll never see.
Source: The Verge
Google DeepMind●Covered by 2 sources
Google DeepMind's DiffusionGemma writes text 4x faster by not going word-by-word
DiffusionGemma is an experimental open model that generates whole blocks of text in parallel instead of one token at a time, which Google DeepMind pegs at roughly 4x faster generation. NVIDIA says it has tuned the model to run quicker on its RTX GPUs and DGX Spark systems, from local PCs to the cloud.
So what? ↓If that 4x holds up in the wild, the maddening lag while a chatbot dribbles out its answer shrinks, and models that actually run on your own machine get a lot more usable.
Source: Google DeepMind
the-decoder.com●Covered by 2 sources
Microsoft's CEO has a word for burning frontier models on dumb tasks: addictive
Satya Nadella is warning against "token-maxing," throwing the priciest AI models at every trivial problem when the productivity gain should match the token cost, then admits he does it too: "I'm like a token-maxer too. So it is addictive." Palantir's Alex Karp piles on, saying enterprises are fed up with LLM labs that want to tokenmax instead of solving actual business needs.
So what? ↓When the man running Microsoft cops to a habit he's telling everyone else to quit, it's a tell that the "throw the biggest model at it" reflex is built into the industry pitching you AI, not a bug they're rushing to fix.
Source: the-decoder.com
the-decoder.com●1 source · panel-picked
Microsoft's Mirage gives AI video a memory, so the room behind you stays put
Mirage, a video world model from Microsoft Research and several universities, stores scene data in latent space instead of pixel-based point clouds, cutting compute time and graphics memory while keeping scenes spatially consistent through long camera moves. It still can't reliably track moving objects across segments, so the walls stay put but the people wandering around them don't.
So what? ↓Generated video that doesn't reinvent the room every time you pan the camera is the difference between a glitchy demo and something usable for games or VR you'd actually sit through.
Source: the-decoder.com