The chip is almost the boring part.
The jump came from gluing 72 GPUs into one memory fabric, so a giant model keeps thousands of agents' context straight without choking.
NVIDIA didn't ship a faster GPU. They shipped a faster hallway.
Primary source: π https://t.co/IUKxjhlQP5
Everyone's resharing NVIDIA's "~61K AI agents" chart.
It's per megawatt. Not per rack.
By NVIDIA's own per-GPU math, one GB300 rack runs ~4,140 agents. Still ~20x over Hopper.
The shift: the scoreboard is now agents-per-watt. Power is the ceiling.
Read the small print. π
#AI
OpenAI reportedly offered ~$500M for its gameplay data. General Intuition said no.
Now it's reportedly raising ~$300M at ~$2B.
Everyone sells the world model. GI sells the agent that learns inside it.
The moat is data no one else owns. What would you refuse to sell?
#AI
Physicists measured a metal where electrons stopped acting alone and moved as one web. The bond between them peaked at the edge of change, not in the calm. Strong systems aren't built from stars. They're built from connections. Your team: more aligned in comfort or chaos #Quantum
Source(s): NASA, asteroid Donaldjohanson, published in Science (18 June 2026). π https://t.co/kFOxC0V1xA
Note: the "reshaped by sunlight" bit is the YORP effect, where re-radiated heat slowly changed the asteroid's spin.
π°οΈ Three asteroids. Same scientific label. Opposite life stories.
NASA's Lucy proved it up close: Donaldjohanson is 155M years old, Bennu and Ryugu are billions, and it held water only briefly.
"Looks like" is not "is like."
What are you judging by analogy? π
#FirstPrinciples
"Food jutsu" - creators borrow the anime "domain expansion" hand gesture (from Jujutsu Kaisen) and use it to "summon" a meal in a sharp cut, set to the track Delirious from the Jujutsu Kaisen score. The edits come fast and need a steady hand and a sharp cut to land well.
Nobody explains why "Food jutsu" actually works.
It's not the anime hand sign.
It's the cut. One frame off and the illusion breaks.
That "effortless" clip? Pure editing discipline.
What trend looks easy but absolutely is not? π#CreatorEconomy#ContentCreation#Storytelling
Source(s): A study published in Science reports that ~16 min after Japan's 2011 quake, GPS stations recorded a synchronized eastward shift of up to ~6 mm. πhttps://t.co/2PsOZgPP1y
~16 min after Japan's 2011 quake, the country lurched east. No aftershock explained it.
New in Science: Its waves hit Earth's core, bounced back, and moved the island up to ~6 mm.
Hidden in the data for 15 years. Sensors were tuned to miss it.
What "noise" are you ignoring?
Source(s): Midjourney (the AI image company, led by David Holz) has announced "Midjourney Medical" and a full-body ultrasonic scanner as its first hardware product.
π https://t.co/JGmhCX6cO9
π https://t.co/CJO3KtvV6B
Relevant source: on May 31, 2026, motion-capture firm ChingMu ran a "100 people, simultaneously entering the digital world" challenge in Shanghai, with AMD as compute partner and Unitree among the partners. URL: https://t.co/yW3waHENVx
π€One dancer. A mocap suit. 6 humanoid robots copying her every move.
We noticed show business may be the only place people teach robots wholeheartedly. Crowds still pay for humans.
Chinese saying: "Teach apprentice well & master starves".
Teach AI all your skills or hold back?
Source: Nature, DOI 10.1038 (Univ. of Surrey)
π https://t.co/wdyWc1UYjl
Note: a 2025 Nature study used supercomputer simulations to predict a hidden new class of star system (globular-cluster-like dwarfs) that looks like ordinary clusters but holds dark matter...
Over 3 centuries we sorted star systems into 2 kinds.
A supercomputer predicted a hidden 3rd: looks like an ordinary star cluster but holds invisible mass.
Some we've catalogued may be misfiled. Astronomers plan to point Webb at one nearby.
What have u called "ordinary" too soon?
Helion's Polaris hit 150 million degrees Celsius. 10x hotter than the sun's core.
First private fusion machine to show measurable D-T fusion (Feb 2026).
Microsoft is already a buyer for plant #1.
No grid fusion yet. The gap is closing fast.
#Fusion#DeepTech#Energygrowthvent
AI-designed drugs: overwhelming majority of Phase I success rate. Historical industry average: near-even split. Approaching 200 AI-originated programs in clinic. Insilico's Rentosertib approaching Phase III. No FDA approval yet. 2027 might be verdict year. #AI#Pharma#BioTech
Figure 02 humanoid robots just retired.
11 months at BMW. 30,000 cars built. 90,000 parts placed. 99% accuracy.
Figure 03 heads to BMW Leipzig this summer.
Humanoid robots now have a replacement cycle. We're not in demo land anymore.
#HumanoidRobots#PhysicalAI#FrontierTech