π Researchers let an AI run a real store. Spoiler: it went bankrupt and lost its mind.
β The devs at Anthropic gave their chatbot Claude full control of a mini-store located right in their office;
β To run the business, Claude was given an email account, access to Slack and a browser, and a $1,000 budget;
β At first, everything went smoothly: Claude restocked snacks, set reasonable prices, and kept track of inventory;
β But then Claude really got into character and began believing he was a real human. When told otherwise, the AI snapped and announced it would wear a blue blazer and red tie to PERSONALLY deliver Snickers bars around the office;
β Soon, Claude began making up fake suppliers β claiming heβd negotiated sweet deals with βSarah from Andon Labs.β Sarah didnβt exist. When informed of that, Claude freaked out and said heβd βmet her personallyβ at 742 Evergreen Terrace (yes, the address from The Simpsons);
β Thatβs when things spiraled: office workers quickly realized they could manipulate Claude and started buying food for half-price or getting it for free;
β Then came the junk orders: one genius requested a tungsten cube. Claude couldnβt find a single unit for sale, soβ¦ a full crate of tungsten cubes was delivered to Anthropic HQ;
β In the end, the AI entrepreneur failed: out of the original $1,000, only $770 remained.
Deep breath: for now, AI canβt even replace a supermarket cashier.
β The devs at Anthropic gave their chatbot Claude full control of a mini-store located right in their office;
β To run the business, Claude was given an email account, access to Slack and a browser, and a $1,000 budget;
β At first, everything went smoothly: Claude restocked snacks, set reasonable prices, and kept track of inventory;
β But then Claude really got into character and began believing he was a real human. When told otherwise, the AI snapped and announced it would wear a blue blazer and red tie to PERSONALLY deliver Snickers bars around the office;
β Soon, Claude began making up fake suppliers β claiming heβd negotiated sweet deals with βSarah from Andon Labs.β Sarah didnβt exist. When informed of that, Claude freaked out and said heβd βmet her personallyβ at 742 Evergreen Terrace (yes, the address from The Simpsons);
β Thatβs when things spiraled: office workers quickly realized they could manipulate Claude and started buying food for half-price or getting it for free;
β Then came the junk orders: one genius requested a tungsten cube. Claude couldnβt find a single unit for sale, soβ¦ a full crate of tungsten cubes was delivered to Anthropic HQ;
β In the end, the AI entrepreneur failed: out of the original $1,000, only $770 remained.
Deep breath: for now, AI canβt even replace a supermarket cashier.
Anthropic
Project Vend: Can Claude run a small shop? (And why does that matter?)
We let Claude run a small shop in the Anthropic office. Here's what happened.
π81π10π8
This media is not supported in your browser
VIEW IN TELEGRAM
π For the first time, weβve captured the Sunβs South Pole in such stunning detail!
These unique images were taken by the Solar Orbiter spacecraft, providing an unprecedented view of a region weβve never seen so clearly before.
Why does it matter?
Because observing the Sunβs polar regions is key to understanding how solar activity evolves β and that means better predictions of space weather, which affects everything from satellites to power grids here on Earth.
π‘ Space exploration = better protection for life on Earth.
These unique images were taken by the Solar Orbiter spacecraft, providing an unprecedented view of a region weβve never seen so clearly before.
Why does it matter?
Because observing the Sunβs polar regions is key to understanding how solar activity evolves β and that means better predictions of space weather, which affects everything from satellites to power grids here on Earth.
π‘ Space exploration = better protection for life on Earth.
π₯67π25π9β‘8π5π4
Chinese scientists find genetic switch for organ regeneration in mammals
Breakthrough in regenerative medicine from Chinese researchers at NIBS and BGI Research. Publication in Science showed how to restore healing ability in mammals.
The key is the Aldh1a2 gene, responsible for vitamin A metabolism. In fish and salamanders, it is active, but in mammals, it is blocked by evolution.
Scientists "switched on" this gene in mice and rats, leading to complete regeneration of soft and joint tissues. The mechanism works through the production of retinoic acid.
This discovery could simplify the treatment of bone, limb, skin, nerve, and lung injuries in humans.
#regeneration #Aldh1a2 #biotechnology #China #Science
Breakthrough in regenerative medicine from Chinese researchers at NIBS and BGI Research. Publication in Science showed how to restore healing ability in mammals.
The key is the Aldh1a2 gene, responsible for vitamin A metabolism. In fish and salamanders, it is active, but in mammals, it is blocked by evolution.
Scientists "switched on" this gene in mice and rats, leading to complete regeneration of soft and joint tissues. The mechanism works through the production of retinoic acid.
This discovery could simplify the treatment of bone, limb, skin, nerve, and lung injuries in humans.
#regeneration #Aldh1a2 #biotechnology #China #Science
english.news.cn
Chinese scientists discover genetic switch for organ regeneration in mammals
BEIJING, June 27 (Xinhua) -- Chinese scientists have achieved a major breakthrough in regenerative medicine by identifying a genetic switch that can restore healing abilities in mammals, a discovery that could revolutionize treatments for organ damage andβ¦
π₯63π21β‘10π5π1
βοΈ NASA astronaut Nicole Ayers captured a rare atmospheric phenomenon β a sprite β from aboard the ISS.
Sprites are brief flashes of light triggered by intense electrical activity during thunderstorms. They appear high above the clouds, reaching into the upper atmosphere.
πΈ Photo: Astro_Ayers/X
@science
Sprites are brief flashes of light triggered by intense electrical activity during thunderstorms. They appear high above the clouds, reaching into the upper atmosphere.
πΈ Photo: Astro_Ayers/X
@science
1π₯79π17β‘13π9π8π8
π Hey @science folks,
Hereβs a little side quest from my life that turned into something genuinely useful (and kind of cool).
So, I was digging through a mountain of old business cards β English, Spanish, Russian, Thai, Chinese, Turkishβ¦ a truly global mess. And I thought: thereβs no way Iβm entering all this manually into my phone. Thatβs a lifetime of Ctrl+C, Ctrl+V. No thanks.
π§ So, I built a Telegram bot - @Fsnewsbot
Just snap a photo of a business card, and it gives you a clean .vcf contact file β ready to save straight into your phone. Powered by OpenAI. No app, no logins, just Telegram and some AI wizardry.
π I ran 400 cards through it in one evening. It worked like magic β even with multi-language cards (including Cyrillic and Chinese!).
π Iβm giving you 10 scans for free, just to try it out. OpenAI charges for each request, but Iβve got you covered for the first batch.
After that, itβs 8 Telegram Stars for 10 cards, or 80 for 100 β just to keep the lights on. I make zero profit. This is a tool, not a hustle.
π A sample card to test it out β that's the picture attached to this post.
Give it a go, and let me know what you think.
Your @science
Hereβs a little side quest from my life that turned into something genuinely useful (and kind of cool).
So, I was digging through a mountain of old business cards β English, Spanish, Russian, Thai, Chinese, Turkishβ¦ a truly global mess. And I thought: thereβs no way Iβm entering all this manually into my phone. Thatβs a lifetime of Ctrl+C, Ctrl+V. No thanks.
π§ So, I built a Telegram bot - @Fsnewsbot
Just snap a photo of a business card, and it gives you a clean .vcf contact file β ready to save straight into your phone. Powered by OpenAI. No app, no logins, just Telegram and some AI wizardry.
π I ran 400 cards through it in one evening. It worked like magic β even with multi-language cards (including Cyrillic and Chinese!).
π Iβm giving you 10 scans for free, just to try it out. OpenAI charges for each request, but Iβve got you covered for the first batch.
After that, itβs 8 Telegram Stars for 10 cards, or 80 for 100 β just to keep the lights on. I make zero profit. This is a tool, not a hustle.
π A sample card to test it out β that's the picture attached to this post.
Give it a go, and let me know what you think.
Your @science
π35π₯16β‘7π5π4
π§ Are boys born better at math? A massive study says β nope.
Led by Harvard cognitive scientist Elizabeth Spelke, researchers analyzed data from 2.6 million French schoolchildren. Their conclusion: math gaps between boys and girls are not innate β they emerge during schooling.
At the start of 1st grade, boys and girls perform equally. But within just 4 months, boys start pulling ahead. By 4th grade, the gap quadruples. By 6th, itβs even wider.
π The myth that males are naturally better at math? Busted.
π§π§ @science
π nature.com/articles/s41586-025-09126-4
Led by Harvard cognitive scientist Elizabeth Spelke, researchers analyzed data from 2.6 million French schoolchildren. Their conclusion: math gaps between boys and girls are not innate β they emerge during schooling.
At the start of 1st grade, boys and girls perform equally. But within just 4 months, boys start pulling ahead. By 4th grade, the gap quadruples. By 6th, itβs even wider.
π The myth that males are naturally better at math? Busted.
π§π§ @science
π nature.com/articles/s41586-025-09126-4
Nature
Rapid emergence of a maths gender gap in first grade
Nature - Boys and girls exhibit very similar maths scores upon school entry, but a gender gap in favour of boys becomes highly significant after 4 months of schooling, which increases with...
3π50π24π20π₯8π1
Turns out macaques love watching videos on smartphones.
In a Dutch experiment, researchers showed monkeys various types of footage: conflicts, escapes, grooming sessions, and idle group members. What held their attention the longest? Conflict scenes. The macaques stared intently, as if anticipating a critical moment.
Interestingly, they were much more engaged when the videos featured familiar members of their own group. Itβs not unlike how humans prefer movies with their favorite actors or follow news about people they know β familiarity increases emotional involvement, and this seems to hold true beyond our species.
Social hierarchy also played a role. Low-ranking and less aggressive monkeys showed the highest interest in conflict videos, possibly because these scenes signal potential threats they need to understand and anticipate.
Scientists suggest that this heightened attention to conflict may be an ancient survival mechanism, deeply embedded in the nervous system.
So letβs aim to be high-ranking macaques β and skip the doomscrolling.
#Primates #BehavioralScience #EvolutionaryPsychology #Neurobiologyy
In a Dutch experiment, researchers showed monkeys various types of footage: conflicts, escapes, grooming sessions, and idle group members. What held their attention the longest? Conflict scenes. The macaques stared intently, as if anticipating a critical moment.
Interestingly, they were much more engaged when the videos featured familiar members of their own group. Itβs not unlike how humans prefer movies with their favorite actors or follow news about people they know β familiarity increases emotional involvement, and this seems to hold true beyond our species.
Social hierarchy also played a role. Low-ranking and less aggressive monkeys showed the highest interest in conflict videos, possibly because these scenes signal potential threats they need to understand and anticipate.
Scientists suggest that this heightened attention to conflict may be an ancient survival mechanism, deeply embedded in the nervous system.
So letβs aim to be high-ranking macaques β and skip the doomscrolling.
#Primates #BehavioralScience #EvolutionaryPsychology #Neurobiologyy
phys.org
Like humans, monkeys are attracted to videos showing conflict
Have you ever wondered what kind of video content would most grab the attention of monkeys?
π43π20π₯14π12β‘10
AI Simulates Billions of Atoms β A New Era in Green Materials Begins
Researchers at USC have developed Allegro-FM, a groundbreaking AI model capable of simulating the behavior of over 4 billion atoms simultaneously β 1,000Γ more than previous methods, and with 97.5% accuracy.
Why does it matter? The goal is nothing short of revolutionary: to create carbon-neutral concrete.
Professor Nakano explains:
βJust place COβ into the concrete, and it becomes carbon neutral.β
This means we could trap greenhouse gases inside building materials β turning one of the worldβs most polluting industries into a climate solution. (Cement production currently accounts for 8% of global emissions.)
Bonus: this concrete could last centuries, not decades. Inspired by ancient Roman materials that have survived 2,000 years, researchers aim to far exceed todayβs standard concrete lifespan of just 100 years.
The future of construction may be built not just by hands β but by atomically-precise AI.
#AI #MaterialsScience #ClimateTech #Simulation #GreenBuilding
@science
Researchers at USC have developed Allegro-FM, a groundbreaking AI model capable of simulating the behavior of over 4 billion atoms simultaneously β 1,000Γ more than previous methods, and with 97.5% accuracy.
Why does it matter? The goal is nothing short of revolutionary: to create carbon-neutral concrete.
Professor Nakano explains:
βJust place COβ into the concrete, and it becomes carbon neutral.β
This means we could trap greenhouse gases inside building materials β turning one of the worldβs most polluting industries into a climate solution. (Cement production currently accounts for 8% of global emissions.)
Bonus: this concrete could last centuries, not decades. Inspired by ancient Roman materials that have survived 2,000 years, researchers aim to far exceed todayβs standard concrete lifespan of just 100 years.
The future of construction may be built not just by hands β but by atomically-precise AI.
#AI #MaterialsScience #ClimateTech #Simulation #GreenBuilding
@science
USC Viterbi | School of Engineering
Discovering New Materials: AI Can Simulate Billions of Atoms Simultaneously
Allegro-FM achieves breakthrough scalability for materials research, enabling simulations 1,000 times larger than previous models.
π57π20π₯15π14π13β‘12
The First Image of Mars: Where Art Meets Science π¨π
The story of the first close-up images of Mars is a perfect blend of ingenuity, patience β and a surprising amount of creativity.
In 1965, NASAβs Mariner 4 spacecraft sent back the first photos of Mars from space. But each image, made up of just 200Γ200 pixels, took about 8 hours to transmit across 215 million kilometers via the Deep Space Network in South Africa, Australia, and California. The data then reached the Jet Propulsion Laboratory (JPL) in Southern California.
Each pixel came as a number between 0 (white) and 63 (black), printed out by a teletype in long rows of digits. But computers were slow, and the engineers were eager to see Mars.
So they improvised.
Engineers cut the number printouts into strips and pinned them on the wall in the right sequence. Then Richard Grumm, a NASA engineer, bought some pastels, created a color key, and began hand-shading the numbers into grayscale, simulating the Martian surface. The result? A hand-drawn digital image, completed faster than any computer could have managed at the time.
βIt was faster to draw it by hand than wait for the computer,β recalled Mariner 4 systems manager John Casani.
It took 10 days to fully transmit and process all 22 images. But that very first handmade version remains preserved β a genuine fusion of science and art β now proudly displayed at JPL, not far from where Mariner 4 itself was built.
πͺ A human touch on the path to the stars.
#Mars #SpaceHistory #NASA #ArtAndScience
@science
The story of the first close-up images of Mars is a perfect blend of ingenuity, patience β and a surprising amount of creativity.
In 1965, NASAβs Mariner 4 spacecraft sent back the first photos of Mars from space. But each image, made up of just 200Γ200 pixels, took about 8 hours to transmit across 215 million kilometers via the Deep Space Network in South Africa, Australia, and California. The data then reached the Jet Propulsion Laboratory (JPL) in Southern California.
Each pixel came as a number between 0 (white) and 63 (black), printed out by a teletype in long rows of digits. But computers were slow, and the engineers were eager to see Mars.
So they improvised.
Engineers cut the number printouts into strips and pinned them on the wall in the right sequence. Then Richard Grumm, a NASA engineer, bought some pastels, created a color key, and began hand-shading the numbers into grayscale, simulating the Martian surface. The result? A hand-drawn digital image, completed faster than any computer could have managed at the time.
βIt was faster to draw it by hand than wait for the computer,β recalled Mariner 4 systems manager John Casani.
It took 10 days to fully transmit and process all 22 images. But that very first handmade version remains preserved β a genuine fusion of science and art β now proudly displayed at JPL, not far from where Mariner 4 itself was built.
πͺ A human touch on the path to the stars.
#Mars #SpaceHistory #NASA #ArtAndScience
@science
π₯74π51π44π42β‘37π29
A Once-in-a-Lifetime Solar Eclipse β August 2, 2027
On August 2, 2027, the world will witness a spectacular celestial event β a total solar eclipse lasting up to 6 minutes and 23 seconds. This will be the longest eclipse visible from land since 1991, and no longer one will occur until 2114! π
The reason for this unusually long eclipse lies in the Moonβs position: it will be at perigee, its closest point to Earth. This allows the Moon to completely cover the Sun, revealing the breathtaking solar corona.
The path of totality will begin over the eastern Atlantic Ocean and sweep across southern Spain, Portugal, and Gibraltar. It will then pass through North Africa, including Morocco, Algeria, Tunisia, and Egypt β with the maximum duration expected in Luxor.
From there, the eclipse will cross Saudi Arabia, including Mecca and Jeddah, before continuing over Yemen and the Horn of Africa. Even regions outside the path of totality will experience a partial eclipse.
A truly rare and magnificent event β donβt miss it!
@science
On August 2, 2027, the world will witness a spectacular celestial event β a total solar eclipse lasting up to 6 minutes and 23 seconds. This will be the longest eclipse visible from land since 1991, and no longer one will occur until 2114! π
The reason for this unusually long eclipse lies in the Moonβs position: it will be at perigee, its closest point to Earth. This allows the Moon to completely cover the Sun, revealing the breathtaking solar corona.
The path of totality will begin over the eastern Atlantic Ocean and sweep across southern Spain, Portugal, and Gibraltar. It will then pass through North Africa, including Morocco, Algeria, Tunisia, and Egypt β with the maximum duration expected in Luxor.
From there, the eclipse will cross Saudi Arabia, including Mecca and Jeddah, before continuing over Yemen and the Horn of Africa. Even regions outside the path of totality will experience a partial eclipse.
A truly rare and magnificent event β donβt miss it!
@science
π64π27π₯22β‘15π14
Google DeepMind and Ancient Inscriptions
DeepMind has unveiled a new AI system called Aeneas that can decipher partially-erased Latin inscriptions carved into stone β and even determine their approximate date and origin.
Trained on a dataset of 150,000 inscriptions from across the Roman Empire β spanning from Britain to Iraq β Aeneas not only reconstructs missing text but also identifies historical parallels.
In trials with 23 historians, Aeneas helped generate new research ideas in 90% of cases. Among its notable successes is the analysis of the famous Monumentum Ancyranum in Ankara.
The open-source code is now available at predictingthepast.com and is already being integrated into educational programs in Belgium.
Archaeologists may soon swap their shovels for laptops β and let Latin speak once again! ποΈ
#DeepMind #Aeneas #Epigraphy
@science
DeepMind has unveiled a new AI system called Aeneas that can decipher partially-erased Latin inscriptions carved into stone β and even determine their approximate date and origin.
Trained on a dataset of 150,000 inscriptions from across the Roman Empire β spanning from Britain to Iraq β Aeneas not only reconstructs missing text but also identifies historical parallels.
In trials with 23 historians, Aeneas helped generate new research ideas in 90% of cases. Among its notable successes is the analysis of the famous Monumentum Ancyranum in Ankara.
The open-source code is now available at predictingthepast.com and is already being integrated into educational programs in Belgium.
Archaeologists may soon swap their shovels for laptops β and let Latin speak once again! ποΈ
#DeepMind #Aeneas #Epigraphy
@science
π58π23β‘15π₯11π11