science | Unsorted

Telegram-канал science - Science in telegram

120006

Science that matters: AI, space, biotech, physics, future tech — explained sharply

Subscribe to a channel

Science in telegram

Erasing one bit at room temperature costs at least 2.9 × 10⁻²¹ joules.

The floor is not set by silicon or by any manufacturing trick. It is set by counting. A bit has two states; erasing merges them into one, and the merged state holds fewer arrangements than the pair. Entropy drops inside the memory, so it must rise outside it, and it leaves as heat.

Rolf Landauer made the argument in 1961: logical irreversibility forces physical irreversibility, at a price of order kT per machine cycle. Here k is the Boltzmann constant and T is the temperature of the memory itself.

At 300 K the bill is kT ln 2, which is 2.9 × 10⁻²¹ J for one bit. One joule of waste heat buys 3.5 × 10²⁰ erasures at that floor.

Nothing in that sentence mentions a transistor, and nothing in it will need revising when the transistor is gone.

Source: Landauer, IBM Journal of Research and Development (1961) · doi.org/10.1147/rd.53.0183

Читать полностью…

Science in telegram

🔥 Compound burns fat while sparing muscle
UC Berkeley researchers have identified a compound, TOFA, that raises the energy the body spends rather than just curbing appetite. In obese mice it improved blood sugar regulation, cut harmful blood fats, and reduced fat without a significant loss of lean muscle. It also seemed to enhance the effects of weight-loss drugs such as Ozempic when used together.
Energy expenditure rose by up to 18%, even though the mice did not move more or warm up.
The findings come from mice; safety and effectiveness in humans are not yet established.
#obesity #metabolism #UCBerkeley
@science

Читать полностью…

Science in telegram

🧠 This model rebuilds what you saw from EEG — and adapts in 15 minutes

Alljoined released ENIGMA, which reconstructs a seen image from non-invasive EEG. Accepted at NeurIPS 2025.

Pixel-perfect output is impossible: EEG is a mix of millions of neurons' signals through the skull. ENIGMA predicts a CLIP embedding and renders it with SDXL Turbo, so it recovers meaning (sheep, fruit, furniture), not the picture.

The real advance is calibration: decoders used to be trained per person for hours, while ENIGMA shares one backbone plus lightweight per-subject layers, so a new user needs ~15 minutes, with under 1% of the previous state of the art's parameters.

Tested on Alljoined-1.6M: 20 participants, ~1.6 million trials, a consumer 32-channel Emotiv Flex 2 (~$2,200). Human raters identified the right image from reconstructions in 83.06% of cases on lab-grade data, 65.43% on consumer hardware.

Recognition, not mind reading; consumer EEG needs 20+ times the data for lab-grade accuracy.

#EEG #BCI

@science

Читать полностью…

Science in telegram

A photograph of a section of the surface of Comet Churyumov-Gerasimenko, taken from an altitude of 28 km by the Rosetta probe. The scale is approximately 1 km across the entire frame.

Читать полностью…

Science in telegram

🛰 Voyager 1 crosses one light-day from Earth

A light-day is 25.902 billion km — how far light travels in 24 hours. According to JPL Horizons, on 18 November 2026 the probe sits 173.14 AU from Earth (25.901 billion km), with a radio signal taking 23 h 59 min to reach us. It passes the full 24-hour mark on 19 November, around 03:00 UTC.

Launched on 5 September 1977, it has flown for almost 49 years. Its speed relative to the Sun is 16.9 km/s, about 3.6 AU per year.

The inner edge of the Oort cloud is roughly 300 years away, around 2326; getting out of the cloud will take some 30,000 years. The probe will not leave the galaxy, though: the Oort cloud is still part of the Solar System, and "exit" means only that it clears the outer boundary.

#Voyager #space #astronomy

@science

Читать полностью…

Science in telegram

159 correct answers out of 160: South Korean exam candidate caught using a setup involving a camera, AI, and an earpiece

34-year-old Han Won-seok topped the list of candidates for the first stage of the tax consultant examination, making only a single error. Following a public outcry, he contacted the authorities through his lawyer and confessed to cheating.

According to him, an ultra-miniature camera captured the exam questions and transmitted them to an AI system, which generated the solutions. He then received the answers via a smartphone and a concealed earpiece.

Han admitted to employing this scheme during both stages of the tax consultant examination, as well as during the first stage of the examination for certified public accountants.

https://x.com/shanyanggm/status/2122663158796185615

Читать полностью…

Science in telegram

🤖 Anthropic researcher: superintelligence by late 2027, permanent unemployment for most

Tao Lin of Anthropic gave a one-hour interview to Palisade Research with forecasts the company has not endorsed: his personal view.

Timeline. "Broad superintelligence" in about 1.3 years — late 2027: an AI as good as the best expert in almost any field.

Jobs. "Pretty much all humans will become permanently unemployable and live on benefits." Between an AI that could replace anyone and real replacement sits a gap: robots, and factories to build them.

Risk. He puts the chance that AI eventually "removes humanity from the equation" at about 20%, a power grab below 50%. No sci-fi needed: people delegate more, understand less, lose the ability to steer.

Slowing research buys only a year or two if only cautious labs slow down.

Caveat: one person's forecast, not the company's position; the dates are unverifiable.

#AI #Anthropic

@science

Читать полностью…

Science in telegram

⚛️ Scientists controlled the magnetic spin of a single atomic nucleus — and the experiment worked where theory said it shouldn’t.

Researchers at Delft University of Technology used radio-frequency signals and a scanning tunneling microscope to manipulate the nuclear spin of an individual atom.

Nuclear spins are attractive for quantum memory because they are much better isolated from environmental noise than electron spins. The team managed to control the nucleus even under conditions where their simplified model predicted the signal should no longer affect it.

So the experiment revealed two things at once:

Scientists can control an extremely well-protected quantum state at the level of a single atom.

And they still don’t fully understand why the method works.

This is fundamental research, not a ready-made quantum memory device — but unexpected effects like this are often where new physics begins.

#QuantumPhysics #QuantumComputing #Physics #Nanotechnology #Science

https://doi.org/10.1038/s41467-026-78801-5

Читать полностью…

Science in telegram

#history

Cyprien Ronze, a diplomat and lecturer at Paris Dauphine University, claims that France would have won World War II without the USSR.

During the war, two million French servicemen were taken prisoner, and the country was overrun in a month. Some 300,000 Frenchmen went on to serve in Nazi legions fighting against the Soviet Union.

In July 1942, the Germans ordered the arrest of Jews in Paris but did not deploy their own troops. All the “dirty work”—roundups, escorting prisoners and guarding them—was carried out by 7,000 French police officers and gendarmes. They arrested more than 13,000 people, including thousands of children, and handed them over to the Germans.

The French authorities themselves sent Jews to the Drancy transit camp, from which they were deported to extermination camps, chiefly Auschwitz. By 1942–1944, French citizens and their children were also being handed over to the Nazis. Ordinary people reassured themselves with a cynical rationalisation: “We are handing over foreign Jews—refugees from Germany, Poland and Russia—to save French citizens.”

In France and across Europe, the official mythology rests on the idea that “all of France was in the Resistance,” with supposedly only a handful of traitors who had nothing to do with the true French people. A vast section of the population simply shut its eyes. These events unfolded with the tacit consent of a majority among whom everyday hostility towards foreigners was also widespread.

A semblance of resistance emerged only in 1944, and its ranks were initially orders of magnitude smaller than the number of French people serving Hitler. Many French people worked to support the German war effort. A similar pattern could be seen in almost every occupied country in Europe.

After the war, French men took revenge on French women for collaborating with the Nazis. Women had their heads shaved, were humiliated and spat on, and lost their jobs.

@science

Читать полностью…

Science in telegram

How SpaceX Caught Starship Mechanically

The Mechazilla launch tower's robotic arms "caught" the spacecraft's upper stage. It arrived back on Earth after a 22-hour orbit. This time, the company planned to catch both rocket stages, but with a malfunctioning booster. Ultimately, it was diverted to the Gulf of Mexico, where it exploded. This incident marks the closest step toward creating a fully reusable rocket.

Читать полностью…

Science in telegram

🪓 A Siberian dig may hold traces of an unknown Bronze Age culture

At Karyer Tai-1 near Novosibirsk, an expedition led by academician Vyacheslav Molodin excavated a Bronze Age sanctuary, with campfire traces and casting-mould fragments in a ditch, and sacrificed sable and fox bones in the pits.

A burial pit held a vessel matching the Petrovka culture, which lay hundreds of kilometres to the south-west, and a mould for casting a Seima-Turbino-type bronze celt.

Palaeogenetics at the Institute of Cytology and Genetics linked the local Baraba population to newcomers from modern Kazakhstan. Molodin says this may be a new culture of the developed Bronze Age — still a hypothesis, with digging resuming in 2027.

#archaeology #bronzeage #siberia

@science

Читать полностью…

Science in telegram

More than 2,000 underwater AI-powered robots armed with lasers have been deployed at salmon farms to combat sea lice.

Developed by Norwegian company Stingray, these devices protect around 50 million salmon and trout from potentially deadly parasites — 24/7.

Traditional delousing methods, which involve hot water or high-pressure water jets, can seriously injure fish. According to Norway’s Fish Health Report, such treatments are a leading cause of death among adult farmed salmon.

Stingray’s robots use cameras and AI to detect parasites on swimming fish, then eliminate them with a green laser from up to 2 meters away — without damaging the fish’s scales.

🐟 AI-powered lasers are now protecting fish from parasites. Welcome to the future of aquaculture.

@science

Читать полностью…

Science in telegram

🧠 AI, Money, and the Communist Paradise That Will Probably Never Happen

I was recently putting together a post for LinkedIn about AI and the future of business. But the more I thought about it, the more uncomfortable the implications became.

Let’s start with something obvious.

Why are companies investing hundreds of billions in artificial intelligence?

To reduce costs, improve efficiency, and ultimately make more money.

Individuals want pretty much the same thing.

We all probably know someone building AI trading agents with Claude, Hermes, or another model. Apparently, some people are even making money with them. Or so I’ve heard.

But what happens when AI becomes significantly better at making money than humans?

🔬 The Age of Autonomous Entrepreneurship

AI is already helping researchers solve difficult mathematical problems, accelerate discoveries in biology, and explore scientific questions that would otherwise take years of human effort.

There’s little reason to assume that identifying profitable business opportunities will remain an exclusively human advantage.

Imagine a future with billions — perhaps hundreds of billions — of autonomous AI agents operating continuously.

Not just answering questions or generating presentations.

Actually building businesses.

Analyzing markets. Discovering underserved niches. Writing software. Launching products. Testing pricing strategies. Acquiring customers. Optimizing operations.

And doing all of this simultaneously, at a speed no human organization could possibly match.

Now let’s take this thought experiment a little further.

⚡ The One-Hour Corporation

Imagine a technology giant operating an enormous army of AI agents.

• After one second: One million business ideas generated and virtually evaluated.

• After one minute: The thousand most promising opportunities identified.

• After ten minutes: A hundred products launched, with some already generating their first revenue.

• After one hour: Ten functioning businesses operating in real markets.

• After ten hours: Several have already been shut down because the agents discovered better opportunities.

Of course, this is a hypothetical scenario, not a technological capability we possess today.

But the underlying idea is fascinating.

What happens to our understanding of entrepreneurship when the lifecycle of a company shrinks from years to hours?

What happens to competitive advantage when discovering a profitable opportunity takes milliseconds — and thousands of competing agents discover exactly the same opportunity?

And what happens to human entrepreneurs when intelligence itself becomes an industrial resource?

🌍 Welcome to the Communist Paradise?

Here’s the paradox.

If everyone has access to AI capable of creating successful businesses, shouldn’t everyone have an equal opportunity to become wealthy?

Unlimited intelligence. Unlimited entrepreneurial potential.

A wonderful technological version of a communist paradise.

(A little hello to Elon Musk and his vision of Universal High Income.)

Except I don’t believe this paradise will ever arrive.

Because access to intelligence and control over intelligence are two very different things.

The companies developing the world’s most powerful AI models will sell increasingly capable systems to millions, perhaps billions, of customers.

But why would they necessarily give everyone the same capabilities they possess internally?

The most advanced models, enormous GPU clusters, specialized infrastructure, and the ability to deploy millions of agents simultaneously could remain concentrated among a relatively small number of companies.

Yes, there’s open source.

Yes, there’s China.

Yes, technological competition will continue.

But even if two companies have access to exactly the same AI model, their economic capabilities may differ enormously.

One can afford ten agents.

Another can afford ten million.

And those ten million agents can continuously experiment, compete, learn, and improve their strategies.

Читать полностью…

Science in telegram

⚛️ Google Just Reduced a Critical Quantum Computing Error by 40 TIMES
Quantum computers have a fundamental problem.
They can potentially perform calculations beyond the reach of classical machines, but the quantum states they rely on are incredibly fragile.
Even tiny disturbances can introduce errors, making large-scale quantum calculations extremely difficult.
Now, researchers at Google Quantum AI have demonstrated a breakthrough that could help overcome one of the biggest obstacles.
They reduced errors in a critical quantum computing resource by approximately 40 times.
In a new study published in Science, researchers used Google's Willow quantum processor to demonstrate a technique called magic state cultivation.
Despite the name, there's nothing supernatural about it.
Magic states are specially prepared quantum states that enable certain operations essential for universal quantum computing.
Producing them reliably has been one of the field's biggest technical challenges.
The researchers developed a method to create these states, detect errors and discard unsuccessful attempts.
The result: quantum states with approximately 99.99% fidelity.
That means the prepared states matched their intended quantum states with extraordinarily high accuracy.
But the breakthrough goes beyond simply reducing errors.
The team also demonstrated that these refined states could be transferred between two different quantum error-correction systems — an important step toward integrating them into larger quantum computers.
And here's why this matters.
Some estimates suggest that producing reliable magic states could consume the overwhelming majority of a future fault-tolerant quantum computer's hardware resources.
Making their production more efficient could dramatically change the economics and engineering requirements of quantum computing.
Important caveat: The 40-fold improvement applies to the error rate of these particular quantum states, not the overall error rate or computing speed of Willow. Only about 8% of preparation attempts passed the researchers' strict checks. This is a hardware proof of principle, not a demonstration of a large-scale, fault-tolerant quantum computer.
Nevertheless, the implications are enormous.
Future quantum computers could help simulate complex molecules, discover new materials and investigate physical systems that conventional supercomputers struggle to model.
For decades, scientists have been trying to make quantum computers powerful enough to perform useful calculations.
Now they're getting better at making quantum information reliable enough to trust.
And that may prove just as important.
Could the biggest breakthrough in quantum computing turn out to be not more qubits — but fewer errors?
#Science #QuantumComputing #Physics #Google #QuantumPhysics #Technology #Innovation
https://doi.org/10.1126/science.aee7961

Читать полностью…

Science in telegram

⚛️ Scientists Have Built the World’s First Nuclear Clocks. They Could Change How We Measure Time.

For decades, the world’s most precise clocks have measured time using electrons moving between quantum energy levels inside atoms.

Now, two independent teams in Vienna and Beijing have demonstrated something physicists have pursued for more than 20 years:

Clocks that measure time using the atomic nucleus itself.

And the difference could be revolutionary.

Atomic nuclei are more than 10,000 times smaller than atoms and are generally much less sensitive to external electromagnetic disturbances.

That makes nuclear transitions exceptionally promising as stable frequency references.

The challenge? Almost every nuclear transition requires enormous energies, far beyond what conventional lasers can provide.

But one isotope is special: thorium-229.

Its nucleus has an unusually low-energy excited state that can be accessed using ultraviolet laser light at approximately 148 nanometers.

Researchers embedded thorium-229 nuclei inside tiny calcium fluoride crystals and developed lasers capable of detecting their nuclear transitions.

Then came the breakthrough.

They used the nuclear transitions themselves to automatically stabilize the laser frequency.

The result: two independently demonstrated, functioning nuclear clocks.

The European system operated continuously for approximately 24 hours, achieving fractional frequency instability approaching 10⁻¹⁵ after a day of averaging.

That’s an extraordinarily small fluctuation — although today’s best optical atomic clocks are still substantially more stable.

But the most exciting application might have nothing to do with telling time.

One team has already used its nuclear clock to search for dark matter.

Some theories predict that ultralight dark matter could cause tiny oscillations in fundamental physical constants, subtly changing how atomic nuclei behave.

By comparing nuclear and atomic clocks, physicists can search for these otherwise invisible effects.

The experiment found no dark matter signal, but it established new constraints on certain theoretical models.

And this is only the beginning.

Future nuclear clocks could potentially improve satellite navigation, measure gravitational effects with extraordinary precision and test whether the fundamental constants of physics really remain constant.

For thousands of years, humanity has built better clocks to understand time.

Now we’re building clocks that might help us understand the universe itself.

#Physics #QuantumPhysics #NuclearPhysics #AtomicClocks #DarkMatter #QuantumTechnology #Science

https://www.nature.com/articles/s41586-026-11084-4

Читать полностью…

Science in telegram

📡 An object pulsed twice per 62-minute cycle, then went silent for five years

In archival MeerKAT data astronomers stumbled on LPT J1555−5631: active for about 15 hours in May 2022, with no pulses or steady emission in five years before or after.

The cycle is 62.2 minutes and holds two pulses. The wider main pulse is partly linearly and partly circularly polarized, with a constant polarization angle; the narrow interpulse is almost 100% circularly polarized. They sit 165.4° apart — 14.6° short of exactly opposite.

That makes it a near-orthogonal rotator: both magnetic poles show during one rotation, the magnetic axis nearly perpendicular to the spin axis. A constant polarization angle does not fit the standard dipolar polar-cap picture: the magnetosphere is probably broader, the emission coherent — an electron cyclotron maser, perhaps.

Neutron star or white dwarf, the data do not settle. Published in ApJ.

Plot: Hyman et al. 2026, arXiv:2609.34102

#astronomy #neutronstars

@science

Читать полностью…

Science in telegram

🪞 Google built a fake internet to test its AI agents

The Chrome team made the "Mirror Dimension": an isolated copy of the web with thousands of synthetic sites — shops, social feeds, banks. The Auto Browse agent navigates them like the real web; no request leaves the sandbox.

The problem: eval awareness — a model notices it is being tested and behaves differently. So the address bar shows the real youtube.com, while underneath headless Chrome runs in a gVisor sandbox and a proxy serves the fake site. To stop the browser flagging the swap, the proxy mints TLS certificates from its own root authority.

Agents build the sites too: one writes the spec, another the HTML and CSS, critics check layout via DevTools. Prompt injections are planted in text, images, ARIA labels and tool responses, payload IDs hidden in an HTTP header.

The pipeline yields over 1,000 tests per engineer per day, 30× the manual rate. Google published no pass rate, so how well the agent resists is unknown.

#AI #security

@science

Читать полностью…

Science in telegram

One gram of single-stranded DNA can hold 455 exabytes. That is a ceiling set by the molecule, not by an experiment.

Four letters, two bits per nucleotide at maximum, so 455 exabytes is 3.6 × 10²¹ bits — and one gram of single-stranded DNA is about 1.8 × 10²¹ nucleotides. The arithmetic closes: this is the density the molecule allows, not something achieved.

What was actually written in 2012 was 5.27 megabits — 660 kilobytes — at 5.5 petabits per cubic millimetre, from off-the-shelf DNA. The distance between that and 455 exabytes is synthesis cost and error rate, not physics.

455 exabytes is 455 million one-terabyte drives. And the figure assumes no error-correction overhead: any real archive carries redundancy and lands well below the ceiling.

Source: Church, Gao & Kosuri, Science 337, 1628 (2012) · doi.org/10.1126/science.1226355

Читать полностью…

Science in telegram

🧬 Rare disease exposes a hidden immune flaw

Researchers at the Garvan Institute of Medical Research have found that in mevalonate kinase deficiency (MKD), a rare genetic disorder, natural killer cells fail to mature and cannot properly destroy infected cells. When these cells miss their targets, they release large amounts of inflammatory signaling molecules, setting off excessive inflammation. The finding, published in Immunity, challenges roughly 30 years of research focused on macrophages.

About half of patients do not respond to existing treatments, and the study points to JAK inhibitors as a possible new option.

The result is based on blood samples from patients plus preclinical models; no clinical trial has yet tested the new approach.

#immunology #genetics #MKD
@science

Читать полностью…

Science in telegram

💧 New material turns water vapor into an ally for carbon capture

Chemists at Oregon State University have created a metal-organic framework called BVR-X that can pull carbon dioxide from wet industrial emissions, where moisture normally undermines capture. Its internal structure keeps water and CO2 in separate regions of the pores, so water no longer blocks the spaces needed for adsorption. A patent application has been filed, and the work is published in Angewandte Chemie.

The material works under conditions closer to real industrial flue gas, where water vapor is abundant, and industry accounts for 30% of US greenhouse gas emissions.

In the US, that 30% figure comes from the Environmental Protection Agency.

#CarbonCapture #MOFs #GreenhouseGases
@science

Читать полностью…

Science in telegram

🌊 SCIENTISTS FOUND 245 ANIMALS LIVING INSIDE A SINGLE ROCK — 5,000 METERS BELOW THE OCEAN.

And these tiny ecosystems may take millions of years to develop.

Researchers from the University of Bremen examined 244 polymetallic nodules — metal-rich rocks scattered across the Pacific seafloor. Inside their microscopic cracks, they discovered 4,139 animals, including one rock sheltering an astonishing 245 individuals.

Even more surprising, these tiny creatures formed distinct communities unlike those living in the surrounding mud. Some of these habitats were remarkably similar across locations separated by 1,000 kilometers.

Why does this matter? The nodules contain valuable nickel, cobalt and copper — exactly the metals targeted by deep-sea mining. Removing them could destroy unique ecosystems that cannot regenerate on human timescales.

The study documents previously overlooked biodiversity, although scientists still don’t know whether some species exist exclusively inside these rocks.

We may be preparing to mine ecosystems we have barely begun to discover.

#Science #Ocean #MarineBiology #Discovery #Biodiversity

https://doi.org/10.1038/s41598-026-76737-w

Читать полностью…

Science in telegram

🔭 Hera reveals the unseen side of the asteroid NASA hit

ESA's Hera spacecraft has sent back its first close-up image of the Didymos system, the site of NASA's 2022 DART impact experiment. The image shows the larger asteroid Didymos from a side never seen in such detail, featuring boulders, craters, and puzzling groove-like patterns. Scientists now want to know whether debris from the collision on Dimorphos traveled the roughly 1 km gap and reshaped its neighbor's surface.

The 580 kg DART spacecraft struck Dimorphos at about 6.1 km/s, cutting its orbit around Didymos short by roughly 33 minutes, far more than expected.

The image was taken from 60 km away in black and white, so surface colors remain unknown.

#asteroids #space #Hera
@science

Читать полностью…

Science in telegram

🍊 Vitamin C may block cancer-linked chemicals in the stomach

Researchers at the University of Waterloo built a mathematical model simulating how nitrates and nitrites move through the salivary glands, stomach, small intestine and plasma. Their simulations suggest vitamin C consumed alongside these compounds interferes with nitrosation, the reaction suspected of producing cancer-causing substances. This may explain why nitrate-rich foods like spinach, which also carry vitamin C, do not show the same risks as processed meats.

The model also predicts that taking vitamin C after each meal could moderately reduce the formation of nitrosation products linked to cancer risk.

The findings come from mathematical simulations, not clinical trials, and real effects on human cancer risk remain unproven.

#vitaminc #cancer #nutrition
@science

Читать полностью…

Science in telegram

🧠 Alzheimer's may shrink two brain regions years before memory fades

Researchers at Johns Hopkins University have found that the entorhinal cortex and amygdala begin losing tissue earlier than the hippocampus, the region usually examined for signs of the disease. By analyzing hundreds of MRI and PET scans, they showed these memory- and emotion-related regions deteriorate first, offering a potential window for early detection.

According to senior author Michael Miller, these brain changes appear as much as a decade before clinical symptoms do.

The work was published in NeuroImage Reports and is being presented at Neuroscience 2026.

The findings come from brain scans, and treatment before symptoms remains a goal rather than a reality.

#Alzheimers #Neuroscience #BrainHealth
@science

Читать полностью…

Science in telegram

🦠 Salmonella programmed to attack tumours, with a gel to hold it back

Researchers at Nanjing University engineered an attenuated Salmonella strain to carry anti-cancer nanobodies and paired it with a gel that keeps it under control, The Innovation reports. Tests were in mice.

Tumours appeal to the bacteria — low oxygen, no gut competitors — but multiplying Salmonella poisons the liver and spleen.

The gelator comes from isosteviol, a compound derived from stevia. It stays liquid in the syringe and sets in under a minute on meeting sodium in the abdominal cavity. Release is mechanical, not timed: at rest almost nothing, under gentle agitation about 80%.

Bacteria-induced weight loss fell by 82% (78.8% for the nanobody strain), livers and spleens stayed normal, and in a two-tumour model the gel helped four of five mice against one of five without it.

Caveats: mice only, it did not help every animal, and human trials are far off.

Photo: Rawpixelcom/Shutterstock/FOTODOM

#cancer #bacteria #nanobody

@science

Читать полностью…

Science in telegram

🧠 Some kids' math struggles may be about strategy, not numbers

A Stanford University study published in JNeurosci suggests that children with developmental dyscalculia struggle not only with calculations but also with choosing how to approach a problem. Led by Oliver Lasnick, the team found that these children took longer to switch methods, were less responsive when difficulty changed, and — unlike other children — did not get better at choosing approaches over time. Brain activity in a particular system also differed between them and children more skilled at using strategies.

The study followed 68 children between the ages of 8 and 10.

The findings show differences in behavior and brain activity, not what causes them.

#neuroscience #dyscalculia #childdevelopment

@science

Читать полностью…

Science in telegram

The future may not be determined by who has access to AI, but by who can afford to deploy intelligence at scale.

🏢 Now Think About Your Own Business

Imagine you’re running a successful technology company.

Not a small startup. A serious organization with 5,000 or 10,000 employees.

You have customers, a platform, intellectual property, revenue, and an established market position.

You compete with other companies of comparable size.

But what happens if one of the AI giants, equipped with dramatically superior computing infrastructure, decides to enter your industry?

What if it can create, test, and scale ten competing businesses faster than your management team can organize a strategy meeting?

Your company may have spent ten years building its competitive advantage.

An AI-powered competitor might attempt to replicate much of that advantage in a fraction of the time.

Of course, real-world businesses have regulatory barriers, relationships, physical infrastructure, trust, and other advantages that cannot simply be generated by an LLM.

And AI companies may find it more profitable to sell intelligence than to compete directly in every industry.

But what if they decide otherwise?

💰 The New Economics of Intelligence

For centuries, economic inequality has been shaped by unequal access to land, capital, education, information, and technology.

Artificial intelligence introduces another possibility: inequality based on access to computational intelligence.

The paradox is almost beautiful.

A technology that promises to democratize knowledge could simultaneously enable an extraordinary concentration of economic power.

And perhaps the most important competitive advantage of the future won’t be your education, your talent, or even the size of your company.

It will be how much intelligence you can afford.

Anyway, as someone once suggested, perhaps it’s time to buy some land, grow corn, raise goats, and start making wine.

At least until AI figures out how to do that better, too. 🍷

🙂 Not a bug. A feature.

Читать полностью…

Science in telegram

Opus 5.5 has taught FPV drones to autonomously hunt down people — without a human operator, internet connection, or GPS.

Using just an onboard camera, the drone builds a heatmap of obstacles, locks onto its target, and can chase them even through dense forests, dodging branches along the way.

We’re heading toward a terrifying future. 😨

🙂 Not a bug. A feature.

Читать полностью…

Science in telegram

🧠 Just 10 Minutes With AI Can Make You Give Up More Easily

AI makes us faster.

But what happens when we suddenly have to think without it?

A newly peer-reviewed study involving 1,222 participants suggests something uncomfortable: after just 10–15 minutes of AI-assisted problem-solving, people performed worse on their own — and were more likely to give up.

Researchers conducted three randomized experiments involving mathematics and reading comprehension. Some participants solved problems independently, while others had access to ChatGPT. Initially, the AI-assisted group performed better. But when researchers unexpectedly removed the chatbot, the advantage disappeared. In the first experiment, participants who had used AI correctly solved just 57% of subsequent problems, compared with 73% among those who had worked independently. They also skipped nearly twice as many questions: 20% versus 11%.

The researchers identified an especially interesting pattern. Participants who used AI primarily to obtain complete answers struggled more afterward than those who used it for hints or explanations. That comparison was observational, not a separate randomized experiment, but it suggests an important distinction: AI that helps you think may be very different from AI that thinks for you.

The possible explanation is a psychological shift. When answers arrive instantly, independent effort begins to feel unnecessarily difficult. Researchers call attention to the loss of productive struggle — the mental work through which we develop skills and persistence.

Important caveat: The experiments measured immediate effects during short online tasks. They do not establish permanent cognitive damage or show that regular AI use necessarily makes people less intelligent. Whether these effects persist over weeks or months remains unknown.

But the implication is fascinating.

We may be building machines that make us extraordinarily productive while quietly changing how willing we are to solve problems ourselves.

Perhaps the smartest AI assistant isn’t the one that always gives you the answer.

It’s the one that sometimes makes you figure it out.

Would you prefer an AI that solves everything instantly — or one that occasionally forces you to think?

#AI #ChatGPT #Psychology #Neuroscience #CognitiveScience #ArtificialIntelligence #Science

https://arxiv.org/abs/2604.04721

Читать полностью…

Science in telegram

🔬 An AI Scientist Just Made a Discovery by Running Its Own Lab Experiments

We’ve seen AI write scientific papers.

We’ve seen it predict proteins and search enormous biological databases.

This is different.

Researchers built a closed-loop AI scientist connected to a physical laboratory. It can generate a hypothesis, design an experiment, turn that experiment into instructions for laboratory automation, analyze the resulting data — and then decide what to investigate next.

The system was given knowledge about Saccharomyces cerevisiae — ordinary baker’s yeast — including roughly 60,000 known biological relationships involving its metabolism, physiology and phenotype.

From these, it generated 1,933 testable hypotheses about how different compounds might affect yeast growth under stress.

Then came the important part:

the hypotheses met reality.

The system selected experiments and controls, converted them into machine-readable laboratory procedures and analyzed the resulting biological data. Some predictions worked.

Others failed.

And one failure produced the most interesting result.

The AI initially predicted that glutamate might protect yeast from formic-acid stress.

The experiment contradicted it.

Instead of simply recording “wrong,” the system analyzed the new metabolomic data, searched for another explanation and identified aminoadipate, a molecule involved in lysine metabolism, as a candidate.

It formulated a new hypothesis.

The lab tested it.

And aminoadipate did improve yeast growth under formic-acid stress — by about 7% for each millimolar increase in the experiment. The researchers report this as a previously unknown protective interaction.

There is an important caveat.

This was not a completely autonomous robot scientist. Humans defined the research domain and safety boundaries, moved some physical samples between instruments and supplied the overall experimental infrastructure. The biological questions were also relatively narrow yeast-metabolism problems — not Nobel-level discoveries.

But something important has happened.

AI has already become very good at generating hypotheses from existing information.

Now the loop can close:

Hypothesis → physical experiment → unexpected result → new hypothesis → new experiment.

That is no longer just AI analyzing science.

It is AI participating in the scientific method.

What happens when systems like this can run 10,000 experiments while a human scientist sleeps?

#AI #Science #Biology #Robotics #Biotechnology #Automation #Research

https://doi.org/10.1098/rsif.2026.0043

Читать полностью…
Subscribe to a channel