Hi Reader, some science this week made us laugh and think. The 2025 Ig Nobel Prizes were awarded last week at Boston University. Yes, the awards that honour weird, quirky, unexpected research (not to be confused with the rather more prestigious Nobel prizes awarded in Scandinavia next month). This year’s winners include cows painted with zebra stripes to avoid fly bites, a Teflon-filler diet to reduce calories, babies sucking longer on garlic-flavored breastmilk. That same spirit of curiosity runs through this week’s discoveries.

Here’s what we’re covering this week:
​✨ DeepSeek’s Secrets Revealed in New Peer-Reviewed Paper
🤖 Can a Robot Do Emergency Intubation Better Than Human Doctors?
🧬 How inflammation jolts dormant cancer cells awake
​+more

TECHNOLOGY

DeepSeek’s Secrets Revealed in New Peer-Reviewed Paper

For years, the race to build frontier AI seemed like an exclusive club. With training budgets running into the hundreds of millions, only Silicon Valley giants such as OpenAI and Google could afford to play (and U.S. chip export bans seemed to lock China out entirely).

That narrative broke down in January, when a Hangzhou start-up called DeepSeek claimed in a preprint that it had trained a reasoning-focused AI model, R1, for just US$294,000 (after pretraining a base model at a cost of about US$6 million). The announcement rattled markets but left many asking: was it real?

This week brings the answer. Nature published the peer-reviewed account of R1’s creation, the first time a frontier model has undergone such scientific vetting. The paper shows that DeepSeek chiefly used reinforcement learning, with minimal direct reasoning supervision (augmented in later stages by other techniques), rewarding only for correct final answers. From this minimal signal, the model developed its own strategies, self-checking, reflecting mid-calculation, even showing “aha moments.” On the American Invitational Mathematics Examination benchmark, accuracy rose from 15.6% to 77.9%, and reached 86.7% with self-consistency decoding, outperforming the human average.

By making R1 open-weight and peer-reviewed, DeepSeek has set a transparency precedent in a secretive field. The DeepSeek-R1 code has already been downloaded over 10 million times, and the paper suggests that billion-dollar budgets may no longer be the entry fee for frontier AI, a shift with potentially large economic and geopolitical implications.

TL;DR: Once thought impossible without billion-dollar budgets, DeepSeek’s R1 AI has now been peer-reviewed in Nature, confirming how a Chinese start-up trained a reasoning model on restricted chips for just US$300,000 in fine-tuning costs.

BIOLOGY

Why Cancer Comes Back Years Later

Even when a primary breast tumor is successfully treated, cancer isn’t always gone. Stray cells may have already slipped into distant organs, where they go quiet. These “sleeping,” or dormant, cancer cells don’t divide, making them invisible to scans and resistant to drugs. For months or even years, they can remain dormant before suddenly reawakening and sparking relapse. That hidden danger explains why metastasis causes most cancer deaths.

Now researchers have uncovered one of the triggers that can jolt these dormant cells back to life: inflammation. In a study published in PNAS, researchers showed that when lung tissue in mice became inflamed, dormant breast cancer cells switched back on. The team found that the chemotherapy drug bleomycin, which scars lung tissue, inadvertently woke the cells. Once reawakened, the cancerous cells don’t return to dormancy. They remain active and multiplying, even after the inflammation fades.

The discovery exposes a troubling paradox: cancer treatments that inflame healthy tissue might also help metastasis along. But it also points to a new strategy: targeting the inflammatory signals that rouse dormant cancer cells could help keep them permanently in check.

TL;DR: Inflammation can wake dormant cancer cells - even after treatment - helping explain why relapse happens and offering a new target to prevent metastasis.

ROBOTICS

Can a Robot Do Emergency Intubation Better Than Humans?

Intubation, placing a breathing tube into the trachea, is one of the most critical emergency procedures, but even trained professionals can struggle, especially under pressure. Now, researchers have built a soft robot that, in simulator tests, can do it faster and more reliably than humans.

The device, described in Science Translational Medicine, is made of flexible materials that allow it to “squirm” its way down the throat, steering into the trachea without rigid pushing. In high-fidelity simulation trials, the robot outperformed even extensively trained medical personnel, consistently placing endotracheal tubes in less time and with fewer errors.

Autonomous intubation could one day change emergency medicine in ambulances, military field units, and disaster zones, where expert anesthesiologists are scarce. By offloading one of the riskiest and most stressful procedures to a robot, medics might save lives with greater confidence. Still, questions remain about how much autonomy should be granted to machines in critical care and how results in manikins will translate to human patients.

TL;DR: A soft, flexible robot can autonomously intubate in simulator tests, placing breathing tubes more quickly and reliably than trained clinicians.

In Other News

Salt + Ice = Electricity? Researchers discovered that adding salt to ice strengthens its flexoelectric response. This means that salted ice, unlike pure ice, can produce an electric current when it’s bent - hinting at a novel, if very low-power, way to harvest energy in cold places.

Thousands of worlds beyond our solar system. NASA has confirmed more than 6,000 exoplanets (planets that orbit stars outside our solar system). This shifts astronomy’s focus from counting planets to probing how they form, evolve, and maybe even support life.

Could lab-grown brains one day feel pain? Lab-grown brains are becoming more complex, and some researchers fear they could one day sense pain, forcing a rethink of research ethics.

Primordial black hole may have spewed the highest-energy neutrino. Physicists propose that the most energetic neutrino ever detected might have come from a primordial black hole’s final explosion which, if true, could be the first evidence of Hawking radiation and a clue to dark matter.

Scientists clash over “mirror life” research. At a scientific meeting in Manchester last week, scientists debated whether to restrict research into mirror versions of DNA and proteins that might one day enable synthetic cells. Such work could yield new drugs and, if uncontrolled, unleash organisms our bodies couldn’t fight.

25 years later, still no dark matter. Two new experiments have refuted DAMA’s decades-old claim of detecting dark matter, closing a long-standing controversy and clearing the way for new searches.

Turning windows into solar panels. Chinese researchers built a transparent coating that turns ordinary windows into solar panels, offering a scalable path to clean energy if efficiency hurdles can be overcome.

This Week in History

This week is notable for the anniversaries of the passing of several pioneering scientists and mathematicians:

Girolamo Cardano (d. 21 Sep 1576) Cardano was a Renaissance polymath who solved cubic and quartic equations, invented the combination lock, and described the universal joint still used in cars. He also gave the first clinical description of typhus and published encyclopedias of natural science. Sadly, he ended his extraordinary life by suicide at 74.

Sigmund Freud (d. 23 Sep 1939) Freud laid the foundations of modern psychotherapy by proposing that the mind is structured into the id, ego, and superego, with much of human thought driven by unconscious processes. His method of psychoanalysis, using free association, transference, and talk therapy, established the basic framework for treating mental illness that still underpins clinical practice today, even as his specific theories remain debated.

Paul Erdős (d. 20 Sep 1996) One of the most prolific mathematicians in history, Erdős published hundreds of papers in number theory and combinatorics. He proved at age 20 that there’s always a prime between any number and its double, a deceptively simple, beautiful result. Famous for wandering the globe with “proofs in his pocket,” he collaborated so widely that mathematicians still track their “Erdős number” to measure their distance from him.

This Week’s Puzzle

🧩

Three Friends
​

Three friends (A, B, and C) play ping-pong with this rule: two play at a time, the winner stays on, the loser steps aside and the third person plays.

At the end of the day each has played:

• A played 10 games
  ​
  ​
• B played 15 games
  ​
  ​
• C played 17 games
  ​
  ​

Question: Who lost the second game?

​
​

Until next time,

The Ve Team 👋

Know someone who would enjoy this newsletter? Forward it their way.

Research by Pablo
Written by Ines

​Unsubscribe · Preferences​

🧩

​
Solution

​

If you add up all the games each person played, you get: 10 + 15 + 17 = 42 participations. But in each match there are two players, so divide by 2 → 21 total matches.

If someone plays every game (always wins or always loses), they'd have 21 games. But someone who always loses and waits after each loss can only play in every other match:

If they started in Game 1, they'd be in Games 1,3,5,… up to 21 → that’s 11 matches.

If they sat out Game 1 and only played Games 2,4,6,…,20 → that’s 10 matches.

A played 10 games, which matches the pattern of playing only every even-numbered game (i.e. stepping in for Games 2,4,6,…,20). That matches someone who never wins a game, just keeps playing in those even turns, then sits out others.

Because A played only the even games and lost every one of them, A lost the second game.