There are some stories that stay with you forever, the big scientific moments that change the course of history. These are not those moments – but we can’t stop thinking about them anyway.

Push my button

Surgeon Stuart Meloy was performing a pain-relief operation on a woman’s spine when she started exclaiming emphatically. “You’re going to have to teach my husband to do that,” the woman said. This inspired Meloy to invent an implant to help women who are unable to achieve orgasms in the usual way. New Scientist broke the news in 2001, sparking a frenzy of worldwide media excitement. But 13 years later, Meloy was still struggling to raise funds for a full clinical trial, and the “orgasmatron” never reached a climax.

Monkey pleasure

We aren’t the only primates capable of coming up with creative ways to pleasure ourselves. In a study of long-tailed macaques (Macaca fascicularis) in Bali, Indonesia, at the University of Lethbridge in Alberta, Canada, and her colleagues filmed hundreds of examples of male and female monkeys using stone tools as sex toys, rubbing or tapping them on their genitals. According to at the University of Western Australia, the Balinese monkeys get plenty of food handouts from humans, and this frees up time for sexual experimentation. But some monkeys in other places have unusual kinks, too: another study reported a wild macaque in Japan rubbing its genitals on the back of a sika deer.

Head transplants

In 2015, surgeon Sergio Canavero announced plans to carry out the first human head transplant, intended to extend the lives of people with muscle and nerve degeneration or widespread cancer. He claimed his technique could overcome technical hurdles, including how to fuse the spinal cord and prevent the body’s immune system from rejecting the head. In the following two years, Canavero published animal experiments supposedly backing up his methods, but other surgeons remained highly sceptical. A human head transplant never did happen, but Canavero has reportedly moved on to an even bolder ambition:

Rats driving

Have you ever wondered if a rat can drive a car? Kelly Lambert at the University of Richmond in Virginia has. She and her colleagues built a tiny car with three copper bars that would steer it left, forward or right. Sure enough, 18 rats learned to drive towards food rewards. They even found it relaxing, as measurements of their stress hormones indicated. Lambert proposed that driving challenges could replace traditional maze tests when using rat models to study neuropsychiatric conditions.

The sniff test

Humans are more similar to rats than we like to think. In a psychology experiment, and his colleagues at the Weizmann Institute of Science in Rehovot, Israel, surreptitiously filmed 153 volunteers. After shaking hands with someone of the same sex, people smelled their shaking hand more than twice as much as they did before the handshake. Puzzlingly, after shaking hands with someone of the opposite sex, people were more likely to sniff their non-shaking hand. The team suspected that humans use chemical signals all the time without being aware of it.

Urine fountains

Communication through scent is, of course, widespread in mammals, but some never-before-seen behaviour in dolphins took biologists by surprise in 2025. While studying Amazon river dolphins (Inia geoffrensis), or botos, in Brazil, Claryana Araújo-Wang at CetAsia Research Group in Ontario, Canada, saw a male “flip his belly up out of the water, expose his penis and then proceed to urinate into the air”, she told New Scientist. They saw dolphins do this a total of 36 times, and most of the time another male was nearby and would pursue the stream of urine. The purpose of this behaviour hasn’t yet been established, but hormones in urine might signal a male’s social position or physical condition, said Araújo-Wang.

Giving the finger

As for human modes of communication, many of us struggle to put our smartphones down – so the last thing we’d want is for them to crawl across the table towards us. Nevertheless, Marc Teyssier at the University of Paris-Saclay in France and his colleagues gave phones this ability in 2018, with a robotic appendage dubbed the MobiLimb. It plugs into a device’s USB port and, for some reason, looks eerily like a human finger, giving it a spooky quality reminiscent of Thing from The Addams Family. The MobiLimb can stroke your wrist, wave, wag like a tail or, most disconcertingly, drag the phone around. Thankfully, the idea hasn’t taken off.

Horror frog

If you find the thought of a phone with a finger horrifying, wait until you hear about the nightmare-inducing hairy frog Trichobatrachus robustus. In 2008, researchers reported that these frogs can break their own bones to produce claws that burst through their skin. at Harvard University’s thought these weapons were for defence, but two years later, that they were used in fights between males – and that the frog’s unusual hairs were a kind of armour to protect its vulnerable parts from sharp claws.

Corpse shrinking

In 1957, when New Scientist was 1 year old, we reported on a new method for shrinking an adult human body to the size of a baby. At an exhibition of inventors in Winnipeg, Canada, the unnamed inventor of this process claimed it would make disposing of dead bodies simpler, as they could be slid into underground tubes and take up less space. The method involved applying high pressure to the corpse and squeezing out fluids so its flesh could be compressed into a mould. “All this is, of course, highly ingenious but it is uncertain why the inventor thinks that his newly proposed form of burial is in any way better than the current and very popular method of cremation,” the writer mused.

Transient anus

We may take it for granted, but the anus, which allows food to pass through the body, was an evolutionary breakthrough, both figuratively and literally. The earliest animals, like jellyfish today, had only one gut opening, where food went in and waste came out. In 2019, scientists discovered that the warty comb jelly (Mnemiopsis leidyi, pictured top) forms an anal opening when it needs to defecate, but it vanishes straight afterwards. This, they suggested, could represent how the anus evolved in the first place.

Fart science

Flatus – gas expelled from the anus – has been a topic of endless fascination for many readers (and journalists). In 2001, we featured the work of Michael Levitt at the Veterans Administration Medical Center in Minneapolis, a leading authority on the subject. He developed a “Toot Trapper” that reduces odours by trapping certain gases using activated charcoal. We also reported on a device for monitoring gas emissions from dogs, used to test dog snacks that were supposed to reduce flatulence. In 2016, researchers in Australia worked out which foods cause the smelliest farts by mixing faeces with different foods. They recommended eating less protein and more slow-release carbohydrates to minimise hydrogen sulphide production. In 2022, at the Mayo Clinic in Rochester, Minnesota, and his colleagues worked out that some poos float because of gas from gut microbes, confirming a hypothesis first put forward by Levitt 50 years earlier.

Brew your own

Even stranger things go on inside some people’s intestines, as we found out in 2019. A 46-year-old man started experiencing unexplained mental fogginess, dizziness and memory loss, and had to give up his job. Later, doctors found high levels of alcohol in his blood when he hadn’t been drinking. Eventually, they realised his gut contained high levels of Saccharomyces cerevisiae – brewer’s yeast – which got him drunk any time he ate a lot of carbohydrates. This rare condition, known as auto brewery syndrome, had been triggered when the man took antibiotics for a thumb injury, which disrupted his gut microbiome. This year, a study reported that another man with this condition fully recovered after getting a transplant of faeces from a healthy donor. Both articles were written by Alice Klein, who we also have to thank for the monkey sex toys, car-driving rats, smelly farts and floating poos. Thanks, Alice!

The rise of artificial intelligence has pushed mathematics to a crisis point. In our latest video, we speak with Terence Tao, one of the world’s top mathematicians and author of Six Maths Essentials, to diagnose the state of the field right now and explore if it has a future even in the face of increasing automation.

Why should anyone learn to do maths with a pen and a piece of paper right now? What is the power of abstract thinking and can we afford to lose it? And what is on the research wish list of one of the most prolific and decorated mathematicians of our time? Join us as we take a grand tour of modern mathematics, following one of its best guides through the rapidly changing science. We’ll explore the fundamental character of mathematics research, as well as the principles and questions that guide the field – and how AI can be in discord with them.

Why prime numbers might not be random after all

The Riemann hypothesis may hold the answer to the distribution of the primes, the only problem is, no-one has yet been able to prove it

I was sitting in a conference in August, drifting as I so often do between listening to what was on stage and pretending to be doing important work on my laptop (by which I mean sending memes on WhatsApp).

All of a sudden, from the stage, the renowned biologist came out with a sentence that made every single person in the room sit up and pay attention: “Of course,” she said, “we now know that ��������𾱳����’s starts in the nose.”

I’m sorry, what? Starts in the nose? I didn’t know ��������𾱳����’s starts in the nose.

Free newsletter

Sign up to Professor Hannah Fry’s column

Sign up to newsletter

My own lovely dad died from ��������𾱳����’s last year, and ever since his diagnosis, I’ve done an incredible amount of digging into the disease, trying to understand what was happening to him. I was astonished I’d never come across this idea.

And perhaps you haven’t either. Which is why I want to share my astonishment at discovering that Aviv Regev was right – not only is ��������𾱳����’s closely associated with our sense of smell, but there is a growing body of research suggesting that our noses might have something to do with why the disease develops in the first place.

The first bit you might already know: people with ��������𾱳����’s tend to lose their sense of smell, sometimes years before any symptoms of memory loss start to show up. But it turns out that our ability to smell is a surprisingly good indicator of our overall health (looking at you, covid-19). One study from 2014 found that of whether they would be dead within five years than a cancer diagnosis.

It’s not that losing your sense of smell itself will kill you. More that the olfactory nerves are deeply intertwined with some of the most important memory structures in the brain, so when they start to deteriorate, it can be an early warning that something much more serious is going on.

And this connection seems to work in both directions. One intriguing study (admittedly with quite a small sample size) , thus exposing them to all kinds of wonderful scents while they slept. After six months, the people who had the olfactory enrichment performed better on memory tests than those in a control group.

All very interesting. But not what Regev had been suggesting. Her words seemed to imply that noses are actually responsible for the trouble in the first place.

It turns out that the reason scientists are even entertaining this possibility is because of how fiercely our brains are protected from the outside world. Our skulls are like an actual fortress: even our blood isn’t allowed in without passing a particularly fussy bouncer called the blood-brain barrier.

Can we harness the brain’s deep-cleaning system to prevent ��������𾱳����’s?

The glymphatic system is the brain's recently discovered network for flushing out toxins. Now, we're uncovering surprising ways to boost it to protect brain health

One of the only real exceptions is a little bit of bone right at the top of your nose that is riddled with holes. It’s called the cribriform, and it’s where the olfactory fibres dangle down into your nose to give you your sense of smell – a bit like if you pushed some Play-Doh through a sieve.

And that means that, despite all the elaborate defences protecting your brain, there is a potential route in, sitting right in the middle of your face.

So, here’s the idea. Maybe the beta-amyloid plaques, the sticky little protein fragments that appear to play such a key role in the cascade of ��������𾱳����’s in the brain, are what’s left over after the brain has fought an infection. And maybe the nose is how the infection arrives.

To see if this held any water, in 2022, a group of scientists of some mice.

(I should add, for the sake of the mice’s reputation, that it was actually Chlamydia pneumoniae, a cousin of the bacterium that causes the STI. But I hope you’ll forgive me the less scientifically accurate version. In a long career, you get very few chances to write sentences that good, and I wasn’t about to waste mine.)

The remarkable thing, though, was that within 72 hours, the bacteria had travelled up the nerves in the nasal cavity and got through to the brain. What’s more, the infection triggered the accumulation of beta-amyloid. Now that’s not definitive proof of how ��������𾱳����’s starts in humans. But there is some other, rather unsettling evidence that infections aren’t the only unwelcome visitors that manage to get in that way.

A group in Mexico . Inside the dogs’ brains, they found microscopic particles of metals – that seem to have entered through the nose – wrapped in pockets of inflammation and that same beta-amyloid protein. You don’t see the same thing in dogs that lived in cleaner, more rural environments.

Now these studies have been extended to who died from gun violence or car accidents – and they find something similar. Tiny, metal–rich particles, thought to originate from exhausts and brake dust, turned up in their brain tissue, alongside inflammation and accumulations of beta-amyloid. The very sorts of changes associated with ��������𾱳����’s, appearing in brains that were sometimes only a few years old. Put together, it implies that the disease might have an environmental cause, one that begins much much earlier than we thought.

��������𾱳����’s, stroke, depression: The preventative power of sauna

Sustained heat stress is bad for our health and can be deadly. But we’re discovering that heat therapies like sauna, when used in the right way, have surprisingly wide-reaching benefits for health

So, where does ��������𾱳����’s start? The answer is no one nose (pun very much intended). There is still a great deal we don’t understand, but the implications from all this research are rather extraordinary. What if the beta-amyloid plaques we’ve spent so long trying to get rid of are, at least sometimes, the aftermath of the brain trying to protect itself? And what if, rather than focusing entirely on clearing up the damage, we could work out what set it off in the first place?

But more than anything, I can’t help but laugh as I think about all this in the context of my dad, who had many wonderful qualities, but a discreet nose wasn’t one of them. We used to say that his nose would enter a room 15 minutes before he did. He would have found it very funny indeed if it turned out to have been the cause of his downfall.

Hannah Fry: How OpenAI has caused an existential crisis in mathematics

OpenAI has been throwing everything at solving long-standing maths problems. This should have been a glorious moment for mathematicians – but instead we're having a massive collective wobble, says New Scientist's new columnist, Professor Hannah Fry

ChatGPT was never meant to start a revolution. When it launched in November 2022, OpenAI CEO Sam Altman called it “an early demo” and expectations were low. Personally, my initial reaction was that it seemed no more capable than previous large language models (LLMs) OpenAI had released; it just had a nicer interface. Obviously, I was wrong to dismiss the importance of the interface, but in 2026 I think most people who use AI are making a similar mistake without even realising it. What you think of as ChatGPT or Claude or any other chatbot from a big AI brand is actually many pieces of software wrapped into one, and we are ignoring the most important: the harness.

To understand what a harness is and how it works, it helps to break down the multiple things we call ChatGPT. At the very bottom is the LLM, currently a version of GPT-6, but this gets switched out when OpenAI releases new model families. LLMs do one thing: given a string of text, they predict the most likely next characters. OpenAI’s first model, GPT-1, was released in 2018 as a piece of software that you had to download and run on your computer through a terminal, the text-based interface to a computer that most of us avoid. The same was true of GPT-2 in 2019, until an independent software developer called Adam King created Talk to Transformer, a now-defunct website that allowed you to enter a prompt and get a GPT-2 response in return. Back-and-forth chat wasn’t possible, but this was an early example of an interface, in this case a website, sitting on top of a model.

In 2020, OpenAI released GPT-3 and, for the first time, allowed people to access it through an application programming interface (API), its first commercial product. APIs are a standard way for one computer to communicate with another – for example, Google has a Google Maps API that lets a piece of software send a request for location data. OpenAI’s API let anyone create software that could plug in to GPT-3. At the time, I played with a few of these and found them interesting but clunky – the text they produced was often incorrect or flawed in some way, though it was cool that this level of text generation was even possible. This was essentially still a time of “raw” LLM output, with no guard rails – if you asked GPT-3 to .

That changed with the release of GPT-3.5 and, at the same time, ChatGPT, which was first called “Chat with GPT-3.5” until a last-minute name change. That original name reveals the way OpenAI saw ChatGPT, as simply an interface to a model, and this explains my reaction as well – the results I got from playing with ChatGPT seemed to be similar to GPT-3, just easier to use.

Actually, there was a crucial difference. With GPT-3.5, OpenAI had attempted to install guard rails that would prevent the model from producing text it deemed harmful, with software that would screen inputs and outputs. These didn’t work very well – you could get around them by , for example – but this was the beginning of a move away from raw LLMs to ones mediated by a harness sitting between the model and the interface.

Today, the ChatGPT product, whether that’s the app on your phone or the website, has an extensive harness that is mostly hidden from the user. It allows the model to do things like search the internet, read documents, or “reason”, meaning to continuously prompt itself through a “chain of thought” to produce better results. OpenAI also has another, more complex harness for its models, called Codex, which allows the LLM to more directly control a computer and so is particularly useful for writing computer code.

Why does any of this matter? Well, the dirty secret of the big AI firms is that their best models are, to some extent, interchangeable. The technology behind an OpenAI GPT model, a generative pre-trained transformer, is the same as the one used by Anthropic, Google or any other AI company. There are important differences in the data and computing power used to train these models, but generally they are all headed in the same direction, which is why the companies are constantly racing to claim the crown of “best” model before another one takes its place.

What really makes the difference is the harness. To illustrate this, I gave the same prompt – “Who am I?” – to the same model, GPT-6, in three different ways. Via API, I got a response beginning: “I don’t know your identify from this chat.” From chatgpt.com, without being logged in, I got a response that started “That’s a big question. 🙂”. When logged in to my personal account, the response began “You’re Jacob Aron, a senior editor at New Scientist, with a strong interest in mathematics, science, technology and the limits of what we can actually know.”

But the harness matters much more than simply knowing a few facts about the user. In the ARC-AGI test, considered to be one of the leading measures of model “intelligence”, models are tested using either a . The difference is remarkable – for GPT-6 Astra, OpenAI’s leading public model, when using the provider adapter harness rather than the standard one. For an earlier model, GPT-5.6 Sol, OpenAI found that .

If harnesses are so important, why don’t we talk about them more often? Why don’t AI firms shout about their new harness, rather than just their latest model? I think there are a few reasons. The first is simply that they are complicated and technical – I have taken a whole article to explain them – and your average chatbot user doesn’t really need to know about it. But the second is that they are, in a way, boring. The cost of training a leading LLM is approaching $1 billion, a Herculean effort that requires number crunching across a vast army of computer processors. Once LLMs are trained, we still don’t really know exactly how they produce the results that they do, providing an air of mystery. Harnesses, on the other hand, are just ordinary software, lines of code written by humans that anyone can read.

But here is where it gets really interesting. If a harness is just code, can an AI write and improve its own harness? That is currently an open research question, both inside the AI firms and for academics. For example, in a paper , one research team tested models by identifying repeated failures and proposing tweaks to their harnesses that would address the failures, seeing improvements of around 10 to 40 percentage points on benchmarks. This is why I think we are going to be hearing much more about harnesses in the near future. If the goal of the AI firms is recursive self-improvement – an AI that can build a better AI – then the harness, not the model, could be the key to achieving it.

Further AI reading:

• OpenAI appears to have overturned a big assumption in computer science, as part of a release that saw new results in maths and physics as well. Interestingly, it is taking a slightly different approach for each of them . For the maths result, related to a problem about spinning a needle, OpenAI has released the AI output in full. It is doing the same for the computer science result, a refutation of a conjecture called the exponential time hypothesis, while also funding a conference to discuss it. The physics result is being held back until it is reviewed by external experts. It seems like OpenAI is still trying to figure out what to do about its models stepping on academics’ toes.

• Anthropic has ahead of a planned initial public offering (IPO) on the stock market later this year. It calls AI “the most consequential technology in human history” (no surprises there) while also warning that it may pose “catastrophic or existential risks to humanity” – again, not particularly surprising given Anthropic’s public record, but it does raise eyebrows to see this language in documents filed with the US Securities and Exchange Commission.