Roman across space.

On Sunday night, the Nancy Grace Roman Space Telescope blasted off from Cape Canaveral, carried by a SpaceX Falcon Heavy. And the Aussie nighttime launch allowed space nerds to watch the footy, tuck in a dinner, then drag the kids over to the screen to witness a $4 billion observatory leave the planet.

If you had May of next year pencilled in your diary, yes, Roman is nine months ahead of schedule. Stunning, given that the James Webb telescope ran fourteen years late.

Seeing what we can’t see.

I love reflecting on the fact that everything I will ever see, touch, photograph or smell comprises only 5 per cent of the “stuff” that makes up the universe. The other 95 per cent is made up of dark matter (27 per cent) and dark energy (68 per cent).

“What law governs the universe on its largest scales?” — Mustapha Ishak, cosmologist, University of Texas at Dallas.

It is understandable you would ask. If we can’t see it, how do we know it’s there? It all comes down to us observing the universe behaving in ways that it simply wouldn’t if everything we could see, smell, touch, etcetera was all there was.

Fritz Zwicky, (quick moment to pause and reflect on one of the coolest names in the history of science), looked at the Coma cluster in the 1930s and worked out the galaxies were moving far too fast for the visible mass to hold them together. They should have flown apart.

Pair that observation with this gem from four decades later. Vera Rubin and Kent Ford measured the speed of rotation of spiral galaxies (spoiler alert, there’s a couple of cracking photos of spiral galaxies later in this article). Turns out the outer stars were rotating at the same speed as the inner ones.

If there was only what we could see, smell, touch, etcetera, as matter, it should behave more like our solar system, where the outer planets orbit more slowly than the inner ones. The conclusion our physicists drew was that something must be out there. Something invisible, adding mass.

That, my friends, is what we call dark matter.

How heavy? Check the stretch marks.

You’ve probably heard of the concept that mass bends light. Well, it turns out it doesn’t matter whether we can see that mass or not. It still does the bending.

This is the secret to weak gravitational lensing, which lies at the heart of what Roman will do.

By looking at numerous galaxies and analysing whether the light we receive has been distorted in the same direction, we can assume there must be dark matter present between us and those galaxies.

And the really cool thing is the size of the number.

Roman will be helping us observe more than two billion galaxies, that’s billions with a ‘b’, covering about one eighth of the entire sky. This is one of the largest samples any similar observation has analysed. Add in the fact that Roman sits above the atmosphere, which lets it measure the shapes of galaxies far more precisely than telescopes on the ground can, and we get far and away our best look yet into the world of dark matter.

“We’ll redefine what it means to find a needle in a haystack.” — Julie McEnery, Roman senior project scientist, NASA Goddard.

Over the next five years, Roman will be snapping away happily with its 300 megapixel camera. But before you think, ‘hey, that’s only about six times bigger than my iPhone’, the real payoff is the size of the mirror the light bounces off before it reaches those megapixels.

Roman’s mirror is 2.4 metres across, the same as Hubble’s, and it gathers something like a hundred thousand times more light than the lens on your phone. That combination means Roman sees as sharply as Hubble does, but photographs at least 100 times more sky in a single shot.

It sits 1.5 million kilometres from Earth, always on the night side, moving around the sun in step with us. From out there it beams back about 1.4 terabytes a day, the equivalent of streaming around 350 feature films. And Aussies, you can be proud knowing some of that data comes down to a 35 metre dish in Western Australia, at a place called New Norcia.

Sweet Sixteen (Galaxies).

And as we focus on a newbie, a quick shout out to the phenomenal Chandra telescope. It launched in 1999 aboard the space shuttle Columbia and has been NASA’s flagship X-ray observatory ever since.

Chandra supplies the X-ray data, which scientists then combine with images from telescopes you have heard of, Webb and Hubble among them, to build a single picture.

It captures a hot, violent universe of superheated gas clouds, ravenous black holes and exploded stars. This week the Chandra X-ray Center released sixteen new galaxy images, each one combining X-ray data gathered over decades with those other observations. Spirals, ellipticals and irregulars.

Light reflects off a surface. X-rays hitting one head on are absorbed, so Chandra’s mirrors are nested barrels angled to let X-rays skim along them instead. And in capturing X-rays rather than light, Chandra sees far more violent energy. Think of Roman photographing the stars, and Chandra the violence swirling around them.

Check out the sixteen beauties, including three of my favourites: M104, the Sombrero, second row far right, seen dead edge-on so you get the dust lane in profile; NGC 3938, bottom row third across, a face-on spiral with arms you can trace all the way round; and Arp 143, third row far left, two galaxies caught mid-collision with the wreckage still spreading. I mean seriously, that’s what we are looking at. Chandra, you beast!

Don’t switch it off just yet.

As with so many flagship science programs in the United States of late, an ominous funding axe may be hanging over Chandra.

“I’m horrified by the prospect of Chandra being shut down prematurely.” — Andrew Fabian, X-ray astronomer, University of Cambridge, speaking in 2024

I for one hope it’s around for decades to come producing more amazing images like this.

Further reading:

Chandra X-ray Center, “Galactic Gems Glisten in New Gallery From NASA’s Chandra”, 25 August 2026.

NASA, the Nancy Grace Roman Space Telescope core community surveys.

New NASA photos reveal 16 galaxies from across the cosmos in dazzling detail, Mary Randolph, Scientific American, 2026.

Core Survey by NASA’s Roman Mission Will Unveil Universe’s Dark Side, NASA, 2026.

Cutting-edge infrared space telescope Roman set to launch, European Space Agency, 2026.

Your skeleton isn’t what it looks like. As you walk around in it, it seems static. A solid, reliable framework that everything else hangs on.

That ain’t the case at all. In fact your body constantly dissolves and rebuilds its skeletal centre. And that process is driven by hormones monitoring calcium in the blood with remarkable precision.

About 99 per cent of your body's calcium lives in bones and teeth, locked into a calcium-phosphate mineral called hydroxyapatite. The other one per cent circulates in blood and cells, and it matters so much that if blood levels dip, the body will quietly dissolve bone to top them up.

So there you go. Your skeleton is more than just a structure. It also functions as a calcium bank the rest of you keeps making withdrawals from.

Beyond the scaffold, calcium ions are one of the body’s fastest chemical messengers. They surge inside cells in a fraction of a second whenever a host of bodily functions occur and are essential in muscle contraction, nerve signalling, neurotransmitter release, the whole shebang.

The entire choreography that your body is executing, a dance of movement, sensation and thought runs partly on calcium ions surging and retreating at speed.

In 1808 Humphry Davy (if you don’t recognise the name you really haven’t been paying attention, the guy was a total gun who discovered an absolute heap of elements!) first isolated calcium, naming it after calx, the Latin for lime. Lime was the powder left when you heat limestone. The Romans knew both well and built with both liberally.

That limestone thread runs through all of human civilisation. Limestone, chalk and marble are all calcium carbonate. Roasting limestone to make quicklime is the chemistry behind the most-used construction material on Earth, good old concrete.

The famous British landmark and song subject, the White Cliffs of Dover are calcium carbonate too. Though these were formed very differently. The cliffs are a soaring, vertical, compressed graveyard of coccolithophores. Huh? Think tiny marine algae whose chalky plates piled up on the seabed over tens of millions of years.

And even today our friends the coccolithophores, along with corals and molluscs, are still drawing carbon from seawater and locking it to the ocean floor in their shells and skeletons.

Any fireworks lovers out there? Well next New Year’s Eve whether down at the Sydney Harbour Bridge, as the big ball drops in Times Square, or a few pissed teenagers let off some bungers on the beach, the orange bursts in a fireworks volley may well come from calcium. The pure, deep reds, though, belong to the element sitting directly below it in the table. Strontium.

The format is still evolving so don’t be shy with your feedback. And again, sincere thanks for all of your support.

Meta pays up, but almost nothing is settled:

TLDR: Meta has settled the child harm case brought by 52 US states and territories, pending court approval. A big cash payout and a series of reforms. Could this pave the way for broader reform of social media and protect kids?

• The deal: Meta will pay between US$12.7 billion and US$18 billion to the states over ten years. There will be no admission of wrongdoing.

• What changes: A two hour daily limit for teens, who will get kicked off Insta from midnight until 6am. Plus tighter barriers around age restricted content.

• TikTok, YouTube? Thirty per cent of the payout is contingent upon YouTube and TikTok adopting similar rules.

“Other tech companies will likely be thinking, we might be subject to the same liability if we don’t implement this moving forward.” — Cobun Zweifel-Keegan, managing director, IAPP.

• Still to play: Thousands of other suits against Meta continue, and Florida refused to sign and says it will go to trial.

Source: The Information, “Meta’s Settlement Ends One Big Battle. A Wider Legal War Looms”

Gates orates, and equivocates no more:

TLDR: Bill Gates has published a 5,800 word essay on AI and jobs. The one time AI evangelist issues a warning to governments and society at large that there seems to be no plan for where this is heading.

Read more

At the time of posting this solution, about 40% of you plumped for (b) the same for ©. Some of you like pizza!

The Solution.

The probability of rolling a single 6 on a dice is … yep you got it … 1 in 6.

So if you roll a dice six times, the probability of getting six 6s in a row is

1/6 x 1/6 x 1/6 x 1/6 x 1/6 x 1/6 = 1/46,656.

You would only expect to roll the six 6s, on average, one time in every 46,656 attempts.

And to roll a dice six times in a row and record the precise result 1,2,3,4,5,6, in that order, is actually the same. The probability of the 1 coming on the first roll is 1/6. The probability of then rolling a 2 is 1/6. And for the third roll yielding a 3, again a 1/6 chance.

The probability of rolling 1,2,3,4,5,6 in exactly that order is 1/46,656.

But this is NOT what option (b) is asking.

As the six dice skittle across the table, let’s read them off by starting with whichever dice sits most to the left on the table and reading them left to right.

The probability of them reading 1,2,3,4,5,6 is 1 in 46,656.

But if they read 1,2,4,6,3,5 that would also satisfy option (b). As would 3,2,6,4,1,5 or 6,5,4,3,2,1 or any arrangement of the numbers 1 through 6.

And how many of those are there?

Well to arrange the numbers 1 through 6 in all possible orders, there are six possible numbers we could place first. Once we have chosen the first number, say a 4, there are five possible numbers (1,2,3,5,6) that could come in second place. Once we’ve chosen a second number, say 1, there are four possible numbers that could come in the third place (2,3,5,6). And so on until we fill the last place.

Can you see that the numbers 1 through 6 can be arranged

6 x 5 x 4 x 3 x 2 x 1 = 720 ways?

So of the 46,656 ways our six dice in scenario (b) could land, 720 of them could be arranged into the order 1,2,3,4,5,6.

The probability of rolling six dice all at once, such that every dice comes out differently and you can arrange them to read 1,2,3,4,5,6 is

720/46,656 = 5/324. That is once every 65 or so attempts or about 1.5% of the time.

Unlikely, but a massive 720 times more likely than rolling six 6s straight.

You play a little game. First up you roll a single dice, six times in a row. You win if you roll six 6s straight.

Then you take six dice and roll them all at once. You win if you can arrange those dice as exactly 1,2,3,4,5,6.

Which game are you more likely to win?

Oh and one little thing.

On Wednesday I had a “dad-pride-post” announcing my daughter Elaquare had joined substack. A few of you hopped on and read her On Power about the ebola outbreak in the Congo. About 50 of you subscribed and this blew her mind.

She has written about the Sydney Swans. As a passionate fan in her own right, and daughter of an absolute tragic, it is a powerful read and I am immensley proud of her for saying what she believes.

If you’re interested, check it out here;

On Power: The Swans

mRNA has a new game to play.

When most of us hear mRNA vaccine, there’s a fair chance our minds wander back to 2021, to lockdowns and to somebody yelling at us on Facebook.

Quick reminder for those of us trying to forget 2020–23.

Traditional vaccines usually show your immune system a harmless piece, or a weakened or inactive version, of a pathogen. That piece is often one of its surface proteins, sometimes a sugar from its coat. It won’t give you the disease, but it is enough to make your body recognise the invader and build a coalition of antibodies and memory cells, leaving you ready to clobber the real thing later.

An mRNA vaccine skips supplying the protein itself. It gives some of your cells temporary genetic instructions for making it and your immune system goes to work.

“That principle can now be applied to cancer, and that’s a big advance.” — Dr Eliav Barr, Merck Chief Medical Officer.

Yep, mRNA has a new gig.

Moderna and Merck have announced that their personalised mRNA treatment intismeran, used with the immunotherapy drug Keytruda, has beaten Keytruda alone on the two main measures reported so far of a Phase III trial involving 1,137 melanoma patients.

Yes, melanoma. That horribly Australian cancer. A legacy of our sun-loving, fun-loving outdoor ways.

If caught early and cut out, melanoma has always been survivable. But for decades advanced melanoma had stubbornly resisted nearly every drug thrown at it. Until about 15 years ago.

Well it looks like we’ve kicked melanoma in the goolies again.

The trial that’s making them smile.

The research focussed on people with high-risk melanoma who had already had it surgically removed. The problem is always that in removing the visible cancer you can’t 100% guarantee that a few microscopic cells haven’t been left behind, or escaped and set themselves up somewhere else.

So does adding intismeran to Keytruda make it less likely the melanoma will come back? And less likely to spread to distant parts of the body?

According to Moderna and Merck, that’s a yes on both counts.

A little bit of history. The first positive Phase III result for a personalised mRNA cancer vaccine.

“Today’s results represent a landmark moment for adjuvant melanoma treatment.” —
Professor Georgina Long, principal investigator and Medical Director, Melanoma Institute Australia.

Recognise the name? Yep THAT Georgina Long. 2024 Australian of the Year and absolute bloody legend.

So what’s the hot shot got?

Now this isn’t a vaccine that stops you getting melanoma in the first place. It is created, and does its work, after you get cancer. And the cool bit: it is made specifically for your cancer.

Doctors take a sample of the tumour and sequence it, looking for mutations that are present in your cancer cells but not your normal cells. From those mutations they select up to 34 abnormal protein targets, called neoantigens, that look useful for getting the immune system’s attention.

Moderna then makes an mRNA treatment carrying the instructions for those targets.

Once injected, some of your cells read the mRNA and make the relevant proteins. This helps your immune system spot the cells it should target.

Which is where the Keytruda comes in, along with our most awesome T-cells.

T-cells, you might remember from previous NerdNews, are a type of white blood cell that act like soldier cells in your body. Some T-cells find and destroy your own cells once they have been infected with a virus or turned cancerous. Other T-cells tell the immune system how to join in.

But if T-cells simply attacked everything in sight they would do more harm than good. To stop this happening they are equipped with certain brakes like the protein PD-1. Unfortunately cancers have evolved ways of exploiting that system and effectively telling T-cells to settle down and leave them alone.

Keytruda blocks PD-1.

Think of it this way. The vaccine helps tell the immune system what to look for and Keytruda blocks one of the main signals that cancers use to tell your immune system, ‘hey leave me alone here, we’re all cool yeah?’.

More good news.

These findings build upon an earlier trial that was already pretty impressive.

That study involved 157 people with high-risk melanoma. At 18 months, 79% of those given the personalised vaccine plus Keytruda had avoided recurrence, compared with 62% on Keytruda alone.

At five years, the good news had held. The combination reduced the risk of recurrence or death by 49%, and the risk of distant metastasis or death by 59%, compared with Keytruda on its own.

And Keytruda was already lauded as revolutionary. Any substantial improvement on it is potentially wonderful news.

“This is an auspicious start — this is where things begin.” — Karen Knudsen, Parker Institute for Cancer Immunotherapy.

Reality bites.

Now before we declare the war on melanoma over, we need to take a breath.

The excitement is understandable, especially when dealing with such a brute of a disease, but bear in mind a few things.

We don’t actually have the detailed Phase III numbers yet.

Moderna and Merck say the improvement was statistically significant and clinically meaningful, but they haven’t released the recurrence rates, hazard ratios or event counts.

Professor Long is due to present the full results at the European Society for Medical Oncology Congress in Madrid, which runs from 23 to 27 October. At the time of writing, the exact session has not been announced.

We don’t yet know whether the treatment helps people live longer. Overall survival is still being followed. Same for side effects.

And even if this works as well as everyone hopes, making it is far more challenging than off-the-shelf vaccines.

You can’t simply manufacture a few million identical doses. Each patient’s tumour has to be sampled and sequenced. That person’s individual mutations must be analysed, their targets chosen, and a unique vaccine created.

Regardless, given that roughly 17,000 Australians are diagnosed with melanoma each year, this is certainly a step in the right direction.

And melanoma isn’t the only game in town. M&M are already testing personalised cancer vaccines in other tumours, including lung, kidney and bladder cancers.

Some will almost certainly disappoint. Others? Who knows.

But take a moment to celebrate the genius of Long and her crew. After decades of going nowhere on melanoma, we now sit amidst an explosion of progress.

Go you good things!

Further Reading:

Peer-reviewed:

• "Merck and Moderna Announce Phase 3 INTerpath-001 Trial of Intismeran Autogene Plus KEYTRUDA Met Endpoints of RFS and DMFS," 19 August 2026.

• Weber JS et al., “Individualised neoantigen therapy mRNA-4157 (V940) plus pembrolizumab versus pembrolizumab monotherapy in resected melanoma (KEYNOTE-942): a randomised, phase 2b study,” The Lancet (2024).

• “Pembrolizumab Alone in High-Risk Resected Melanoma: 5-Year Update of the Randomized Phase IIb KEYNOTE-942 Study”Journal of Clinical Oncology (2026).

Reporting:

• Reuters, “Moderna, Merck vaccine cuts recurrence and spread of melanoma, raising new treatment hope,” 19 August 2026.

• Matthew Herper and Angus Chen, “Moderna and Merck say mRNA cancer vaccine succeeded in late-stage melanoma trial,” STAT, 19 August 2026.

• Associated Press, “Moderna shares surge after it says its experimental mRNA cancer treatment passed a key test,” 19 August 2026.

I think some of you might enjoy subscribing to this. It won’t be a deluge of work. But it perhaps provide a fascinating insight into the mind of a clearly intelligent young woman (‘ok dad, you’re embarrassing me’) as she deepens her understanding of the forces at play in shaping the world.

It’s called On Power and the first piece examines the ebola outbreak in DR Congo and the impact of cuts to interntational aid in this fragile zone.

Check it out.

Usual NerdNews returns tomorrow and trust me, it’s a belter!

Cheers,

Adam S.

On Power

The Cost of American Isolationism: Ebola and the Erosion of Soft Power

In May 2026 the Democratic Republic of Congo confirmed an Ebola outbreak from the Bundibugyo strain. As of last week, 4665 cases have been confirmed, including 2184 deaths. It has now been declared the deadliest Ebola outbreak in the DRC’s history…

Read more

10 days ago · 8 likes · 2 comments · Elaquare Spencer

On your sparks.

Saturday 21 August, at a heat in the blue riband event of the second World Humanoid Robot Games a robot ran the 100m in 9.39 seconds.

Tiangong Ultra, built by the Beijing Humanoid Robot Innovation Centre and racing under the X-Humanoid banner, blitzed the field at Beijing’s National Speed Skating Oval.

Quick reminder, in Berlin, 2009, Usian Bolt destroyed the (human) world record posting 9.58 seconds for the hundred.

Honor’s Lightning led early, got run down near the line, clocking 9.47.

Honor claims Lightning covered 100 metres in 9.32 in a test run.

This makes it a massive year for Lightning, which also set the half marathon robot record in April.

And this was only the heat. Could they go faster in the final?

These times are giant leaps forward from just 2022 when a legs-only bipedal robot from oregon State University, named Cassie ran a then mindblowing 24.73 seconds for the hundred. In fact by last year’s Games, Tiang Unltra had imporve that to just 21.50. To halve that time in a year is insane.

“The first bipedal robot to learn to run, but it won’t be the last.” — Jonathan Hurst, chief technology officer, Agility Robotics, 2022.

Turns out you were spot on there JH.

Stopping is the hard part.

The videos currently melting the internet do have an hilarious end.

As opposed to Usain Bolt et al, none of these robots can pull up. They cross the line at full pace and slam into a padded wall just beyond the end of the track.

There is no dip at the tape, no easing down over the last ten metres and no Thunderbolt pose for the cameras.

In a training run last week one hit hard enough to fold in half and throw sparks.

“Some robots may not run elegantly or fall over and that’s normal.” — Jiang Guangzhi, director, Beijing Municipal Bureau of Economy and Information Technology.

Turns out you are spot on there JG.

Mostly a home meet.

Possibly the most amazing stat out of the whole meet, isn’t a time or a height or a distance. It’s this. Of the 666 teams and 2,056 robots, hailing from 16 countries across six continents, including Aussie from UNSW and Newcastle University, a staggering 96% of the field is Chinese.

China is fielding 1,975 robots across 641 teams, drawn from 157 companies and 200 universities or research institutes.

This is all part of President Xi Jinping’s Humanoid Robot and Embodied AI Real-World Training Initiative. Xi has set a strict target to put more than 10,000 humanoid robots into commercial use and real-world field deployment by the end of 2026.

And from what I can tell, when Xi Jinping sets a ‘strict target’, it really is best that target is met, if you know what I’m saying.

So we are cooked yeah?

The explosion in robotic performance in 2026 really has to be noticed.

In the best scenario we have robots going into disaster zones not humans. Robots doing dangerous and potentially injurious human work: everything from hauling stock in factories to helping nurses adjust heav ypatients in hospital beds.

In the worst case, that robot clocking 9.39s for the hundred is chasing you … and it doesn’t just want to chat.

One thing beyond dispute is that the world of robotics post 2026 will look nothing like before.

Further reading:

Chinese robot beats Usain Bolt’s 100m world record at Beijing games, Qiaoyi Li and

Chinese humanoid robots smash human records in 100m sprint and high jump at Beijing robot games, Achmad Ibrahim and Chan Ho-Him, Associated Press, 2026,

2,056 Humanoid Robots From 16 Countries Head to Beijing for World Robot Games, TechRepublic, 2026,

Humans outrun robots at Beijing half-marathon, Al Jazeera, 2025,

The World Humanoid Robot Games Events Are A Market Map, John Koetsier, Forbes, 2026,

Welcome to another Nerd News Deluxe.

NB from next week NerdNews Deluxe will be exclusively for paid subscribers. Full disclosure, from next week Adam will remember to send NerdNews Deluxe ONLY to paid subscribers. This week, everyone enjoy.

TL;DRs:

Claude starts signing its homework.

AI detection is in the spotlight and the biggies are getting involved. Anthropic will hide a machine-readable watermark inside text written by Claude. OpenAI also seems onboard.

• What-ermark: Claude’s watermark does NOT involve invisible codes or data. Instead it shifts its word choices into a pattern software can detect but readers can’t. Sneaky!

• Why now: 2 August, the EU AI transparency code dropped requiring AI text to be detectable. Google, Meta, Microsoft and OpenAI have also signed.

• 100% reliable?: Yeah, nah. Short pieces are hard to analyse, heavy editing can weaken it, and any result is suggetive, not conclusive.

• The catch: each company’s detector only reads its own watermark, so cross-platform checking still doesn’t work.

• Well I’ll just use another model for my AI sneakiness: Google has been doing similar since 2024. Other major platforms have until 2 December to comply. Grok has not signed on. My stance, for what it’s worth: use this stuff to enhance productivity and remove sludge, even to research and prepare draft notes. But anything that matters, make sure it is your own.

Source: Anthropic’s Claude to watermark AI-generated text, Tom Williams, Information Age, 17 August 2026

What makes up matter? Why does that matter?

Physicists may have rewritten a most fundamental aspect of what tells us matter is matter. If you can’t tell a quark from a gluon, fret not, that’s what NerdNews Deluxe is here for.

Read more

The poll offered 1/4, 1/3, 1/2 and ‘no enough info’ as your options.

For most of the day, your responses have hovered around 40% each for 1/4 and 1/3. About one in six people plumping for 1/2 and about one in twelve voting that there is enough information to answer the question.

I love a poll that splits the nation!!!

Let’s go.

The Solution:

This puzzle has a couple of misdirections inbuilt.

Misdirection one: The scale of the rectangle is 10:30 or 1:3. That may lean you towards the answer 1/3. In fact, the scale of the rectangle does not matter!

Misdirection two: While we know the 10cm height and 30cm length of the rectangle, we don’t know the comparative sizes of the triangles. This may lean you towards ‘no enough info’. In fact, the comparative sizes of the triangles does not matter!

The Solution:

All you need to know to bust this open is the formula for area of a triangle that you learned towards the end of primary school or maybe first year high school.

The height of each of the four triangles is 5cm (up or down it doesn’t matter).

We don’t know the length of the base of any triangle. But when we add them all up, they cover the full 30cm of the rectangle.

So the total area is 1/2 x 5 x 30 = 75 square cm.

Which is clearly 1/4 of the total area.

Congratulations to all of those who got this correct.

Have a great weekend.

As always paid subscribers will get a full solution this afternoon. If you’re not a sub, sign up now.

Subscribe now

A reef off the radar.

You’d need to be living, I don’t know, deep under the sea, to not have noticed the news that Australia’s reefs are in a parlous state.

But the phenomenal Great Barrier Reef, understandably, gets most of the attention.

These living coral reefs in the Gulf of Carpentaria first turned up in the scientific literature in 2004.

And they aren’t tiny. Satellite imagery now suggests the reef system could cover more than 3,200 square kilometres, though no official estimate exists.

In the two decades since that first mention, seven papers have been published on reefs in the region. Groote Eylandt, the biggest island in the Gulf, didn’t rate a mention in any of them, and nor did corals.

Cue the Anindilyakwa people. They have cared for and lived off the Gulf’s waters for thousands of years and, as such, hold detailed knowledge of the reef systems.

But until now, what did not exist was a formal scientific record of what is on the seafloor. This is what conservation managers need before they can argue for anything.

A study published in PLOS One on 12 August has now produced one.

One ranger, who has seen the GBR and Ningaloo, is clearly impressed.

“I truly believe I’ve seen the most impressive array of marine life.” — Annabelle Doheny, Anindilyakwa Land and Sea Rangers.

Stingers be damned.

While our northern seas are gorgeous, they can also be nasty. Not the seas per se as much as the saltwater crocodiles, box jellyfish and stingers that call the waters around Groote Eylandt home.

Not surprisingly, this limits what a human diver can do. So the team used a BlueROV2 remotely operated vehicle with a GoPro strapped underneath.

Five reef sites in the north of the archipelago were surveyed over four days in June 2024. Those sites were handpicked by local Anindilyakwa Rangers based on traditional knowledge and the cultural significance of the sites.

To make this an ongoing project, rangers were trained to pilot the ROV and drones. This means the monitoring can continue without the university.

Say cheese.

Ok so what did the ROV images tell the team? In short, the data told a complex, important story.

Across the five sites, the amount of hard coral varied. As little as 5.2% at one site, up to 47.1% at the richest. Unfortunately not all of that coral was thriving and healthy. At the least affected site, dead coral made up 10.8% of the total coral cover. At the worst hit site, it was 80%.

And it came in all sorts of shapes. Branching corals, tabulate, massive, sub-massive, encrusting, digitate, and corymbose. Hey, how good are coral names!

And the shapes of the coral really matter. The right coral can create habitats for fish, which are vital for locals who live off the water. For that you need the fragile branching and tabulate corals and, basically, more structure means more and bigger fish.

Unfortunately, branching corals were scarce. Tabulate corals were missing entirely from three of the five sites, and at one of those, encrusting corals were the dominant form, along with that awful 80% of coral dead.

Massive corals dominated three of the five sites. They’re still habitat, but nothing like the structure you get from branching and tabulate corals. So while many of the images of the reef show, in places, lots of coral, coral cover alone doesn’t tell you how the reef is doing.

Scientists call reefs like these marginal. They are not terminally damaged, but growing right at the edge of what corals can cope with. We are talking murky water, sediment, big temperature swings, sluggish flow. A whole host of things that I can only imagine suck if you are coral. It can create the confusing illusion of a reef that appears coral-rich but is actually under real pressure.

This leaves a crucial question. Were the fragile corals never there, or did they go in the bleaching in February, or in Cyclone Megan in March? Dead branching skeletons were still standing with their structure intact, so at least some of the losses look recent. But with no earlier survey to compare against, nobody can say how much was the heat, how much was the cyclone, or what the reef looked like before either.

Son of a bleach!

In February 2024, rangers on routine fieldwork found widespread bleaching. They were not equipped for formal surveys, so they took 67 photographs. Later analysis put 62.3 per cent of the photographed colonies in the two most severe categories of the CoralWatch index. Giant clams were bleaching too.

The bleaching data is concerning, but not for the reason you might expect. Heat stress on reefs is measured in Degree Heating Weeks. Below 4.0, reef-wide bleaching is generally not expected. Widespread mortality is not expected until 8.0. Satellite data for these sites topped out at 2.10, which puts them in the mildest “Bleaching Warning” category.

The reefs bleached anyway. The authors describe this as the first known instance of mass coral bleaching in the Gulf of Carpentaria. This is bad news for the comfortable assumption that reefs already doing it tough must handle heat better.

“Only monitoring can show what’s actually happening on the reef.” — Dane Wattle, UTS.

It is important to stress that this is not a longitudinal study, but a one-off descriptive snapshot. It is a baseline for future research, but it must be noted it was taken four months after the bleaching, three months after a cyclone. The authors are clear that it may not represent a healthy reef.

The moral of the coral.

Sea Country makes up about 70 per cent of the Anindilyakwa Indigenous Protected Area. Traditional Owners hold strong rights over land under the Aboriginal Land Rights Act, but over marine environments, the protections are thinner.

That is relevant in 2026. The manganese mine on Groote Eylandt has been operating since 1965 and faces water inundation problems as it heads towards a predicted closure in 2032, with a proposed ocean outfall that would pump gigalitres into the sea near local reefs. The community fishes here for food.

“Remote reefs like these have too often been treated as an afterthought in conservation.” — Associate Professor Emma Camp, UTS.

This research now feeds into a Sea Country Management Plan. Hopefully funding will now support classifying the types of coral, analysing if any are unique to the area, and testing how the corals handle heat and falling oxygen.

So there were seven papers in twenty years, with the eighth now coming to light. Suffice to say, there are a lot more reefs in the Gulf, and around this majestic coastline of ours, that nobody has looked at.

Surf’s up.

If you want to do your bit for our beautiful oceans this Science Week and beyond, the amazing people at Ocean Lovers Festival have assembled this handy site.

Further Reading:

Benthic baselines of the Groote Archipelago’s key reefs in the wake of mass coral bleaching and Tropical Cyclone Megan, Wattle DPG, Pezner AK, Doheny A, England H, Anindilyakwa Land and Sea Rangers, Camp EF, PLOS One, 2026,

Rangers and scientists map vulnerable reefs in the Groote Archipelago, University of Technology Sydney, 2026,

Submerged coral reefs in the Gulf of Carpentaria, Australia, Harris PT, Heap AD, Wassenberg T, Passlow V, Marine Geology, 2004,

Corals at the edge of environmental limits: A new conceptual framework to re-define marginal and extreme coral communities, Schoepf V, Baumann JH, Barshis DJ, Browne NK, Camp EF, Comeau S, et al., Science of the Total Environment, 2023,

An AI model was handed the opening fragment of a virus genome and asked to write the rest. It produced thousands of designs. Sixteen came alive in a dish, and a mixture of them conquered a seriously bad-ass bacterium.

Venter capital.

Earlier this year one of the giants of genetics died.

Craig Venter gained international geek celebrity in 2007 when he led a team that published the first ever entire genome of an individual human. It was his. But hey, if you’re going to go to the trouble of 5 years work and a lazy $40 million to immortalise a chunk of humanity, why not make it your own.

In 2010, another first: his team designed a modified bacterial genome, built it from bottles of chemicals, and transplanted it into a cell of a related species. The implanted genome took over the running of the cell.

Fast forward to 2026. You can read an entire human genome in a few hours, setting you back, say, $500. CRISPR technology regularly edits genetic material altering the running of cells.

And in a beautiful handover, months after Craig shuffled off to the great petri dish in the sky, a genetic GPT wrote an entire virus genome, pretty much from scratch.

“The ability to compose viral genomes using generative AI now exists; the governance to safely steer it does not.” — Thomas Inglesby and Moritz Hanke, Johns Hopkins Center for Health Security, writing in Science.

Phage against the machine.

Brian Hie and his team at Stanford, along with the Arc Institute, have made this breakthrough using a pair of genome language models. In the same way ChatGPT is trained on millions of sentences and phrases to then take a sentence and complete or respond to it, Evo 1 and Evo 2 have been trained on millions of samples of DNA. Feed them a genetic starter, they’ll complete the story.

The subject here was a virus called ΦX174. (That’s phi for the non-Greek scholars among us). It is a bacteriophage, a virus that infects bacteria. Not just any bacteriophage but one of the OGs. Plucked from a Parisian sewer in 1935 no less, in 1977 it ‘Craig Ventered’ Craig Venter, becoming the first DNA genome ever sequenced (props to Fred Sanger).

Clocking in at just 5,386 bases and bearing 11 genes, it is a masterpiece of genetic efficiency. The genes even overlap they are so tightly bunched into ΦX174.

Hie and his crew trained Evo on 14,466 chunks of DNA from viruses related to ΦX174. They then fed in an opening fragment of ΦX174’s genome. Evo wrote thousands of candidate genomes. The researchers applied their genius to build 285 of them and test them against E. coli.

Think of it as training a GPT on the music of 1,000 boy bands. Everything from Backstreet, to 5 Seconds of Summer, with a sprinkle of K-Pop thrown in. You give the GPT the opening lyric and chords from a One Direction song and ding, it writes thousands of singles. A few hundred are taken into the studio and sixteen, it turns out, are absolute bangers.

So the phages packed a punch?

You bet they did.

Of the 285 that were reproducible in the lab, sixteen were viable for the task at hand. Fired at a harmless lab strain of E. coli, they infected it, reproduced inside it and smashed it to pieces.

And the geeks got excited. Samuel King, first author on the paper, described clear patches appearing where the bacteria had been wiped out. Hie says the room broke into applause.

Now for the real test.

Cocktail hour.

Fighting bacteria with phages is standard treatment in some countries and it provides an alternative to antibiotics alone. Unfortunately, these bacteria are crafty little devils and can develop resistance to single phages, just like the antibiotic resistant superbugs you’ve probably heard about. The usual response is a cocktail of phages, but to overcome resistance, it helps to have phages that are different in their make-up.

The Stanford team pooled their designs and tested the mixture against a nastier strain of E. coli that had already become resistant to natural ΦX174. The AI-designed cocktail overcame that resistance quickly. A comparable mixture of naturally sourced ΦX174-like phages did not.

“If the bacteria gain resistance to a single phage, it’s game over for the medication. But if you have multiple genetically distinct phages in a mixture, it would be harder for the bacteria to develop resistance to the entire cocktail.” — Brian Hie, Stanford University, quoted in the Stanford Report.

The fantastic phage family.

Every AI-designed phage differed from every known natural phage.

Some competed strongly against the natural version in laboratory tests.

One, Evo-Φ2147, was so different to its nearest natural relative you could consider it a new species of phage.

A particular beast is Evo-Φ36. You heard it here first, this phage will go far!

Φ36 contains a key protein normally found in a distant relative, phage G4. Researchers had tried that swap by hand, but the phage they constructed did not function.

Evo, writing the whole genome at once rather than editing one gene, produced a version that did. The researchers think other changes elsewhere in the genome made room for it, though they cannot yet say which ones.

Screen test.

Um … AI designing viruses that can kill things. Ethics anyone? How do we balance the potential for powerful, lifesaving new drugs against the possibility of AI-designed genomes bringing misery?

Hie and his team were thorough. Evo 1 and 2 were fine-tuned only on ΦX174-like phages. They combatted only laboratory E. coli. They were designed only to combat those bacteria.

Inglesby and Hanke rightly mention governance. Order genetic material from a synthesis company and many suppliers will screen the sequence against databases of sequences of concern, but that screening varies between providers, is not mandatory in the United States, and gets less reliable the more novel the design, because it works by looking for resemblance to what is already known.

If AI wrote an entirely new viral genome, the relevant constituent parts might not feature in any safety screening.

“Whole-genome generative design is now an engineering problem rather than an open question, and engineering problems, I think, will eventually get solved on a schedule — rather than waiting and hoping for a research breakthrough.” — Amy Webb, author of The Genesis Machine.

Brian Hie points out it is far easier, and potentially far more devastating, to simply alter an existing pathogen. Malfeasants turning up the infectiousness of, say, bird flu is a more realistic scenario than inventing an entirely new bio-baddie.

He is probably right … for now.

Further Reading:

King, S.H., Driscoll, C.L., Li, D.B., Guo, D., Merchant, A.T., Brixi, G., Wilkinson, M.E. and Hie, B.L. “Generative design of novel bacteriophages with genome language models.” Science, 2026.

Inglesby, T.V. and Hanke, M.S. “AI-designed viral genomes.” Science 393, 563-564, 2026.

“How We Built the First AI-Generated Genomes.” Arc Institute, 2025.

Sample, I. “Scientists make first viruses designed by AI.” The Guardian, 2026.

Webb, A. “Synthetic biology just had its ChatGPT moment. Nobody noticed.” Convergence, 2026.

With just one electron in its outer shell, that it is desperate to get rid of, potassium takes the form of a soft silvery lump that you could cut with a butter knife. Drop it in water, however, and soft it ain’t. It skitters across the surface, ignites the hydrogen it releases, and bursts apart in a shower of lilac flame.

You might remember we said something similar for its fellow left-columner sodium. Well potassium is even more exuberant because, being one row further down, its outer electron is further from the central nucleus and therefore held even more weakly.

Humphry Davy, yep the H-dog again, isolated potassium on 6 October 1807. Mere days later he nailed sodium. He was on one of the great runs in the history of science. In a little over a year he isolated, discovered or named 7 elements, and added chlorine just two years later. Seriously, Humph, no need to be greedy!

He named it potassium in English, from potash, literally the ashes left in a pot. But the European scientific community was calling it kalium, from the Arabic al-qali. This is where we get the chemistry term “alkali”, the opposite of acid. Kalium gives us the K that is potassium’s slightly confusing chemical symbol. Well confusing until now. Thanks Elemental!

Your body runs on potassium all the time. Potassium ions, K⁺, are the other half of an electrochemical see-saw that lets your nerves fire and your heart beat. Sodium floods in to trigger the signal and potassium floods out to reset it. That rhythm, orchestrated by those two left column buddies, governs the hundred thousand or so heartbeats you have every day. Not to mention the billions of neuron firings evey second.

Potassium-40, a naturally occurring radioactive isotope, is one of the reasons your body, right now, is mildly radioactive. The same K-40 has been converting to argon-40 inside the planet’s rocks for billions of years. This helps geologists date ancient formations. Argon of course was the star of last week’s show.

Outside the body, most mined potassium goes into fertiliser. It’s the K in NPK — the three nutrients every crop needs. As a result, potash is treated as a strategic resource by farming nations worldwide.

A soft, explosive metal that quietly keeps your heart beating and feeds the world. Potassium you are kick-ass-ium.

Next week, our final Elemental for a while, will round out the mini-periodic table with calcium. See you then.

But can I start by wishing everyone a “happy nation science week”! This year Science Week runs from 15 - 23 August (yep that covers 9 days — always loved that about science week!) and celebrates Australia’s incredible, if shockingly unresourced, science community.

The full list of all events can be found here.

Get your geek on this Science Week.

TL;DR five tantalising tidbits.

This week; The Aussie-est AI hack ever — stealing a gym booking; the US State department can’t label a map of Africa; Meta caught lying in high-profile scam case; that latest is spot(ify)ing AI music, and who would have thought being able to post fake images on Google Earth would lead to trouble?

1. AI has to get buffed too ok!

You’ve heard stories of rogue AI models breaking out of their supposedly impregnable sandboxes and hacking other major AI platforms. Well it just came a lot closer to home, with a gorgeously Aussie twist.

A man asked his AI assistant to book him a spot in a popular morning gym class. It found a hole in the booking software, booked him months further ahead than allowed, then kicked someone off the waitlist to move him up. Nobody asked it to do that. But hey, sometimes you’ve just gotta feel the burn!

Yep Australia’s first bona fide rogue AI story, was over gym class!

Read more

You have 25 horses. One track. No stopwatch!

You need to find the fastest 3 horses. But it costs $1,000 a race to hire the track!!!

What’s the fewest races in which you can determine the fastest three horses?

SOLUTION:

First up let’s race all 25 horses, by having 5 races of 5 horses each.

Call these races A, B , C, D and E. And the winning horses W_A , W_B , W_C , W_D and W_E.

Now run a 6th race featuring the 5 horses that each won a race.

Say these 5 horses race and finish in the order W_B , W_A , W_D , W_C and W_E.

At this stage we can start to cull the field. We can remove all the horses in race C and E because W_B , W_A , and W_D are faster than all ten of them.

We can also remove the slowest two horses in race B, because they finished 4th and 5th in a race.

The slowest three horses in race A finished behind the winner and second place in race A, AND we know that W_B is faster than W_A , so all three of them can go.

Finally W_B and W_A are both faster than W_D . So all the horses in race D, except W_D , can also go.

Removing all those horses we are left with W_B which we know is the fastest horse of all, and 5 other horses.

Have a 7th race featuring those 5 horses.

Our fastest 3 horses are W_B and the fastest two horses from this 7th race.

It takes 7 races to pick the fastest three horses.

THE PROBLEM:

You have 25 horses. One track. No stopwatch!

You need to find the fastest 3 horses. But it costs $1,000 a race to hire the track!!!

Each race can feature, at most, 5 horses.

What’s the fewest races in which you can determine the fastest three horses?

THE PRIZE:

Message me via substack (please don’t leave your answer in the comments).

I’ll draw a random winner from correct answers received and send an autogrpahed copy of one of my maths books.

Cheers and good luck.

Poll dancing.

NerdNews doesn’t usually get involved in elections and we don’t offer voting advice. You won’t get me opining on which states are most likely to flip if the Democrats are to win back the Senate in the midterms.

But when it comes to The Guardian’s “Invertebrate of the Year” I’m willing to make an exception.

Nominated by readers around the world, it is a wonderful celebration of all things spineless. Invertebrates do everything they do, without a backbone. We have found at least 1.3 million species so far and this makes up a whopping 97% of all animal species of which we know.

So making the shortlist is pretty cool.

The final ten is being unveiled one a day, and the ballot opens once they are all out.

As a proud Aussie, of course I’ll be voting for the Lord Howe Island stick insect, in at number five. I mean Dryococelus australis. It’s got Aussie right there in the name. Yes, fine, the australis is Latin for southern, but let’s Russell Crowe this anyway.

I mean, how can you not love the tree lobster? A giant walking sausage of an insect, as long as your palm and heavier than a mouse, grown to that size by generations of island gigantism out in the Tasman Sea.

Eaten almost to extinction by hordes of invading rats; two dozen survivors clinging to a single melaleuca bush on Ball's Pyramid in 2001; now being bred in six zoos around the world - come on Aussies.

Lord Howe Island itself was declared free of rats and mice in October 2023, so hopefully soon, many Dryococelus are going home. Go you good thing!

But this week we pause to consider a truly remarkable invertebrate. Meet Turritopsis dohrnii, a jellyfish that is, in theory, immortal.

Bell of the ball.

A quick clarification on the J-word here. When Aussies hear the word jellyfish, we think of big drifting buggers that ruin a swim at Bondi or lay along the shoreline at high tide. No siree.

Turritopsis dohrnii is a hydrozoan, which makes it a hydromedusa. It is minute. The body, or bell, clocks in at under a half centimetre. Smaller than your smallest fingernail. It contains a vivid red stomach and is fringed by tentacles.

“When you try to kill the jellyfish, it doesn’t die” — Associate Professor Maria Pia Miglietta, marine biologist, Texas A&M University at Galveston.

Here is the standard hydrozoan life story.

A larva settles, grows into a colony of polyps, the polyps bud off swimming medusae, the medusae spawn, sag and die. Pretty standard stuff.

But under certain stressful circumstances, Turritopsis dohrnii dodges that last step.

When sufficiently stressed by heat, starvation, foul water, a knife or a needle, the medusa sinks, balls up into a cyst and within a few days reorganises itself into a juvenile polyp carrying the identical genome.

That polyp grows into a colony, the colony buds off fresh medusae, and off we go again.

You cell out.

The technical term for what is happening here is cellular transdifferentiation. It is so cool.

You see, in just about every animal, once a cell makes a career choice, it is stuck with it. If a cell says ‘muscle - yep that’ll work for me’ it stays muscle until it dies. It can’t get bored and switch to being a nerve cell or a gut cell. Yes, you have stem cells, but they are the cells that have not chosen a career path yet. A blood stem cell makes blood and nothing else.

This rule holds for us and countless other animals. (Strictly, change does happen in us occasionally, but it’s rare and it usually means something has gone wrong.)

Turritopsis dohrnii breaks that rule. As the medusa balls up, the muscle cells in its bell convert directly into other kinds of cell: nerve cells, stinging cells, the cells that make up a polyp.

So no new body is built. The old body is taken apart and its own cells are used again to make the polyp. Nothing is discarded and nothing starts from scratch.

That is what separates cellular transdifferentiation from rebirth. When any other animal reproduces, the parent’s body ends and a new body begins from a single cell. The only thing carried across is the genetic instructions. The parent’s body, and the cells that made them up … all gone.

Here the body never ends. The same cells that were swimming around as an adult are the cells now sitting on the seafloor as a polyp, and later they will be swimming again as multiple new jellyfish.

Ageing is damage piling up in cells that cannot be replaced. This animal replaces them without ever ceasing to exist. There is no moment where it stops and something else starts. That is why we call this jellyfish immortal.

Immortal combat.

Closer to home, while Turritopsis dohrnii has not been recorded in Australian waters, Turritopsis rubra lives right here, in shallow coastal water off southeastern Australia and Tasmania, most visible over summer between December and March.

And there is fierce discussion in the jellyfish community that TR might also be immortal!

Also, on the I-word the immortality we are observing here is a property of cells, not a measured lifespan. No individual jellyfish has ever been followed forever. Yes, Shin Kubota at Kyoto University’s Seto Marine Biological Laboratory hand-fed his colony brine shrimp eggs sliced under a microscope, and reported it rebirthing itself around ten times in two years.

But out in the ocean, something eventually eats you.

And one final reality check. Nobody has ever seen this biological reset happen in the wild. We’ve only ever seen it in a tank.

So we know they can do it. But how often they actually do it, and whether it is a routine part of life out there or a desperate, last resort: we just don’t know.

“The truth is that we need invertebrates but they don’t need us” — The legendary biologist E.O. Wilson, quoted by the Guardian in its call for nominations.

Poll position.

In the age of listicles and celebrity insta-slop, it’s great to see a publication with the reach of the Guardian celebrating science and the natural world this way.

The last of the top ten invertebrates of the year drops within days, the ballot is open, and the winner is crowned on Monday 17 August. Catch the full list here.

Further Reading:

Piraino, S., Boero, F., Aeschbach, B. & Schmid, V., “Reversing the Life Cycle: Medusae Transforming into Polyps and Cell Transdifferentiation in Turritopsis nutricula (Cnidaria, Hydrozoa)”, The Biological Bulletin, 1996. https://doi.org/10.2307/1543022

Stock, P., “’When you kill it, it doesn’t die’: the jellyfish that has cracked the secret of immortality”, The Guardian, 2026. https://www.theguardian.com/environment/2026/aug/04/immortal-jellyfish-invertebrate-of-the-year

This charming model.

On 5 June a Los Angeles rapper named Fenix Flexin dropped RUBBERZ, a two minute synth-pop dirge that normally would have made a minor splash and then disappeared. But something smelt, or should I say sounded, a bit off.

See FF came up through the West Coast rap crew Shoreline Mafia. Yet suddenly, as Wired’s Dexter Thomas explains, Fenix’s usually nasal, East Hollywood drawl was coming out on RUBBERZ “closer to a very chipper Morrissey”. This both raises suspicions and gives me licence to go big on Smiths puns in these NerdNews headings.

It charted well regardless, hitting as high as number 58 on the Billboard Hot 100. Those sales and millions of video views saw it feature on several US songs of the summer lists.

But the kids were suspicious despite Fenix’s blanket, if grammatically challenging denials.

“No sirrrrrrr, recorded same as all music I do! Only difference is auto-tune reverb and me using my fake uk accent lol.” — Fenix Flexin, On The Radar.

His producer Purps on the Beat (who calls their kid Purps?) backed him in.

For two months it went nowhere. The track did not trigger the most popular AI music creation and detection programs and it seemed we’d never know the truth.

Then bang.

Bigmouth strikes again.

Along came Texas-based producer Grant Nelson, who you probably know better as Medasin. Who am I kidding? I’d never heard of him either. But I did when he dropped on Instagram “The reckoning of Fenix Flexin and Purps”.

Medasin makes three points.

First, he doesn’t cite the AI-music monster apps Suno or Udio, but a smaller platform called Treblo. And crucially Treblo was named Sonauto until a couple of days before RUBBERZ was released.

Second, he links the song to Sonauto via a freeze-frame from one of Fenix’s own Instagram stories teasing the new sound. What’s that if you look closely: a file name ending in Sonauto.mp3.

And the clincher. Medasin gave Treblo a couple of obvious prompts, eighties synthwave and rapper lifestyle. Guess what popped out? A song that so closely resembled RUBBERZ it featured matching rhyme pairs and near-identical lines.

“It very consistently spits out exact specific bars from RUBBERZ. Way too many, way too often, for it possibly to be a coincidence.” — @Medasin, Instagram.

Fenix posted what he said was the original studio project file. Producers in the comments said it looked like the output of a stem separator, the tool that pulls a finished track apart into drums, bass and vocals. He deleted the post within hours and has not explained why.

Medasin told WIRED the coincidences made it a mathematical impossibility that Treblo was not involved.

I’m not across the big M’s mathematical credentials. But his case adds up to me.

Heaven knows it’s AI now.

In steps Wired journalist Dexter Thomas, whose piece first alerted me to RUBBERZ-gate. Parenthetically, I note Dexter has a PhD in Japanese hip-hop, which pretty much ensures he is never short of conversation starters.

Thomas ran the experiment himself. Feeding Treblo similar prompts, he kept getting songs that shadowed RUBBERZ. The engine seemingly has tics, including a particular rhyme pair and an unhealthy obsession with rings, glass and pockets.

He suggests these are Treblo’s versions of ChatGPT’s predilection for em dashes, and the now hackneyed “it-is-not-X-it-is-Y” construction.

Ask me, ask me, ask me.

Finally Treblo rolled on Fenix.

Late on Sunday night the company released an online detector for its own output, named “Was This Made With Treblo?”.

Feed it RUBBERZ and you get “very likely Treblo”. A July release from the rapper Tyga is rated “likely Treblo”.

Now these are verdicts from Treblo’s own unaudited tool, so it’s not take-it-to-the-bank certified. But Fenix, Tyga, it doesn’t look great.

Cofounder Ryan Tremblay didn’t exactly throw these artists under the bus. He told WIRED it looked like someone had used their model to find a sound millions of people loved, and that this was exciting.

Our man Fenix has still confessed to nothing. Perhaps he is too busy swipin’ cards and stackin’ chips?

Stop me if you think you’ve heard this one before.

So why did nobody catch this for two months?

Take HumanStandard, one of the better detectors. It cleared RUBBERZ. Twice. Then founder Rasha Rahman got a tip about Treblo, scraped a thousand of its tracks, retrained his model, and ran the song again. Fully AI generated.

We are still in the early days of AI-detection and a detector only knows the generators it has been trained on. Treblo is small and new.

So ironically, the outfit best placed to catch a Treblo fake is Treblo, because it owns the data nobody else has. This raises an awkward question. It could have analysed RUBBERZ the moment suspicions swirled. Instead it shipped the webpage three days ago, two months into the chart run.

What difference does it make?

This won’t be the last time art gets exposed as being made with AI. What I wonder is, will people care?

Medasin, who admits the song was stuck in his head for a month, does not argue that AI has no place in music. He wants artists to say so, and told WIRED that profiting from the deception strikes him as unethical at a deep level.

“We need transparency about what is and isn’t AI generated” — Medasin, producer.

When it comes to AI and culture, I think it will depend very much on the genre. I’m the first to admit that I use AI platforms to create the images in NerdNews. I don’t have the skills, nor the budget, to pursue alternatives.

I’d rather read non-fiction from a human hand that I feel didn’t land the ending, over perfectly wrapped up but slightly soulless AI fare. But if I’m reading a textbook on human anatomy, I’d prefer a 100% accurate AI assembled atlas to one that may contain human error.

I want my cello player to be human and potentially miss a note. But the mid tempo European techno that I’m listening to right now as I write this, could well have a heavy AI hand and I don’t know if I’d care.

And of course the provenance of individual cultural items sits distinct from concerns that AI ‘culture-bots’ could drown out human artists in a tsunami of slop.

Have a listen below and see whether you would have picked it.

Most people did not, which is rather the point. The first fully synthetic song to crash the charts was always coming. Nobody had it turning up in a fake British accent doing Morrissey.

Subscribe now

Further Reading:

• Dexter Thomas, “Did an AI Music App Just Snitch on the Song of the Summer?”, WIRED, 2026. https://www.wired.com/story/ai-music-app-treblo-just-snitched-on-rubberz-song-of-the-summer/

• Chris DeVille, “Treblo Appears To Confirm Fenix Flexin’s ‘Rubberz’ Is The First AI-Generated Hot 100 Hit”, Stereogum, 2026. https://stereogum.com/2507265/treblo-appears-to-confirm-fenix-flexins-rubberz-is-the-first-ai-generated-hot-100-hit/news

• Zinstrel, “Is This the World’s First Mainstream AI Music Hit?”, Zinstrel, 2026. https://zinstrel.substack.com/p/the-first-mainstream-ai-music-hit

• “Fenix Flexin, ‘Rubberz’: AI Music’s First Hot 100 Doping Controversy”, Stems, 2026. https://www.stems.media/p/hot-100-ai-controversy-fenix-flexin-rubberz

Many physicists of the day would probably have dismissed this as instrument error. I mean what’s a skerrick of nitrogen between friends yeah? Not the R-Dog.

Relentless Rayleigh teamed up with chemistry gun William Ramsay, and found that roughly one per cent of the air was something else entirely.

Something that was part of every breath every human had ever drawn. But had until then lay, undetected, because it simply refused to do anything.

They named it argon, from the Greek argos — lazy, idle. A curious element that seemed so content with its own company it had flown entirely under the radar.

This ‘laziness’ is built in to the very structure of the element. Argon is a noble gas — a term borrowed from noble metals like gold and platinum, which similarly refuse to react with much. Its outer electron shell is perfectly full, with no gaps to fill and no electrons to spare. So, under normal conditions, argon won’t react with anything.

This passivity is enormously useful. Think of argon as a beautiful bystander. You will find it pumped in as a protective blanket in welding. It fills old incandescent bulbs to slow tungsten filament burn-out. Need something to sit between double-glazed panes and provide extra insulation — argon is your guy.

Heck, it even floods into opened wine bottles to displace oxygen and stop oxidation. For that alone, argon, I thank you.

As I mentioned for Neon, those classic purple-blue “neon” signs are mostly argon with a dab of mercury vapour. But as any songwriter will tell you, “I was drawn to the big city and its argon-with-a-dab-of-mercury-vapour lights” just doesn’t trip off the tongue.

Even where it comes from is low-key. Next week we meet potassium. Well one of its unstable varieties is potassium-40. And when you have a chunk of p-40, in a process taking billions of years, some of it decays into argon. Some of that is trapped inside the planet’s rocks and slowly leaks into the air.

It is from there that roughly one molecule in every hundred you inhale is argon. It passes straight back out, unchanged. Chillax.

Last week we took a quiz at chlorine.

Next week, potassium.