FlashFeed

Full text

Here's an embarrassing thing to admit on the internet: when I was a teenage subscriber to Nintendo Power magazine, I picked the Donkey Kong 64 soundtrack as my "sign-up gift" because I wanted to be able to listen to the DK Rap on CD . Why I was obsessed with the song I can't say 25 years later, but Donkey Kong 64 is a memorable mess of a game—sprawling, ambitious, funny, inventive, and often infuriating, with a predilection for collectibles so infamous that the developers made fun of themselves for it a few years later.

The point is, Donkey Kong 64 is a game that inspires strong feelings in those who've played it. So it seems fitting that members of the DK64 randomizer community, who've dedicated years to picking it apart and putting it back together, would start developing an unofficial PC port with a specific rallying cry:

"Fuck AI vibe coding bullshit."

DK64 Recompiled, announced in a YouTube video in late June, is actually the second project with that exact name, and that's no coincidence. It's the latest example of an increasingly common point of friction in the game emulation and hacking scene, as experienced programmers butt heads with those eager to use new AI tools to spin up new ports of old games in record time. In some cases, ports written with AI tools like Claude are ostracized or ignored. But in the case of Donkey Kong 64, the game's biggest fans weren't about to cede ground.

"The DK64 randomizer dev team is owning this shit now," contributor 2dos wrote in the N64 Recomp Discord server, shortly after the video's debut. "We have decided to take on and own this development due to the poor development direction of another recompilation project of DK64 happening concurrently that is heavily leaning on AI vibe code, and thus is resulting in a poor quality project that will make it progressively harder to manage as time goes on. Our DK64 recompilation project is backed by folks who have worked on and tinkered with the backend of the DK64 code for well over a decade, and do not rely on vibe coding to make code changes."

Static recompilation is a technique that, in short, translates all of a game's code from one system's language to another; unlike emulation, it's not simulating the original hardware holistically. The most prominent hobby project using this technique so far is N64 Recomp, announced just over two years ago , which is a recompilation tool specifically geared towards Nintendo 64 games. Since its debut we've seen modders use it to create PC ports of Banjo-Kazooie , Star Fox 64, Mario Kart 64, and more, while similar projects have sprung up for other consoles like the Xbox 360.

Legally, the freshly written code is clean because these recompiled ports don't contain any of the assets from the original games—players have to provide a ROM file themselves. The porting process has become an ethical, rather than legal, battleground in the last few months, as "vibe coded" projects have begun cropping up practically every day.

One developer has used Claude's AI coding to spin up recompilation tools for the NES, Sega Genesis, PlayStation, Super Nintendo, Game Boy Advance, and Virtual Boy, each with a handful of games at least bootable so far. Another's got GitHub repositories for everything from the PS3 to the ZX Spectrum. Others have used the Xbox 360 recompilation framework ReXGlue to pump out ports at a seemingly astonishing rate—but once you get past the YouTube videos hyping them up, the reality isn't quite so impressive.

"Vibe coding a recomp has started to become significantly more prominent and not in a good way," says Ballaam, a contributor to the DK64 randomizer and now DK64 Recompiled alongside 2dos. "I think GoldenEye was released somewhat recently with being heavily vibe coded and, as a result of that (and probably either dev inexperience or overconfidence in the AI output), it leads to some very critical bugs which make the recomp borderline unplayable. The main N64 recomp server has gone as far as to not promote any recomp which uses AI for a significant portion of it."

(Image credit: Rare, Banjo Kazooie Recompiled) That GoldenEye project has also drawn criticism from those who know the game inside and out. The paradox of current tools making it easier than ever to start projects means anyone can vibe code their way to a partially working port, but shaping it into an accurate, crash-free, feature-rich version of a game is a different story.

2dos warns that a big risk with relying on AI is technical debt—a codebase becoming harder to understand and modify as it grows, with maintenance becoming "a nightmare" if you don't put in the work to thoroughly document it. The knock-on effect is that other humans are less likely to want (or even be able) to contribute. If a developer doesn't thoroughly understand the code the AI writes, that will also require them to spend more tokens to fix bugs. Those costs can add up fast, and the AI's "fixes" won't necessarily be good, either.

In the case of the AI-based Donkey Kong 64 recompilation, 2dos discovered that the agentic coding had modified RT64 , "the base module that makes the N64 graphics render."

"That would lead to significant management and compatibility issues down the line, and cause other unexpected side effects that are currently visible in preview videos in the other recomp," he says.

Vibes: Questionable

Using AI is attractive because it skips past the slow, manual process of figuring out exactly what every function in a game is doing, which can often take years. Many of the polished recompilation ports available currently are for games that have previously been decompiled , creating fresh, human-readable source code. It's an enormous undertaking even for old games, but understanding what the source code is doing helps enormously when troubleshooting.

Decompilation also makes modding much more viable, which is one of the most attractive features for modern ports of retro games. (DK64: Recompiled's debut trailer touts support for a tag anywhere mod that will let players swap between Kongs on the fly, addressing one of the most common complaints about the game's backtrack-heavy design.)

Developer SrBananaMan64 recently released a PC port of Harvest Moon 64 alongside a range of mods to speed up messages, zoom out the camera, and so on. A completed decompilation by another developer helped make that a reality.

"They're using a powerful tool without putting in the work to actually learn what it does"

ReXGlue developer

"Recompilation doesn’t actually need a 100% decomp to work, just enough for the symbols to be mapped out to produce a viable .elf file that N64Recomp uses," SrBananaMan64 says. "But this game seemed like it needed the proper enhancements to spruce [it] up… The quality of the game is lacking so it needs the widescreen and high frame rates to compensate for that in my opinion."

He shares the DK team's view that developers "should never let AI take the wheel," even if it can occasionally be an "amazing tool" for specific uses. "Minuscule tasks such as searching through a decomp for a certain function you would like to patch or mod," for example.

Lines for acceptable AI usage are now being drawn across the open source development scene. The maintainers of the Godot engine are now rejecting "substantial pieces" of AI-generated code . A nascent recompilation project for the GameCube & Wii declares "No AI code is used," and that it's a "human hand-made project by a group of passionate developers, who want the best for the Retro Gaming community." So does a solo project tackling recompilation for the PS1 .

An in-progress recompilation of Castlevania: Symphony of the Night via RecompOne (Image credit: RecompOne, Konami) A port for Sonic Advance also has a "strict no AI policy," championing the endeavor's underlying purpose: "We want humans to use this project to practice their skills. Reverse engineering, documenting, understanding, being part of the process."

Tom, the developer behind ReXGlue, sees the low-effort ports as a "people problem," not an issue with the tool itself. "The real issue is people who don't bother to understand what they're posting," he says. "They're using a powerful tool without putting in the work to actually learn what it does, and then wondering why their project falls apart, performs poorly, or has nothing genuinely new to bring to the table. It's like riding a motorcycle before you can even ride a bike."

One common defense for vibe coding ports boil down to "this is just my hobby project, so ignore it if you don't like it." One counter-argument to that point is ethical. Even setting aside the broader environmental impact and stolen training data of generative AI models, there's an argument that even a fan-made port should endeavor to do justice to the original work—to honor the game's original creators by representing it as accuracy as possible. Another is that a vibe coded port that's likely to have those long-term maintenance issues 2dos warned about could pull interest away from a human-coded project that's more likely to be perfected in the long run. The existence of a half-assed project could even turn another developer away from putting months or years of their work into a game at all.

"That’s a big risk," SrBananaMan64 says. For example the Smash Bros. Battleship project which was completely AI generated will take away from the release of mine." (He's currently working on his own recompilation of the N64 fighter).

Another common defense from those simply eager to play a game they love—that anything is better than nothing—also doesn't necessarily hold up beyond the initial burst of excitement.

(Image credit: Nintendo, DK64: Recompiled) "I think there is some sentiment from those who haven't experienced a vibe coded recomp that 'who cares if its vibe coded as long as it works," says Ballaam, "when they don't fully realize that for the most part, they really really don't.

"LLMs nowadays have lowered the skill floor of producing a functioning piece of code that those who are technologically curious but don't have the patience or passion to fully learn a language can feel like they're coding. They scale up their projects but don't scale up their learning to accommodate their desires because they rely too much on the AI crutch."

Perhaps someday, as agentic coding continues to advance, vibe coded ports really will be sufficient. Perhaps a small portion of this initial wave of vibe coders will even pick up enough interest in the process to learn to do the programming themselves. Or maybe the rising costs of AI models will dissuade casual hobbyists from burning tokens on projects that are hard to see through to completion.

For now we're in the thick of it, and we're sure to see more competing projects like Donkey Kong 64 Recompiled before we come out the other side.

Full text

Rumours continue to swirl around Intel's upcoming Nova Lake launch. The latest of these comes from noted leaker Jaykihn, who has hinted towards some exciting highlights among the desktop lineup. On X, the leaker wrote , "Nova Lake-H has a 12 Xe segment, and Nova Lake-S has a 12 Xe SKU. These are Xe3p."

Basically, this leak suggests there will be both a next-generation Nova Lake mobile CPU rocking 12 Xe cores as well as a desktop variant also sporting the same number of Intel graphics cores (though if that's true, I should probably be calling it an 'APU').

A high-performance iGPU was something sorely missing from the previous Arrow Lake generation of desktop chips, and having heard from our Dave what a 14 Xe core chip can do in the new MSI Claw handheld , it's exciting to think what a slightly upgraded version of Intel's GPU silicon could do in a wee little desktop chassis.

Anyway, drilling down a little more, the Xe 3P core is an upcoming iteration of Intel's microarchitecture. The discrete GPU tech was codenamed 'Celestial' for a time and was even meant to make its debut alongside Panther Lake. However, that chip went with what was essentially Xe2 'Plus Plus' , and Jaykihn also claims what was Celestial ' was canned long ago '.

In other words, Xe 3P has been cooking for a minute. When it finally arrives in the form of the expected Crescent Island GPUs later this year, it'll be primarily geared towards handling AI inference by leveraging LPDDR5X memory, rather than the pricier data centre du jour HBM. It's entirely possible a powerful bit of kit like that could also turn its hand to gaming, but it's unlikely Intel would bring that chip to any gaming-focused cards.

Yes. Nova Lake -H has a 12 Xe segment, and Nova Lake -S has a 12 Xe SKU.These are Xe3p. April 24, 2026

As for a possible mobile APU, Nick recently wrote that a Nova Lake mobile chip with a beefy iGPU would be great news for handhelds, but that it would likely be destined for edge computing only .

But decent integrated graphics? In my desktop processor? That's exciting for sure, but how likely is it? Well, let's look at the leaker in question. Jaykihn claimed back in January the B70 Pro would arrive 'soon' and it did . Then in April, they leaked details about Intel's Core 300 laptop chips that also appeared to mostly bear out . Sure, a broken clock can be right twice a day, but this leaker is noted for having a pretty good track record as these things go.

Full text

The U.S. Embassy in Jordan on Thursday warned Americans about “missiles, drones, or rockets …in Jordanian airspace.” The warning came as the U.S. and Iran reportedly exchanged fire amid resumed hostilities that have entered a third day. This latest round of fighting was touched off Tuesday when Iran attacked three ships in the Strait of Hormuz. On Wednesday, President Donald Trump and Iranian officials declared that the shaky ceasefire agreed to on April 8 was over.

“Seek overhead cover and shelter in place immediately,” the U.S. Embassy in Jordan cautioned this morning. “Remain indoors and pay attention to local announcements and alerts. The U.S. Embassy in Jordan will continue to review the situation and provide additional information as needed.”

Security Alert: U.S. Embassy Jordan – July 9, 2026

Location: Jordan, countrywide

Event: Reports indicate missiles, drones, or rockets are in Jordanian airspace. Seek overhead cover and shelter in place immediately. Remain indoors and pay attention to local announcements and… pic.twitter.com/QOpNpKLXTA

— U.S. Embassy Amman (@USEmbassyJordan) July 9, 2026

The Jordan Armed Forces-Arab Army (JAF) General Command said the country’s air defense systems “intercepted and shot down eight missiles launched from Iran toward Jordanian territory on Thursday. The interception operations resulted in the fall of missile debris, but no casualties or property damage were reported.”

“JAF is closely monitoring regional developments and remains at the highest level of operational readiness to safeguard the Kingdom’s airspace and defend its sovereignty and territorial integrity,” the command added. “JAF will not allow any party to violate Jordanian airspace under any circumstances.”

Armed Forces Intercept, Shoot Down 8 Iranian Missiles Bound for Jordan https://t.co/eNiTJMTBLL pic.twitter.com/KEGuwl6P6w

— Jordan News Agency (Petra)-English News (@PetranewsEN) July 9, 2026

The Islamic Revolutionary Guard Corps (IRGC)-connected Tasnim media outlet claimed “that multiple explosions occurred at the Muwaffaq Salti Air Base in Jordan,” and that “Iranian forces also targeted American ships stationed off the coast of Bahrain with cruise missiles.” However, no visual proof of that has emerged. U.S. Central Command declined to provide any new operational details.

Muwaffaq Salti, as we have previously reported , has come under attack numerous times. The air base has long been a major regional hub for U.S. operations, and is being very actively utilized in the current conflict. It hosts the greatest concentration of U.S. tactical aircraft in the region, and thus is an extremely important target, where even one ballistic missile landing on an apron could destroy multiple prized fighter aircraft and take the lives of U.S. service members.

Scenes from IRGC Navy and Aerospace retaliatory strikes on Thursday answering the American terrorist army's aggression and disloyalty. pic.twitter.com/qwKo0SkLE2

— IRNA News Agency ☫ (@IrnaEnglish) July 9, 2026

Iranian officials also claimed the U.S. carried out new strikes in southern Iran, including on the perimeter of the Bushehr nuclear power plant.

“According to the deputy governor of Bushehr Province, a U.S. projectile hit the perimeter area of the facility, which had already been hit several times during the current conflict prior to the April 8 ceasefire,” the Jerusalem Post reported Thursday .

Iranian state media: A U.S. projectile struck the vicinity of the Bushehr nuclear power plant. pic.twitter.com/lPRClSNFz7

— Amichai Stein (@AmichaiStein1) July 9, 2026

Iran reports that a few minutes ago, additional explosions were heard near the Bushehr nuclear power plant. https://t.co/zVJczmOpb5

— Raylan Givens (@JewishWarrior13) July 9, 2026

Today’s attacks follow a series of strikes carried out on Wednesday by U.S. Central Command (CENTCOM). The U.S. attacked scores of targets across Iran after President Donald Trump warned on Wednesday that “we’ll probably hit them hard again tonight.”

“U.S. forces struck approximately 90 Iranian military targets including air defense systems, coastal surveillance assets, missile and drone storage sites, naval capabilities, and military logistics infrastructure along Iran’s coastline,” the command said in a statement on X. “The latest strikes follow successful execution of offensive strikes in Iran the night before.”

CENTCOM RELEASES THE HIGHLIGHT REEL

U.S. Central Command says American forces completed another round of strikes against Iran on July 8, hitting approximately 90 Iranian military targets along the coastline.

Targets included air defense systems, coastal surveillance assets,… pic.twitter.com/y2HiEMNWdy

— Mossad Commentary (@MOSSADil) July 9, 2026

Iran, however, claims the attacks took place further inland, including on a railway bridge for a line linking the cities of Tehran and Mashhad, where former Supreme Leader Ayatollah Ali Khamenei is scheduled to be laid to rest. He was killed on the first day of the U.S.-Israeli attacks on Iran. The official Iranian IRIB media outlet claimed on X that the bridge attack interrupted rail service.

Passengers on the Tehran-Mashhad train, stranded due to the U.S. striking the railway, are chanting: "Iranians do not accept humiliation, even at the cost of their own lives."

pic.twitter.com/9NhlUQqidr

— IRIB (Islamic Republic of Iran Broadcasting) (@iribnews_irib) July 9, 2026

There were also unconfirmed reports from official Iranian media , an independent media outlet and eyewitnesses that another bridge in that region was attacked. The Agh Tekeh Khan Bridge in Iran’s northern Golestan province is a key link in a railway line running to Central Asia and ultimately Russia and China. The line has reportedly been an important means of trade between Iran, China and Russia during the now paused U.S. naval blockade on Iranian ports.

Eyewitness video shows damage to the Aq Taqeh Khan railway bridge in Iran’s northern Golestan province following overnight US strikes. The bridge is a key trade link connecting Tehran with its strategic partners, China and Russia. #US #Iran pic.twitter.com/0k5y50w79V

— Al Arabiya English (@AlArabiya_Eng) July 9, 2026

More pictures of destroyed Aq Tekeh Khan Bridge on Iran-Turkmenistan Railway Line. pic.twitter.com/02XKyCR4vG

— Open Source Intel (@Osint613) July 9, 2026

According to local reports, several enemy projectiles have struck the "Aq-Tekeh Khan" bridge on the railway line in the western area of Aq Qala, Golestan province. pic.twitter.com/eMO5todo2i

— IRIB (Islamic Republic of Iran Broadcasting) (@iribnews_irib) July 8, 2026

Where all this leaves the ongoing negotiations between the U.S. and Iran remains unclear. As we have frequently noted , the two sides signed a 14-point Memorandum of Understanding (MoU) on June 17. The MoU provided a 60-day extension of the ceasefire to iron out an agreement to end fighting throughout the region, including Lebanon, prevent Iran from seeking nuclear weapons, end U.S. sanctions and resume the flow of traffic through the Strait of Hormuz among other points.

Control over the Strait has proven to be the biggest flashpoint, as evidenced by the aforementioned flare-up of fighting sparked by Iranian attacks on shipping there.

As we mentioned earlier in this story, both sides have declared the ceasefire over.

Speaking to reporters aboard Air Force One yesterday after departing the NATO Summit in Ankara, Turkey, Trump claimed Iran still wants a peaceful solution.

“They called a little while ago,” the president proclaimed. “They want to make a deal so badly — I just don’t know if they’re worthy of making a deal. I don’t know that they’re going to honor the deal. That’s the problem.”

. @POTUS on Iran: "They called a little while ago. They want to make a deal so badly — I just don't know if they're worthy of making a deal. I don't know that they're going to honor the deal. That's the problem." pic.twitter.com/jQTENvyRGM

— Rapid Response 47 (@RapidResponse47) July 9, 2026

Given all this, the future of diplomacy is an open question.

Contact the author: howard@twz.com

The post Renewed Attacks Between U.S. And Iran Stretch Into Third Day appeared first on TWZ .

Full text

Denza reveals tri-motor, four-seat Z supercar at 2026 Goodwood Festival of Speed.

1,582 hp output gets it to 62 mph in 1.96 seconds, battery fills 10-97% in just 9 mins.

Denza confirms Spider, Coupe, Racing trims with a 1,973+ hp hardcore version to come.

Electric supercars have a habit of producing ridiculous power and acceleration numbers, but are usually accompanied by a ridiculous price. Denza wants to change that with the Z, which outsprints a 911 Turbo S to 62 mph (100 kmh) and makes double the power, while costing less than a mid-range GTS.

var adpushup = window.adpushup = window.adpushup || {que:[]};

adpushup.que.push(function() {

if (adpushup.config.platform !== "DESKTOP"){

adpushup.triggerAd("0f7e3106-c4d6-4db4-8135-c508879a76f8");

} else {

adpushup.triggerAd("82503191-e1d1-435a-874f-9c78a2a54a2f");

}

});

Unveiled today at the Goodwood Festival of Speed, BYD-owned Denza’s Z costs from £142,900 (about €167,400 / $191,400 at current rates) in coupe form in the UK, where a 911 Carrera 4 GTS costs £145,900 (€170,900 / $195,400). The open-top Spider starts at £159,900 (€187,300 / $214,100), and the track-focused Racing version costs £172,900 (€202,500 / $231,500), which compares with £199,500 (€233,700 / $267,100) for the new 911 Turbo S.

Denza hasn’t confirmed home-market numbers, but local estimates put Chinese pricing at around ¥400,000 to ¥500,000 ($58,600-73,200), a fraction of the British ask and a sign of what shipping, tariffs, and selling far from Shenzhen do to the sticker.

Three Motors, One Platform

var adpushup = window.adpushup = window.adpushup || {que:[]};

adpushup.que.push(function() {

if (adpushup.config.platform !== "DESKTOP"){

adpushup.triggerAd("89c65a14-4682-4807-a4bb-5436c5361bee");

} else{

adpushup.triggerAd("f25825bd-1260-4574-84ce-2b26e315fad8");

}

});

All three versions of the Z ride on Denza’s new e3 Sports Car Platform and use the same triple-motor setup. One motor drives the front axle while two independently power the rear wheels, producing a combined 1,582 hp (1,604 PS / 1,180 kW) and 915 lb-ft (1,240 Nm) of torque.

More: Denza’s 1,139-HP Wagon Will Now Drift Any Shape You Draw On The Screen

The Coupe reaches 62 mph in 2.25 seconds (Porsche claims 2.5 seconds for the 701 hp / 711 PS Turbo S), while the Racing cuts that to just 1.96 seconds with optional semi-slick tyres and boosts the top speed from 186 mph (300 km/h) to 217 mph (350 km/h).

Building Up To An EV Record Attempt

var adpushup = window.adpushup = window.adpushup || {que:[]};

adpushup.que.push(function() {

if (adpushup.config.platform !== "DESKTOP"){

adpushup.triggerAd("bb7964e9-07de-4b06-a83e-ead35079d53c");

} else {

adpushup.triggerAd("9b1169d9-7a89-4971-a77f-1397f7588751");

}

});

A fourth model, the Z Special Edition, which Denza says is a Nurburgring natural, makes over 1,973 hp (2,000 PS / 1,471 kW), and hits 62 mph in less than 1.7 seconds. But the automaker is saving more details about that one for later, presumably when it takes a crack at the Ring EV record.

Every Z packs a 76 kWh second-generation Blade Battery that supports BYD Group’s latest FLASH Charging technology. On BYD’s new 1,500 kW chargers, Denza claims the battery can jump from 10 to 70 percent in just five minutes, or reach 97 percent in only nine. That’s close to petrol tank-filling territory, though it’ll be a while before those chargers are commonplace outside China, even if their rollout has been confirmed for Europe.

Range varies from 236 WLTP miles (380 km) for the Racing to 254 miles (409 km) for the basic Coupe, which at 2,230 kg (4,916 lbs) is actually 20 kg (44 lbs) lighter. The 2,300 kg (5,070 lbs) Spider claims 248 miles (399 km) between fills.

Denza didn’t leave the chassis out of the fun, adding magnetorheological dampers, carbon-ceramic brakes, torque vectoring and even a compass turn function that lets the car pivot around its front axle. And the long 2,780 mm (109.5 inches) wheelbase (a 911 ’s is 2,450 mm / 96.5 inches) means that while it looks like a traditional supercar, it’s actually got space for four, though the rears do look tiny.

var adpushup = window.adpushup = window.adpushup || {que:[]};

adpushup.que.push(function() {

if (adpushup.config.platform !== "DESKTOP"){

adpushup.triggerAd("b25ecba7-3bbb-4ea7-a3a8-dbea91695c07");

} else {

adpushup.triggerAd("e46c436a-adeb-4b5e-a2c7-56bc36561c10");

}

});

Supercar With Massage Seats

The cabin sounds lavish in typical tech-heavy Chinese EV fashion, with massage seats, a Devialet audio system, ambient lighting, a 12.8-inch Google-equipped infotainment system, and a digital rear-view mirror all fitted as standard. There’s also a configurable virtual engine soundtrack plus dedicated Boost and Track modes.

On paper, it all sounds great, but as Porsche and Rimac know, Western buyers of luxury and high-end performance machinery generally aren’t interested in electric cars. They tend to want combustion power, which is why Porsche recently asserted that it no longer had plans to make an electric 911 . But this package looks strong enough to persuade a few to at least take a look at what’s on offer.

Denza

Full text

Using a quantum computer alongside a supercomputer, scientists have developed a breakthrough pathway for modeling the physics inside a fusion reactor. The world-first experiment could help clear a path to developing clean, abundant nuclear power and solving the global energy crisis, the researchers said.

Using hybrid quantum computing and artificial intelligence (AI) methods, scientists with IBM and Oak Ridge National Laboratory (ORNL) have blueprinted how to make tritium, an extremely rare isotope of hydrogen that's critical to the fusion process.

Although their research — uploaded June 29 to the preprint server arXiv — has not been peer-reviewed, the researchers say it's the first time that different kinds of computing elements have come together to propose the most effective way to create this material.

Fusion reactors are experimental power sources that create energy by fusing atomic nuclei. The heat produced in the subsequent nuclear reaction is then harnessed as energy. This method produces no carbon byproducts or long-lived radioactive waste, making it one of the cleanest potential forms of mass energy production.

It's projected that, at scale, a single fusion reactor could produce about 4 million times as much energy as a coal-burning facility and around four times the amount of energy as a modern nuclear fission reactor.

Current attempts at building a viable fusion reactor have resulted in numerous laboratory experiments that prove the technology works, with magnetic confinement reactors, such as tokamaks, widely considered the front-runner . But many engineering challenges remain before the first commercial reactors could come online.

Turning seawater into fuel

The base fuel for nuclear fusion reactors is a hydrogen isotope called deuterium, which is commonly found in seawater. It's estimated that there are 33 grams of deuterium in every cubic meter of seawater.

But deuterium is only half of the equation. Nuclear fusion also requires tritium — a heavier hydrogen isotope — and the fusion released from just 1 gram (0.04 ounces) of deuterium-tritium fuel equals the energy from about 2,400 gallons (9,100 liters) of oil, according to the U.S. Department of Energy .

Unfortunately, tritium, a radioactive isotope, is extremely rare; only 44 pounds (20 kilograms) of it is produced on Earth each year, and its 12-year half-life makes it difficult to use in nuclear power plants.

Instead, scientists must painstakingly produce tritium in nuclear reactors by bombarding lithium atoms with neutrons. It's then superheated and bound with powerful magnets into a whirling ring of plasma within a tokamak, a special fusion chamber designed to shape and heat plasma using magnetic fields.

A diagram showing the process of nuclear fusion. (Image credit: Designua | Shutterstock) Scientists add more deuterium and then bash the tritium and deuterium together, causing them to fuse into helium. The force of this reaction creates heat that's converted into energy.

The current bottleneck lies in creating enough tritium to sustain fusion long enough to produce energy. But modeling the particle physics and chemical reactions involved in the tritium-creation process has proved beyond the capabilities of classical supercomputers.

In the new study, however, scientists say they have addressed this bottleneck by simulating nine molecular configurations of a liquid salt that contains fluorine, lithium and beryllium (FLiBe) — one of the leading candidate materials for extracting tritium.

This is the first time quantum computers have been used to model reactions inside a fusion reactor. If perfected, FLiBe could provide a near-limitless source of fuel for nuclear fusion reactors, they said, but the chemistry involved is incredibly complex.

Demystifying complex chemistry

A "blanket of molten salt" made of FLiBe is wrapped around the nuclear reaction inside a fusion reactor, IBM researchers told Live Science. This provides both a fuel source and a thermal shield for the device.

To create enough tritium, the researchers had to calculate the physics involved while a process called "neutron bombardment" constantly altered the blanket's chemistry. Designing a salt that holds up under competing demands and keeps releasing tritium is a key problem in building this kind of reactor.

"If tritium grabs onto fluorine in the salt, it forms tritium fluoride, which is corrosive and stubborn to remove," the researchers explained. "If it binds to another tritium atom to form a gas, it bubbles out on its own. Predicting which way the reaction goes means modeling the interaction between tritium and the salt with high precision and accuracy that is challenging for classical methods."

Because no ordinary computer can perform the necessary calculations, the team used a combination of AI running on the Frontier supercomputer at ORNL, alongside quantum computing algorithms running on an IBM Quantum Heron quantum processing unit (QPU) in New York. The resulting workflow demonstrated a proof of concept for offloading complex chemistry computations to a quantum computer.

That workflow relied on a technique called wave-function-based embedding, which fragments the calculation into easier-to-calculate clusters, the scientists said in the study. They used classical computers to solve the smaller clusters and passed off the more difficult chunks to a quantum computer. The classical computers then stitched the molecule back together.

This is a method that study co-author Kenneth Merz , a biochemist and principal investigator at Cleveland Clinic Research, pioneered in previous research. Earlier this year, in collaboration with IBM and the Japanese national research institute RIKEN, he used quantum computers to calculate the structure of a 12,635-atom protein .

Fusing quantum and AI

In the new study, the researchers tested their model against known molecular configurations that were already solved by a nonhybrid classical system and determined that the accuracy was maintained with the addition of quantum computations.

This proof of concept should serve as a direct pathway for scaling the models used to predict tritium production within fusion reactors, potentially solving what may be the biggest hurdle to large-scale fusion energy production.

The broader workflow the scientists outlined in a technical blog post involved three stages. First, AI agents proposed and screened many candidate salts from the ORNL database, and for each candidate, calculations estimated various qualities in the tritium breeding process, including how much fuel the salt would make under neutron bombardment.

Related stories

Scientists trained an AI model using an IBM quantum computer — and it answered questions correctly that the base model couldn't

New data center will be partially powered by human brain cells for the first time

Meet the world's smallest AI supercomputer — it packs 'doctorate-level intelligence', its makers say, and can fit into your pocket

The most promising salts then went to a supercomputer, which modeled them atom by atom, using the density functional theory (DFT) process to approximate how a molecule's electrons would arrange themselves. These are expensive simulations, so the scientists used "AI stand-ins" trained to reproduce the physics to run them fast enough to be useful. The third stage brought in the quantum computer to figure out where the tritium would bind, which is a shortcoming for DFT.

In the future, the research team will model larger molten-salt systems and study more molecular configurations before evaluating whether AI can slash the time it will take to find a promising molten-salt material.

The wider aim, the scientists told Live Science, is to build a reliable computational pathway for fusion-materials discovery that can help researchers predict how well a blanket material breeds tritium, whether that tritium can be recovered, and how the material may perform in the extreme environment of a fusion reactor.

Can you match these ancient devices to their pictures? Find out with our computing quiz!

Full text

Amazfit Active 3 Premium: One minute review

Amazfit has been a company I’ve been keeping an eye on since their Bip 3 ‘budget’ smartwatch blew through all my expectations (and then some), followed by a slew of other watches across a range of price points, aiming to undercut the likes of Garmin — but most notably the $99 / £99 Amazfit Active 2, which earned five stars in our review . So when they put the word “Premium” on a product, it piques my curiosity.

Not only does the Amazfit Active 3 Premium have a bunch of excellent features that rival some of the biggest in the business, like excellent sleep and fitness tracking, but it’s also a fantastic-looking smartwatch.

The ‘Premium’ moniker signifies the addition of the super-tough sapphire glass coating, and the inclusion of the NFC payment feature. This was the distinction that the Amazfit Active 2 Premium used to separate itself from the base Active 2: however, there is no base Active 3, so even though this watch uses the ‘Premium’ model distinction, it’s the only watch in its range.

Despite this, it’s not an expensive buy. Sure, it’s pricier than the older Active 2, but it’s still undercutting the cheapest Apple and Garmin watches while delivering performance that belies its price. Another excellent example of a fitness tracker from Amazfit, a company which has by now shown itself to be a serious contender in the fitness tech space.

Amazfit Active 3 Premium: Specifications

Component

Amazfit Active 3 Premium

Price

$169 / £169 / AU$239

Weight

54.6g

Case/bezel

Metal and plastic hybrid

Display

1.32-inch AMOLED display

GPS

Dual-frequency GPS

Battery life

Up to 12 days

Connection

Bluetooth

Water resistant?

5 ATM

Amazfit Active 3 Premium: Price and availability

(Image credit: Amazfit) Starts at $169 / £169 / AU$239

Not Amazfit’s cheapest, but still cheaper than many rivals

Available in three color options

It’s hard not to look at the Amazfit Active 3 Premium and not feel as though there’s a catch somewhere when it comes to its price, but it really is a sub-$170 / £170 / AU$250 fitness watch with the kinds of features and accuracy you’d have spent double that for not long ago. Our review unit is the Atlas Blue model, but it’s also available in Apex Silver and Aero White.

Value score: 5/5

Amazfit Active 3 Premium review: Design

(Image credit: Future) Heavier than some alternatives

AMOLED display

Regardless of your color choice, the Amazfit Active 3 Premium is a looker. It has a circular body with a metallic finish, with the underside color-matched to the bands for Atlas Blue and Aero White (Apex Silver would be my pick, but the whole body is silver with a black strap).

The strap itself is the standard polymer-style you’ll have felt on just about any other smartwatch, but the fact that the Amazfit Active 3 Premium uses a pin system like a more traditional watch adds a touch of class (and some minor fiddliness). It’s got a 1.32-inch display, which isn’t large —but it’s the same size as the Amazfit Active 2. I didn’t feel like I was losing much in terms of screen space compared to some of Amazfit’s larger watches, but I was surprised to find it’s slightly heavier than the Coros Nomad I tested recently.

With that said, I’m not sure I’d consider it rugged. I wasn’t prepared to hurl my review unit to the ground to test its durability, despite its sapphire glass and scratch-resistant coating. It might not be the watch you’ll want to take on an ultramarathon, but it’ll stand up to runs, rides and gym workouts with ease.

Design score: 4.5/5

Amazfit Active 3 Premium: Features

(Image credit: Future) Full-color maps

HybridCharge is great

170 sports modes

Ok, so the Amazfit Active 3 Premium is a looker, but does it have the smarts to back it up? Despite the relatively low price, the answer is a resounding yes. You’ll need to use the Zepp app, and while I’ve tested it before and found it to be a little cumbersome, I am pleased to report that (at least in the time I’ve been away), it’s much slicker these days.

You can use it as a dashboard, and there are badges to collect that include step goals, monthly targets, and much more.

Fitbit Premium’s Daily Readiness Score has led the charge to distill a bunch of complex metrics into an easy-to-read number that calculates how hard you can push your workout on any given day, and Amazfit’s version is called “HybridCharge”.

It takes into account sleep, heart rate, activity, and recovery, as well as subjective inputs like stress, health, and more, to help you make a judgment on whether today is a workout day or a rest day, and it does so pretty well.

One day during testing, I’d struggled to sleep and had an early start at work, and the app and watch understood that maybe I needed to take it easy and recover that day. It’s something you used to need a premium subscription for with other manufacturers, but I’m glad to see it’s included in the price of the product here.

PAI is also back, so those looking for a quick look at a glance to see how their natural fitness is holding up can check it out by testing their heart rate zones. The watch supports basic notification features, and of course offers that NFC payment digital wallet feature for convenience. A barnstorming inclusion is full-color maps, offering turn-by-turn navigation. This is a premium feature crammed into a value package, offering turn-by-turn navigation during runs, rides and walks.

Features score: 4.5/5

Amazfit Active 3 Premium: Performance

(Image credit: Amazfit) AMOLED display

12 days of battery life

Excellent accuracy

Performance is uniformly excellent across the board with the Amazfit Active 3 Premium. Fitness tracking offers a deep suite of metrics, including running ones like stride length, ground contact, and just about anything else you could possibly ask for as a runner. And, if you’re just learning what all of these measurements are, fear not: Everything is handily explained within the Zepp app.

When you’re done, the app also does a great job at showcasing your fatigue level, which then feeds into the aforementioned HybridCharge metrics.

I very rarely check my sleep outside of Apple’s Sleep Score number, but the Amazfit Active 3 Premium does make me want to dig deeper — and offers plenty of data to do so. I was surprised at how often I was getting up in the night without realising it, and the Aura section of the Zepp app even includes relaxing music to help you settle back down.

Battery-wise, Amazfit claims you can get 12 days from a single charge. In my experience, it’s been closer to 10 days, but it’s still very respectable given the amount of data it’s collecting day in, day out. The AMOLED display is plenty bright for just about everything, even direct sunlight.

The only minor gripe I have with the Amazfit Active 3 Premium is that it doesn’t automatically offer to track a workout if you start walking/running/doing anything. As someone who still forgets to hit the button before setting out, it’s nice when devices like the Apple Watch give a nudge.

Performance Score: 4.5 / 5

Amazfit Active 3 Premium: Scorecard

Category

Comment

Score

Value

Simply outstanding.

5/5

Design

Iterates on the proven Active 2.

4.5/5

Features

Lots of sports modes, tracking and HybridCharge.

4.5/5

Performance

Bright screen, responsive battery, accurate metrics.

4.5/5

Amazfit Active 3: Should I buy?

Buy it if...

You want a watch that looks good

The Amazfit Active 3 Premium is one of the best-looking and still affordable watches around.

You’re looking for a deep dive into your fitness metrics

The Zepp app has a whole bunch of data, and makes it easy to sift through and spot patterns across weeks or even months of use.

Don't buy it if...

You’re looking for smartwatch features

Alas, there’s no contactless payment functionality here, nor can you run third-party apps a la an Apple Watch.

You’re going off-trail

The Amazfit Active 3 Premium is unlikely to take a battering as you scamper up rockfaces and the like.

Also consider

Garmin Forerunner 70

Garmin's new budget running watch

Read our full review

Apple Watch SE 3

Apple's cheapest watch is a little more expensive, but still offers great value for iPhone users.

Read our full review

How I tested the Amazfit Active 3 Premium

I’ve been testing the Amazfit Active 3 Premium while walking, hitting the gym for weights and cardio (primarily using the cross trainer), and wearing it for just about everything in life.

I tested the GPS across various areas of London and more remote areas like the countryside where I live, and wore it to bed, too. I tested many of its measurements by pairing it with my Apple Watch Series 11 on the opposite wrist.