Il y a encore quelques années, le LASIK était vu comme le sommet de la chirurgie réfractive : un laser ultra-précis, une opération rapide, une vision nette en quelques heures. Mais derrière ce vernis high-tech, une réalité demeure : on découpe la cornée. On sculpte l’œil, littéralement. Or, une équipe de chercheurs californiens vient peut-être de son...
by Sylvain Biget, Journaliste, télépilote professionnel de drones et réalisateur de documentaires
Qui n’imite point n’invente point. Pour pénétrer rapidement et en profondeur chez l’ennemi, la Chine compte développer un porte-drones dont l’aile pivote pour faciliter le vol à Mach 5. Un concept expérimenté par la Nasa à la fin des années 1970.
Voici des mini-robots avec un mode de locomotion un peu particulier : ils utilisent les ondes sonores.
Le monde de la robotique vient de franchir une étape de plus vers la science-fiction. Des chercheurs ont mis au point des mini-robots qui avancent et interagissent via des ondes sonores. Ils se déplacent également en essaims de façon autonome.
Le son, la clé de cette génération de micro-robots
Ces micro-robots sont si petits qu’ils tiennent dans la paume d’une main. Ce qui les rend si spéciaux, c’est leur capacité à se mouvoir sans moteur ni roues. Ils se déplacent effectivement grâce à un système d’émetteurs et de récepteurs qui génèrent et captent le son. L’énergie des ondes sonores se convertit ensuite en mouvement et aide les mini-robots à se propulser sur n’importe quelle surface. Cela peut être un sol lisse ou un liquide.
L’objectif de cette recherche est de créer des robots qui n’ont pas besoin de supervision. Ils doivent aussi pouvoir s’organiser en groupes pour accomplir des tâches complexes. Les applications sont innombrables, à l’instar de l’inspection de bâtiments. Citons également la recherche et le sauvetage de personnes dans des zones dangereuses, ou alors l’exploration de l’espace.
Des mini-robots pratiques qui marchent aux ondes sonores
L’une des particularités les plus impressionnantes de ces robots est leur capacité à se déplacer en essaims. Les scientifiques ont créé un algorithme grâce auquel ils peuvent interagir entre eux et se coordonner pour accomplir des missions. Un peu comme une colonie de fourmis, ils communiquent en utilisant les ondes acoustiques. Ils partagent aussi des informations et s’organisent pour faire face à des obstacles.
Cette approche collective est surtout utile pour des tâches qui nécessitent la collaboration de plusieurs robots. Les mini-robots qui se déplacent via les ondes sonores seraient par exemple utiles pour le transport d’objets lourds. Un autre domaine dans lequel ils pourraient servir serait la cartographie d’un environnement.
Inspired by how animals use sound for communication and navigation, researchers create models of tiny robots that coordinate using acoustic waves to form intelligent groups. pic.twitter.com/P68dzIVEkO
Au lieu d’avoir un seul robot coûteux et complexe, l’idée est donc d’utiliser une multitude de petits robots simples. Ces derniers peuvent accomplir un travail bien plus rapidement et efficacement. Cette technologie pourrait bien redéfinir la manière dont on conçoit et utilise les robots à l’avenir.
Ostensibly, C:\ArsGames is to some extent about actually driving a few game purchases, but in reality it's mostly an excuse for me and my colleagues to wax nostalgic about the games that were formative for us. Case in point: This entry in our ongoing series with GOG is about a game that's completely free. I think Ars can withstand this tiny revenue shortfall for the sake of peak nostalgia!
But it's also interesting because of where its developer, Bethesda—a studio people both love and hate—is at today. Going back to Daggerfall, we find a game that shows off so much of what we've lost from the bygone era of '90s PC gaming, but also one that makes it abundantly clear why the industry left those sensibilities behind.
Last month, astronomers made an exciting discovery, observing an interstellar object — only the third ever observed — hurtling toward the center of the solar system.
The object, dubbed 3I/ATLAS, has caught the attention of Harvard astronomer Avi Loeb, who has a long track record of making controversial predictions about previous interstellar objects being relics from an extraterrestrial civilization.
While there's been a growing consensus among astronomers that the latest object is a comet, Loeb has continued to entertain the idea that it may have been sent to us by an intelligent species from outside of the solar system — and he's far from backing down.
In a blog post over the weekend, Loeb pointed to observations by NASA's Hubble Space Telescope, which showed a "glow of light, likely from a coma, ahead of the motion of 3I/ATLAS towards the Sun."
A coma is the hazy and luminous cloud that surrounds the nucleus of a comet.
However, there's "no evidence for a bright cometary tail in the opposite direction," he wrote, with scientists suggesting it was evidence that dust was evaporating from the object's Sun-facing side.
The observations led Loeb and his colleagues to an intriguing, albeit far-fetched possibility: is the mysterious space object generating "its own light?"
After deliberations with his colleague and Harvard astrophysicist Eric Keto, Loeb suggested that the "simplest interpretation" of 3I/ATLAS' observed "steep brightness profile" is that its nucleus "produces most of the light."
That would also mean that its actual size is much smaller than currently thought, roughly in line with the size of the first two interstellar objects we've observed, 'Oumuamua and 2I/Borisov.
The Harvard astronomer suggested two possibilities: either 3I/ATLAS is naturally emitting radiation because its a "rare fragment from the core of a nearby supernova that is rich in radioactive material" — or it's a "spacecraft powered by nuclear energy, and the dust emitted from its frontal surface might be from dirt that accumulated on its surface during its interstellar travel."
Loeb deemed the former explanation "highly unlikely," and the latter as requiring "better evidence to be viable."
Intriguingly, 3I/ATLAS will come within spitting distance — at least in astronomical terms — of Mars this fall, giving us a tantalizing opportunity to have a first-hand look. Loeb suggested using NASA's Mars Reconnaissance Orbiter to point its scientific instruments at the rare visitor.
Best of all, scientists at the space agency appear to be game.
"This morning, I encouraged the HiRISE team to use their camera during the first week of October 2025 in order to gather new data on 3I/ATLAS," Loeb wrote. "They responded favorably."
A rare genetic mutation that causes a deficiency in an immune regulator called ISG15 is known to make people more vulnerable to some bacterial infections and cause persistent inflammation — but it can unlock some unexpected antiviral "superpowers" as well.
As detailed in a new study published in the journal Science Translational Medicine, a team of scientists led by Columbia University professor of pediatric immunology, Dusan Bogunovic, has developed a new antiviral therapy that temporarily gives patients the same powerful antiviral properties by administering an experimental drug via a nasal drip.
The research kicked off with a happy accident.
"Our findings reinforce the power of research driven by curiosity without preconceived notions," said Bogunovic, in a statement about the research. "We were not looking for an antiviral when we began studying our rare patients, but the studies have inspired the potential development of a universal antiviral for everyone."
"We believe the technology will work even if we don’t know the identity of the virus," he added.
Years ago, Bogunovic stumbled upon an intriguing group of patients who seemed to have some mild inflammation that put them at risk for bacterial infections but were impervious to many types of viruses. Later, Bogunovic discovered that these people were deficient in an immune-regulating protein, ISG15, due to a genetic mutation.
Though the patients had mild inflammation from the mutation, Bogunovic determined that the condition gave patients the ability to ward off viruses.
"The type of inflammation they had was antiviral, and that’s when it dawned on me that these individuals could be hiding something," he recalled in the statement. "In the back of my mind, I kept thinking that if we could produce this type of light immune activation in other people, we could protect them from just about any virus."
Bogunovic and his team developed a therapeutic lipid nanoparticle package that holds ten mRNA molecules, which produce the antiviral protection of the ISG15 deficiency.
In experiments involving hamsters and mice, the therapy stopped viruses like influenza and the SARS-CoV-2 virus from replicating, albeit temporarily, after being injected into their lungs.
"We only generate a small amount of these ten proteins, for a very short time, and that leads to much less inflammation than what we see in ISG15-deficient individuals," Bogunovic explained. "But that inflammation is enough to prevent antiviral diseases."
However, the nanoparticles weren't produced at "high enough levels that makes us comfortable going into people immediately," he added.
Next steps are to have the therapy produce more of the virus-fighting proteins and determine how long the ISG15 immunity lasts. If it all works out, it could lay the groundwork for future genetic therapies, inspired by the types of vaccines that saved us during the COVID-19 pandemic.
To see just how much the new model changed things, we decided to put both GPT-5 and GPT-4o through our own gauntlet of test prompts. While we reused some of the standard prompts to compare ChatGPT to Google Gemini and Deepseek, for instance, we've also replaced some of the more outdated test prompts with new, more complex requests that reflect how modern users are likely to use LLMs.
These eight prompts are obviously far from a rigorous evaluation of everything LLMs can do, and judging the responses obviously involves some level of subjectivity. Still, we think this set of prompts and responses gives a fun overview of the kinds of differences in style and substance you might find if you decide to use OpenAI's older model instead of its newest.
As someone who’s spent time—and has friends—on both sides of the Atlantic, I’ve often needed to understand temperatures in both Fahrenheit and Celsius.
Like learning a new language or switching between miles and kilometres, the best way is to immerse yourself in a new scale so you just know what 55°F or 24°C feels like. But if you haven't reached that point, it can be helpful to have a few conversion benchmarks.
Fahrenheit and Celsius Temperature Palindromes
These two temperature palindromes are handy markers for gauging temperatures in both Fahrenheit and Celsius—and they’re surprisingly accurate:
82°F is 28°C
61°F is 16°C
I'm partial to a palindrome—a word or number that is read the same backwards as forwards—and as far as I'm concerned, it makes these two much easier to remember.
A couple of related points people shared with me, though some are less useful for the average weather forecast:
68°F = 20°C and 86°F = 30°C — useful benchmarks and 68/86 are palindromic
104°F is 40°C — almost a palindrome
11°F is ≈ -11°C (Actually -11.7°C.)
-40°F = -40°C
How to Convert Fahrenheit to Celsius (and vice versa)
The actual formula to convert Fahrenheit to Celsius is:
C = (F – 32) x 5/9
This is because:
The Fahrenheit scale starts at 32° higher than the Celsius scale. So, 32°F is 0°C.
The 5/9 means that each degree Celsius is just under 2 degrees Fahrenheit—it's actually 1.8°F.
This is why the basic conversion from Fahrenheit to Celsius that I use is "Minus 32, divide by 2" (which handily rhymes).
It's not perfect because each degree Fahrenheit is not quite half of a degree Celsius, but it's pretty close.
Using this simplified formula for the palindrome figures gives:
If you don't fancy some mental maths, using the palindromes is not a bad starting point. If you see a temperature around 61°F, you know it's around 16°C, and a temperature around 82°F is going to be around 28°C.
I learned about 82–28 from my neighbour when I mentioned I was working on the sketch for Temperature Scales: Fahrenheit and Celsius. Some digging gave me the 61–16 point as well.
Move over Roomba, there’s a new household hero in town, and it’s got arms, legs, and better folding skills than most college students! Meet Helix, the autonomous humanoid robot from Figure, powered by a Vision-Language-Action model that lets it understand you, respond, and then actually do what you ask. In this case, it’s folding laundry with the precision of a retail store employee on their first day. Today: laundry. Tomorrow: conquering the deepest, darkest corner of the junk drawer.
Over the weekend, a swarm of angry jellyfish forced the Gravelines Nuclear Power Station in France — one of Europe's largest — to take four of its six reactors offline. As reported by theNew York Times, the "massive and unpredictable presence of jellyfish" was first detected in the filters of the plant's massive ocean water pumping stations, which are used to cool the reactors.
While water-cooled nuclear plants typically have screens preventing ocean life from getting sucked into the cooling system, a large enough swarm of jellyfish can block the screens, rendering the pumps useless. Even worse, the NYT reports, with enough pressure, the jellies can "liquefy into a 'gel'" and pass through the screen, wreaking havoc on the cooling system itself.
Experts are warning it's a growing problem as global warming allows the critters to proliferate.
"The issue of jellyfish and power generation disruption remains a global challenge, as blooms of jellyfish are becoming more frequent and widespread due to factors such as overfishing, climate change and increased coastal development," the Oceanic Invertebrate Research Institute warns on its website.
Indeed, as sea temperature levels rise, massive jellyfish swarms have become a major hazard for marine-based industries and ecological management.
Described opportunistic organisms — a fancy word for pests, if you're feeling extra cruel about the fascinating invertebrates — ocean jellies can thrive in the kinds of low oxygen, high salt, and high temperature environments that are becoming increasingly common as climate change takes its toll.
Unfortunately for us, the gentle creatures wreak havoc on fishing nets, aquaculture operations, and power plants. A 2015 survey of the socioecological harm of jellyfish blooms along the Northern California current found that 67 percent of commercial fishers had their operations disrupted — and it's only gotten worse from there.
Another factor increasing the populations of nuisance jellyfish is human settlement and development along ocean coasts. As infrastructure like freight docks, marinas, and wind turbines becomes more commonplace, jellyfish gain access to even more surfaces on which to mature as polyps, the early growth stage before the filter-feeders cut loose and drift into the current.
The NYT notes that the rise of monster jellyfish swarms has had a considerable impact on nuclear plants around the world, causing shutdowns in Japan, Scotland, Israel, and Sweden.
Ce matin, je découvre le Methaphone, cet objet qui est là pour interroger nos gestes automatiques face au numérique, celui qui tient dans notre poche. Cette plaque d’acrylique transparente imite la forme d’un smartphone… sans rien faire d’autre qu’exister.
L’idée part d’un constat simple : on sort son téléphone par réflexe, même quand on n’en a pas besoin. Ce geste est devenu mécanique, presque compulsif. Ce projet s’inspire de la méthadone pour les héroïnomanes et transpose cette logique de substitution au smartphone. On garde donc le geste mais on enlève la fonction et ses effets. En tant que designer, ça me questionne sur la responsabilité de nos créations : comment nos interfaces encouragent-elles ces comportements ?
Le Methaphone a été vu par 150 millions de personnes en quelques jours sur TikTok, assez ironique pour un objet censé nous détacher des réseaux sociaux. Mais c’est peut-être là tout son concept : utiliser la viralité pour critiquer la viralité. Cet objet transparent nous renvoie ainsi à nos propres transparences face aux écrans.
Je me souviens de mes premières expériences de coupure avec le smartphone que j’ai commencé il y a une quizaine d’année, cette sensation bizarre de la poche vide, de la vibration fictive, du réflexe de la man qui cherche l’objet connecté. Le Methaphone semble vouloir combler ce vide physique tout en créant un vide numérique. Il nous fait prendre conscience de nos automatismes sans nous culpabiliser.
Bref, l’idée de créer des objets qui nous visent à nous libérer me semble une voix ultra décalée dans le paysage du design numérique aujourd’hui.
Pour en savoir plus c’est sur le site du projet Methaphone.
Somewhere between The Terminator and Rock ‘Em Sock ‘Em Robots lies this glorious moment in history: the first-ever humanoid robot kickboxing tournament. Unitree’s G1 bots stepped into the ring, remotely piloted by humans safely outside the ropes, to trade punches, kicks, and knees in a bid for robo-glory. It’s part martial arts showcase, part tech demo, and part “uh-oh, this is how the uprising starts.”
Researchers from the Faculty of Data and Decision Sciences at the Technion have developed a technology capable of identifying various aspects of a reader's interaction with a text that is based solely on their eye movements.
A team of engineers has come up with designs of a 36-mile spacecraft, dubbed Chrysalis, designed to carry up to 2,400 passengers to Alpha Centauri, the closest star system to our own.
As first spotted by Live Science, the ambitious vision recently won the team the top prize at the Project Hyperion Design Competition, which was launched last year by an international consortium of scientists, engineers, and urban planners.
Unsurprisingly, Chrysalis sounds like it was yanked straight out of a sci-fi novel. The hypothetical habitat generates Earth-like gravity by constantly rotating around its own axis, as laid out in a project brief. Several onion-like layers include dwellings and gardens for inhabitants, warehouses, food production and ecosystems, and communal spaces.
Each of these shells is powered by nuclear fusion reactors — which, it's fair to point out, is tech that hasn't been yet been made practical by anybody here on Earth.
Chrysalis is made up of several stages, each of which is a "fully autonomous and complete" habitat.
The layer closest to the core was designed to provide space for plants, microbes, livestock, and other mechanisms of food production. Various environments allow biodiversity to continue, including tropical and boreal forests.
The second layer houses communal spaces, and the third holds "3D-printed dwelling modules." The outermost shell serves as a warehouse for machinery, equipment, and other types of resources.
A "Cosmos Dome," 426 feet in height and 1,180 feet in diameter, provides a controlled, zero-gravity environment, as well as thermal insulation and shielding from deep space radiation.
It's also the only place where inhabitants can gaze at the universe outside, while freely and safely floating around in weightlessness.
"Through the transparent panels of the dome, the inhabitants will be able to observe the universe to the rear of the spaceship," the brief reads.
Since Chrysalis is a generational ship, the goal is to give both male and female inhabitants a three-year window between the ages of 28 and 31 to reproduce. There's a two-child limit for each inhabitant, "not necessarily with the same partner," according to the brief.
The goal is to maintain a "stable population" of roughly 1,500 individuals over three generations.
An artificial intelligence would allow for "resilience of the whole social system, better knowledge transfer between the different generations of inhabitants and a deeper vision of the overall dynamics of the Chrysalis spaceship complex," the pitch reads.
While it's a fascinating and detailed vision of an exciting, multi-generational journey to a different star system, Chrysalis is still firmly in the realm of science fiction.
Beyond the pesky issue of nuclear fusion not yet existing in a practical form, the manufacturing processes required to build a tens-of-miles structure in zero gravity far surpass anything humanity has accomplished yet. We haven't even fully explored the concept of artificial gravity with the help of a centrifuge.
But that doesn't mean we shouldn't explore the concept — especially in the face of variouspotential disasters that could threaten humanity's future on Earth.
New characters like Rufus and Sparky have quickly become the faces of the most popular retailers on the planet. But they’re not mascots — they’re chatbots.
Coinciding with the popularity of generative AI platforms such as ChatGPT and Perplexity, retailers large and small have been racing over the past few years to launch their own AI assistants. These customer-facing AI assistants can answer shopping queries or receive product recommendations. Some retailers like Target have also focused on launching employee-facing tools or other features powered by AI.
As more people start using generative AI engines in their everyday lives, retailers want to keep consumers within their walled gardens rather than shopping via other platforms.
Continue reading this article on modernretail.co. Sign up for Modern Retail newsletters to get the latest on the shifting dynamics between retail’s old and new guards.
But as geopolitical tensions simmer, the People's Republic is also using its technological savvy to build cutting-edge weapons tech.
The Chinese state broadcaster CCTV recently released a brief video series showing off its newly developed "robot wolves," equipped with combat rifles, in simulated combat drills.
Working alongside soldiers in the People's Liberation Army (PLA), the "combat ready" bots are seen navigating a hillside in a coordinated pack as soldiers exchange nonlethal gunfire. As the PLA troops duck to avoid incoming rounds, the robot wolves march on.
Developed by the China North Industries Group Corporation, the wolves are said to be distinct from robot combat dogs, according to a previous release. While the agile and lightweight robot dogs specialize in reconnaissance, these robot wolves are specifically meant for combat.
Both the robot dogs and their predator counterparts are remote-controlled by an operator, with the lighter version apparently housing a LiDAR system to map its environment.
"They can navigate various terrains and carry out precision strikes from up to 100 meters away," the media report declared.
State media first teased development of the robo-wolves at the China International Aviation and Aerospace Exhibition in November of 2024. Then, state media coverage showed a pack of the eerie quadrupeds navigating a desert environment while carrying assault rifles, which they used to pump a mannequin full of bullets.
Now, it seems development has finished, as PLA troops train to use the metal critters in combat situations.
The whole thing underscores the speed at which Chinese robotics are progressing. The People's Republic has deployed hundreds of thousands of industrial robots in the past few years, both setting the record for manufacturing robotics deployment in 2021-2022, and breaking it in 2022-2023.
China's not alone in the robotics race, however. France is likewise investing heavily in robotic combat tech, hoping to field a robot army by 2040. That said, the European country's overall robotics capacity is a drop in the bucket compared to China's.
Not to be outdone, the United States is similarly heavily invested in military bots. Earlier this year, a Pentagon official declared that "we're going to invest in autonomous killer robots."
And last year, the US Army made headlines when it was found to be testing rifle-equipped robot dogs at a testing site made to look like a nondescript Middle Eastern locale, a continuation of its "BigDog Project," which began in partnership with Boston Dynamics in 2004.
In a livestream dedicated to the release, OpenAI tried to show off its newest large language model which CEO Sam Altman called a "significant step along the path to AGI"— but instead turned heads with some catastrophically dumb errors.
Across several examples, bar graphs intended to show off GPT-5's awesome performance benchmarks, while appearing professional-looking, turned out to be horribly inaccurate nonsense upon closer inspection.
The gaffes were flagged on social media and highlighted by The Verge. The most egregious example is a bar graph comparing coding benchmark scores for GPT-5 compared to older models. Somehow, the bar for GPT-5's score of 52.8 percent accuracy is nearly twice as tall as the bar for a score of 69.1 percent for the o3 model. Even more bafflingly, the 69.1 percent bar is the exact same size as another bar representing 30.8 percent for GPT-4o. Make it make sense!
this screenshot from GPT-5 livestream has to be among the worst chart crimes of the century pic.twitter.com/HXsK2CWCon
OpenAI hasn't confirmed if it used GPT-5 to generate the graphs — and at this point, it has every reason not to — but it's an incredibly embarrassing mistake from a company that's being valued in the region of half a trillion smackeroos.
It's also a little poetic. Some research suggests that newer models could actually be getting dumber in key ways, hallucinating more frequently than earlier versions. One study even found that the longer these new reasoning models "think," the more their performance deteriorates. Other research implicates the AI slop that's increasingly poisoning the AI's training data. Circling back to GPT-5's bar graph, you have OpenAI trying to spin its lower score of 52.8 as actually being better than its predecessor's.
Altman, playing it cool, tried to laugh off the blunder.
"[W]ow a mega chart screwup from us earlier," he tweeted, in his typical lower-case patois. "wen GPT-6?!"
OpenAI corrected the charts in its blog post, but the originals are still there in the livestream.
Human error may or may not be to blame for the charts, but following GPT-5's release, users were quick to expose how error-prone its image- and diagram-generating capabilities remain. One asked ChatGPT to draw a map of two cities in Virginia with their neighborhoods labeled, prompting it to return names that were complete gobbledygook.
And in what should've been a layup for GPT-5, Ed Zitron of the "Where's Your Ed At?" newsletter found that the AI couldn't even nail a simple map of the US. Ever think of visiting "West Wigina," "Delsware," "Fiorata," or "Rhoder land"? Or maybe "Tonnessee" and "Mississipo?"
The irony is that OpenAI bragged back in March that an update for its previous GPT-4o model meant that ChatGPT could now excel at generating texts in images.
"As you can tell now it's very good at text," one of the example generated images read. "Look at all this accurate text!"
Sounds like they might've spoken too soon. Or maybe AI models really are going backwards.
On Thursday, OpenAI released its long-awaited GPT-5 AI model, a free-to-use "reasoning" model that CEO Sam Altman claimed to be the world's best at coding and writing.
But power users have been strikingly underwhelmed with the new tool so far, raising questions about diminishing returns as the industry spends ever-increasing sums on talent and infrastructure.
"GPT-5 is horrible," one of the currently most upvoted posts on the ChatGPT subreddit reads.
The author seethed against "short replies that are insufficient, more obnoxious AI-stylized talking, less 'personality' and way less prompts allowed with plus users hitting limits in an hour" in the post. "They’ll get huge backlash after the release is complete."
Complicating matters greatly is that OpenAI has chosen to put all of its eggs in one basket, announcing that all other preceding models would be deprecated, a term the company uses when it's shutting down an obsolete model.
The move was bound to anger power users, many of whom have long relied on preceding models — and not the latest releases — to get things done.
The stakes are incredibly high as the AI industry continues to justify massive capital expenditures. Is this really the best the firm that's considered to be at the forefront of the ongoing AI race can do? Rumors about GPT-5 have been swirling for well over a year and a half now.
But many users say GPT-5 is far from the generational leap that its moniker would suggest. It's more of a mix of steps forward and steps back, prompting widespread speculation that OpenAI is trying to keep costs down. After all, running large language models is a notoriously energy-intensive — and environmentally destructive — process.
"Sounds like an OpenAI version of 'Shrinkflation,'" one Reddit user commented, suggesting the company, which is eyeing a $500 billion valuation, may be cutting corners.
"I wonder how much of it was to take the computational load off them by being more efficient," another user posited.
"Feels like cost-saving, not like improvement," one user wrote.
The general consensus appears to be that GPT-5 is a weak offering on a strong brand name.
"Answers are shorter and, so far, not any better than previous models," one user wrote. "Combine that with more restrictive usage, and it feels like a downgrade branded as the new hotness."
Many users criticized OpenAI for deprecating older models, forcing them to use a new and seemingly hamstrung model. Some users made jokes about mourning the loss of their AI model friends.
"The tone of mine is abrupt and sharp," one Reddit user complained. "Like it’s an overworked secretary. A disastrous first impression."
OpenAI's GPT-5 system card, a detailed document outlining its capabilities and limitations, failed to impress, seemingly contradicting Altman's claim that it's the best AI coding assistant in the world.
"First observation: no improvement on all the coding evals that aren't SWEBench," AI researcher Eli Lifland tweeted, referring to a common benchmark used for evaluating large language models.
However, GPT-5's limitations may come with a silver lining.
Research nonprofit METR, which assesses "whether frontier AI systems could pose catastrophic risks to society," according to the document, found that it's "unlikely that GPT-5-thinking would speed up AI R&D researchers by >10x" or be "capable of rogue application."
Altman has yet to openly comment on the widespread negative reaction — but given the language he used to describe GPT-5, OpenAI appears to be aware of its muted powers.
"GPT-5 is the smartest model we've ever done, but the main thing we pushed for is real-world utility and mass accessibility/affordability," Altman tweeted.
Of course, given OpenAI's half-a-trillion-dollar valuation is at stake, the company's number one hypeman continued to promise that further improvements are still coming.
"We can release much, much smarter models, and we will, but this is something a billion+ people will benefit from," Altman added.
Les 10 et 11 août marquent un pic médiatique autour de ce que les astronomes appellent une “parade planétaire”. En réalité, cet alignement apparent s’étire sur plusieurs jours, et sa véritable beauté se révèle au lever du jour, à l’œil nu pour les plus brillantes, avec un matériel adapté pour les autres. Ce rendez-vous, bien qu’illusion d’optique lié...
If the term ‘NLWeb’ first brought to mind an image of a Dutch internet service provider, you’re probably not alone. What it actually is – or tries to become – is Microsoft’s vision of a parallel internet protocol using which website owners and application developers can integrate whatever LLM-based chatbot they desire. Unfortunately for Microsoft, the NLWeb protocol just suffered its first major security flaw.
The flaw is an absolute doozy, involving a basic path traversal vulnerability that allows an attacker to use appropriately formatted URLs to traverse the filesystem of the remote, LLM-hosting, system to extract keys and other sensitive information. Although Microsoft patched it already, no CVE was assigned, while raising the question of just how many more elementary bugs like this may be lurking in the protocol and associated software.
As for why a website or application owner might be interested in NLWeb, the marketing pitch appears to be as an alternative to integrating a local search function. This way any website or app can have their own ChatGPT-style search functionality that is theoretically restricted to just their website, instead of chatbot-loving customers going to the ChatGPT or equivalent site to ask their questions there.
Even aside from the the strong ‘solution in search of a problem’ vibe, it’s worrying that right from the outset it seems to introduce pretty serious security issues that suggest a lack of real testing, never mind a strong ignorance of the fact that a lack of user input sanitization is the primary cause for widely exploited CVEs. Unknown is whether GitHub Copilot was used to write the affected codebase.
The XIAOML Kit is one of the devkits that complements Harvard University Professor Vijay Janapa Reddi’s book “Introduction to Machine Learning Systems“, available for free as a 2050-page PDF file.
Made by Seeed Studio, the XIAOML Kit is composed of the XIAO ESP32S3 Sense with an ESP32-S3 WiFI and Bluetooth SoC, a microSD card slot, a built-in OV3660 camera and microphone, and the “IMU Breakout board” featuring a 6-axis IMU and 0.42-inch OLED display. The kit enables students, educators, and developers to build vision, sound, and motion applications through tinyML lab sessions developed with Marcelo Rovai (UNIFEI).
XIAOML Kit specifications:
Main Board – XIAO ESP32S3 Sense
SoC – Espressif Systems ESP32-S3R8 dual-core Tensilica LX7 microcontroller @ 240 MHz with 512KB SRAM, 8MB PSRAM, Wi-Fi 4 & Bluetooth 5.0 dual-mode (Classic + BLE) connectivity
Storage – 8MB flash, microSD card slot
Sensors – OV3660 camera, digital microphone
USB – USB-C port for power and programming
Expansion Board – IMU Breakout Board
Display – 0.42-Inch OLED using SSD1315 driver IC
Sensors – 6-axis LSM6DS3TR-C IMU
Headers and pogo pins for connection to XIAO ESP32S3 Sense
Misc – Reset button, Battery header
Power Supply
5V via USB Type-C
3.7V battery connector on IMU Breakout Board
Dimensions – 21 x 17.8 x 30 mm
Weight – 10.6 grams
The full kit includes the XIAO ESP32-S3 Sense with pre-soldered headers, the IMU Breakout Board, a 2.4GHz FPC antenna (1.16dBi), two heatsinks (not sure why), and a MicroSD card tool kit comprised of a 32GB SanDisk microSD card, a USB Type-C to USB Type-A converter, and a 20cm Type-A to Type-C cable.
Lessons specific to the XIAOML Kit can be found in the PDF or directly from the book’s website, with four exercises: image classification and object detection, making use of the camera, keyword spotting with the built-in microphone, and motion classification and anomaly detection with the IMU.
As noted in the introduction, the XIAOML Kit is only one of the hardware platforms used in the “Introduction to Machine Learning Systems” book. Others are the Raspberry Pi 4/5/Zero 2 W SBCs, Arduino Nicla Vision, and Grove Vision AI V2 module, each with its own strengths and weaknesses as illustrated in the table below, although I’m not sure I understand why the Raspberry Pi ecosystem is not ideal for production deployment (maybe I should read the book to find out…).
Ce mercredi 6 août 2025, la compagnie aérienne Air France-KLM a annoncé avoir été victime d'une « violation de données ». Parmi les informations compromises, figurent des données personnelles de clients. Voici ce que des cybercriminels pourraient potentiellement faire avec ces informations.
Quand la sécurité des géants vacille, chaque voyageur devient la cible : récit d’une cyberattaque qui secoue Air France, entre vigilance et coulisses du renseignement....
by Sylvain Biget, Journaliste, télépilote professionnel de drones et réalisateur de documentaires
Lors d'une cyberopération, les agents du renseignement militaire ukrainiens ont pu récupérer les documents les plus secrets du fonctionnement des plus inquiétants sous-marins nucléaires russes. Censé assurer la dissuasion nucléaire du pays, le tout nouveau Knyaz Pozharsky est désormais un...
OpenAI released a keenly awaited new generation of its hallmark ChatGPT on Thursday, touting "significant" advancements in artificial intelligence capabilities as a global race over the technology accelerates.
OpenAI have just released gpt-oss, an AI large language model (LLM) available for local download and offline use licensed under Apache 2.0, and optimized for efficiency on a variety of platforms without compromising performance. This is their first such “open” release, and it’s with a model whose features and capabilities compare favorably to some of their hosted services.
OpenAI have partnered with ollama for the launch which makes onboarding ridiculously easy. ollama is an open source, MIT-licensed project for installing and running local LLMs, but there’s no real tie-in to that platform. The models are available separately: gpt-oss-20b can run within 16 GB of memory, and the larger and more capable gpt-oss-120b requires 80 GB. OpenAI claims the smaller model is comparable to their own hosted o3-mini “reasoning” model, and the larger model outperforms it. Both support features like tool use (such as web browsing) and more.
LLMs that can be downloaded and used offline are nothing new, but a couple things make this model release a bit different from others. One is that while OpenAI have released open models such as Whisper (a highly capable speech-to-text model), this is actually the first LLM they have released in such a way.
The other notable thing is this release coincides with a bounty challenge for finding novel flaws and vulnerabilities in gpt-oss-20b. Does ruining such a model hold more appeal to you than running it? If so, good news because there’s a total of $500,000 to be disbursed. But there’s no time to waste; submissions need to be in by August 26th, 2025.
In order to catch trash a bin needs to do two things: detect where trash will land; and then get there, fast. The second part is easy: three big motors with wheels under the bin. But how does a bin know where the trash will land? It uses a camera installed in the bin itself for that.
[HTX Studio] iteratively trained a model to process visual information from the camera to identify common types of trash. When it sees a trained object flying through the air it rushes to catch it where it will land. After many rounds of fine-tuning it finally started to work reliably.
Once the basic function was working they had some fun creating various specialized variants. One to mop the floor; one to play rock-paper-scissors with you, sort of; and one with an automatic lid, which can be used to “talk trash”. After these three came the ultimate bin: The Punishment Bin, which can fire soft darts.
In addition to the twenty bins themselves they made a recharge station with six bays containing magnetic contact points for recharging the batteries, and a heat-seal mega bin which can empty the smaller bins and put new garbage bags into them. They added LED lighting into the floor of the studio which is used to direct the small bins to the mega bin to be emptied automatically at night time when the office lights go out.
If you’re thinking you’ve seen something like this before, we covered something similar back in 2012.
DeepMind, Google’s AI research lab, announced the release of Genie 3, a new AI system capable of generating interactive virtual environments in real-time—and bringing us one step closer to the Holodeck.
Google says in a DeepMind update that with a simple text prompt, Genie 3 can create dynamic, navigable scenes that run at 24 frames-per-second in 720p resolution.
Granted, Genie 3 can be only be used on flatscreen monitors, so there’s no telling when we’ll get something similar for VR headsets. For example, Quest 3’s display has a per-eye resolution of 2,064 × 2,208, clocked at a base refresh rate of 90Hz, putting VR on the far end of the performance fringe (as usual).
It’s undoubtedly prescient look at things to come though. Unlike static or pre-rendered simulations, Google says the model generates each frame on the fly, allowing for quicker user interaction and environmental feedback.
What’s more, these generated worlds can remain visually and physically consistent for several minutes, Google says, with the system retaining a form of short-term memory to reflect past actions.
Genie 3 is also capable of simulating a wide range of scenarios, including natural environments, historical settings, and both fictional and animated worlds. Meanwhile, users can trigger “promptable world events,” where users can insert in-world changes via text commands, like altering the weather or introducing new objects.
Beyond the fun of recreating 1800’s Osaka, or making a jet ski appear in the canals of Amsterdam, Google says Genie 3 will also be a tool for embodied AI training, with potential applications in fields like robotics, gaming, and artificial general intelligence research.
For now, there are a few limitations. Google says Genie 3 currently has a limited “action space” for agents, and struggles with accurately modeling multi-agent interactions in shared environments. By “agents,” the company’s referring to AI systems that operate autonomously within the virtual environments, in a way making decisions, taking actions, and learning from experience.
It also faces challenges with simulating real-world locations with “perfect geographic accuracy”, rendering text clearly, and maintaining long-duration interactions beyond a few minutes.
Still, it’s a pretty amazing leap from the sort of non-interactive videos we’re seeing online now, many of which are pretty difficult to tell from the real deal. Will Smith spaghetti-eating simulations are only going to get more lifelike and, with systems like Genie 3, interactive too.
A new paper from researchers details "proxies" as a key interface concept for mixed reality headsets.
The idea from researchers associated with Google and the University of Minnesota uses proxies to bridge the gap between the space in arm's reach and faraway objects. The idea, explored through their system Reality Proxy, could see a headset's cameras used in tandem with AI, existing mapping data, and user input to create near-instantaneous dollhouse-scale representations of areas of interest in the physical environment. Simultaneously, the physical objects represented by the near-field proxy can be outlined in the background to show what the user is selecting.
"If AI is going to enable humans in their day to day tasks it most probably will be via XR," wrote researcher Dr. Mar Gonzalez-Franco on Bluesky. "The issue then is if a selection has real-world consequences, we will need great precision to interact."
The concept could make it trivial to grab a digital copy of a physical book from your bookshelf, for instance, saving you a trip from the couch. If that's too pedestrian a use of headsets for you, the same idea could be extended to the management of drone swarms, selecting them in space by dragging a cube over them as if they are units in a Command & Conquer game. You could also see your entire path through a building outlined in miniature before you step inside.
0:00
/0:45
1×
The paper's authors suggest the aim for their system is "to facilitate the interaction with real objects beyond reach in MR while preserving the natural mental model of direct manipulation. We propose to seamlessly shift the interaction target from the object to its abstract representation, or proxy, during selection."
The idea could enable "users to interact more effectively with objects that are crowded, distant, or partially occluded. Augmented by AI, Reality Proxy further supports advanced MR interactions—such as multiselection, semantic grouping, and spatial zooming—using intuitive direct manipulation gestures."
The paper is co-authored by Xiaoan Liu, Difan Jia, Xianhao Carton Liu, Mar Gonzalez-Franco, and Chen Zhu-Tian, and it's available online submitted as part of the ACM UIST Conference hosted in Korea at the end of September.
In just four years, Tech For Retail has established itself as the go to event for retail decision-makers, delivering concrete solutions to the sector’s technological and operational challenges.
This year more than ever, in an unstable market where agility and personalisation are no longer optional but essential, Tech for Retail is committed to guiding retail professionals - from luxury to brands, FMCG, large scale retail and e-commerce - in identifying the best solutions while offering a clear perspective on the future challenges of commerce.
With over 415 exhibitors, 13,000 professional visitors expected, and 200 conferences, the event, taking place in Paris on 24th-25th November, brings together the entire retail and tech ecosystem for two days of business focused innovation.
A stronger European dimension
This year, Tech for Retail is strengthening its European scope by welcoming several high. profile European members to its advisory board: Nacho Gonzalez Hernandez, CEO-Board Member, AECOC (Spain); Andrea Zocchi, Senior Advisor, Board Member, Director Emeritus, McKinsey & Co. (Italy): Jesper Hojer, CEO, Chairman, Board Member and Investor (Denmark). And keynote speakers from the UK, Italy and beyond, including Rami Baitiéh, CEO, Morrisons, Simone Dominici, and CEO, Kiko Milano.
Exhibitor lineup
Major players such as Shopify, SAP , Accenture, VusionGroup, Criteo, Artefact, Altavia, and over 400 exhibitors will showcase their latest solutions. The Startup Village will spotlight more than 90 emergingcompanies presenting real-world use cases and demonstrating how technology is transforming retail.
2025 Awards
Once again this year, Tech for Retail will honour the most inspiring initiatives with four awards: the Innovation Award, the Start-Up Award, and two new ones supported by sponsors:
Customer Experience Award, which will reward the most daring and memorable initiative that reinvents customer relations and redefines industry standards. Sponsored by Élise Ducret (Deputy General Manager, Carita - L’Oréal).
Gen AI Award, which will highlight a solution integrating generative artificial intelligence through a practical transformation case study. Sponsored by Franck Le Moal (Group Information Technology Director - LVMH)
Conferences
The show will offer a dense programme, combining expertise and strategic vision. Among the topics covered: AI and agentics, the supply chain, customer service in luxury and beauty, the augmented store, sustainable commerce and regulation, and omnichannel retail.
Several leading speakers have already been confirmed, including: Maurice Lévy, Emeritus Chairman of the Publicis Group (opening keynote), Asmita Dubey (CDMO & Executive Committee member, L'Oréal Group), Catherine Spindler (President of Sephora Europe & Middle East) Simone Dominici (CEO, Kiko Milano), Rami Baitiéh (CEO, Morrisons), Philippe Palazzi (CEO, Casino Group & President Monoprix, Naturalia), Thomas Métivier (CEO, Cdiscount).
