I realized something mildly depressing last week while packing for a shoot: I now carry more USB-C accessories than actual personal items.
Yep, in my bag you’ll find a laptop charger, SSD, hub, SSD cable, and hub cable. There’s also a backup cable because one of the other cables inevitably decides to stop behaving at the worst possible moment. Somewhere in there is a dongle I no longer recognize but that I carry purely out of superstition.
This is supposedly the simplified USB-C future.
So when Digiera sent over its 2-in-1 Portable SSD, I honestly wasn’t expecting much. Products that try to combine multiple things usually end up being annoying in at least one direction. Either the SSD part is good and the hub part feels cheap, or the hub works fine and the storage feels like an afterthought.
Two weeks later, it’s still in my bag. Mostly because it quietly replaced three other things.
The built-in connector is the whole point
There’s no cable. A foldable USB-C connector lives inside the body of the drive — you flip it out, plug it directly into your phone, tablet, or laptop, and that’s it. Done. No short SSD cable dangling off the bottom of your phone while you try not to bump it loose mid-transfer.
I’ve mostly used it with an iPhone 16 Pro, and the difference from a conventional cabled SSD is immediately obvious. The drive just sits there, plugged straight into the device. No twisted cable. Nothing sliding around. The hinge feels solid too, which matters — I’ve used enough flash drives to know that “built-in connector” can mean “future warranty claim.” This one clicks firmly into place and folds flush when you’re done.
It’s also a mini USB-C hub
This is the part that sounded gimmicky and turned out to be the actual reason I kept using it. While the drive is plugged into your phone or laptop, you can run accessories and 35W passthrough power through it at the same time. During an actual shoot setup, I had:
the SSD plugged into the iPhone
power running through it to charge the phone
a mic connected
ProRes footage recording directly to the drive
All at once. No disconnects. No “accessory unsupported” warnings. No power throttling halfway through.
That’s the part that earned the bag space. Plenty of USB-C accessories technically support multitasking until you actually try it — then one port flakes out, or charging slows to a crawl, or the SSD drops the connection at the worst possible moment. This one just kept working.
The 35W passthrough specifically matters more than I expected. Recording externally to an SSD drains a phone battery fast. Being able to charge the phone through the drive while filming meant longer sessions without juggling cables, and the drive didn’t turn into a hand warmer during the process.
Cross-platform is the actual superpower
The Magnetic SSD is great if you live on an iPhone. The 2-in-1 is what you want if you live across devices. Same drive, no setup change:
iPhone 15/16/17 Pro — plug in, record ProRes directly to it
iPad — plug in, edit footage off it
MacBook — plug in via the same connector, dump files
Android — same connector, works the same way
For anyone who shoots on a phone and edits on a laptop, or who’s tired of carrying a hub and a drive and the cables to connect both, this is the consolidation play. One device, one plug, every USB-C-port-having device you own.
Speed is fine. Convenience is the point.
You’ll get the usual 2000MB/s read / 1800MB/s write on a USB 3.2 Gen 2×2 port. On a standard USB-C Gen 2 port — what most people actually have — you’ll see closer to 1000MB/s. Either way, ProRes transfers feel instant.
Under the hood it’s the same hardware story as the rest of the lineup: 3D NAND TLC memory, aluminum casing that doubles as a heat sink, 1500G shock resistance, 3-year limited warranty. None of that is the headline. It’s just the floor.
Who it’s for
Skip this if you just need basic file storage at a desk. Get one of the cheaper magnetic drives in the lineup instead.
This one earns its price if you:
shoot on a phone and edit on a laptop
film with external mics, lights, or follow-focus rigs that need their own power
already carry both a hub and an SSD in your bag and resent it
need to charge your phone while recording externally to a drive
The pitch is genuinely “less stuff,” which is rare. Most gadgets add to the pile.
Specs, price, link
Capacities: 1TB / 2TB Colors: Black, White Speeds: Up to 2000MB/s read, 1800MB/s write (USB 3.2 Gen 2×2) Hub: Built-in foldable USB-C connector + extra USB-C port with 35W passthrough power Compatibility: iPhone 15/16/17 Pro, iPad, Mac, Android, Windows Build: Aluminum casing, 1500G shock resistance, 3D NAND TLC Warranty: 3-year limited
Available at digieraglobal.com, $188.99 for the 1TB and $283.49 for the 2TB. If the contents of your tech pouch have started embarrassing you, this is the easiest single swap I’ve found.
Although parents and teachers like to point out the deep link between cooking and chemistry, most people don’t deliberately apply any chemical principles beyond acid/base reactions to their recipes. Not so [Ben Kazez]: he’s written a thorough exploration of the chemical journey to the perfect pancake, and made a calculator for others to use with their own ingredients.
The goal is to optimize the pancakes along four dimensions: interior texture (light and smooth), a tangy flavour, rise, and a crisp, brown exterior layer. The tang comes from residual acids, and since lactic acid produces the best taste, dairy-based acid sources (such as Greek yoghurt or buttermilk) are preferable. Acids also react with baking soda to release carbon dioxide, making them a part of one of the four rising agents. The other three are carbon dioxide released when double-acting baking powder is heated, steam released from the batter, and air bubbles stabilized by egg white foam.
Dairy products, besides contributing acid, also provide a protein structure to keep the interior smooth. In a normal wheat-heavy pancake, two proteins (glutenin and gliadin) interact to form tough strands of gluten. Fats bind to hydrophobic amino acids in these proteins and shorten the gluten chains, hence the name shortening. Adding ricotta cheese also replaces some of this gluten network with a smoother structure of previously-denatured dairy proteins. Dairy products also contribute to the Maillard reaction between reducing sugars (such as lactose, glucose, and fructose) and amino acids, which causes the browning of the pancake’s surface. Besides being brown, the surface should be crisp; since amylose, found in corn starch, forms a brittle, glassy, crackly network when dehydrated, corn starch was added.
The result is a set of chemical equations which can be tuned to create perfect pancakes, combined in the calculator. This summary doesn’t do justice to the depth of the research here; [Ben] also investigated optimal batter resting times, fermentation, cooking fats, cooking surfaces, and spatula properties. If all this has you interested in more about dairy proteins, check out our article on cheesemaking.
At Computex 2026 in Taipei, Taiwan, faytech Sales Director Michael Foss presented the HLD (Hololuminescent Display) series, developed in partnership with Looking Glass. The HLD series recently won the Display of the Year award from the Society for Information Display (SID) at their annual conference. These glasses-free holographic displays are engineered for retail, digital signage, and experiential interactive systems.
—
HDMI® Technology is the foundation for the worldwide ecosystem of HDMI-connected devices; integrated with displays, set-top boxes, laptops, audio video receivers and other product types. Because of this global usage, manufacturers, resellers, integrators and consumers must be assured that their HDMI® products work seamlessly together and deliver the best possible performance by sourcing products from licensed HDMI Adopters or authorized resellers. For HDMI Cables, consumers can look for the official HDMI® Cable Certification Labels on packaging. Innovation continues with the latest HDMI 2.2 Specification that supports higher 96Gbps bandwidth and next-gen HDMI Fixed Rate Link technology to provide optimal audio and video for a wide range of device applications. Higher resolutions and refresh rates are supported, including up to 12K@120 and 16K@60. Additionally, more high-quality options are supported, including uncompressed full chroma formats such as 8K@60/4:4:4 and 4K@240/4:4:4 at 10-bit and 12-bit color.
—
The product family includes 16-inch, 27-inch, and large-format 86-inch versions, which are already in production or entering deployment phases. In addition to these sizes, faytech is developing 43-inch and 55-inch versions to expand the lineup. All initial units, including the 16-inch, 27-inch, and 86-inch models, feature native 4K display panels to output high-fidelity content.
From a hardware and integration standpoint, the HLD series supports various external media players, including single-board computers like the Raspberry Pi, as well as standard Android devices and Windows-based PCs. The displays support both native 2D and 3D content. For media that is not originally designed for light-field or holographic playback, the system can upscale the content to render it in 3D.
Initial manufacturing runs of the large 86-inch display, which started with a mass production batch of 20 units for global distributors, have sold out, prompting additional production runs. This commercial interest spans digital signage integrators, marketing agencies, and rental providers looking to deploy the displays at trade shows, exhibitions, and retail storefronts. Engagement is also growing among content creation and graphic design agencies that render custom professional 3D assets for brands.
Future iterations of the HLD technology aim to combine these light-field displays with interactive AI systems. While the current displays feature built-in speakers, upcoming developments include integrating camera systems and directional microphone arrays to enable real-time communication with 3D AI avatars, providing an immersive, conversational interface for retail and public installations.
La Terre traverse actuellement un nuage de débris radioactifs issus d'une supernova ancienne. Des traces de fer-60, un isotope qui ne se forme que lors de l'explosion d'étoiles massives, ont été...
Une protéine présente naturellement dans nos articulations pourrait bien se révéler un allié contre l'arthrose. Des chercheurs sud-coréens ont en effet découvert que des niveaux élevés de...
[Deal du jour] Amazon casse le prix des trois sets Lego Star Wars équipés de la fameuse Smart Brick, jusqu'à 25 % de remise. De quoi tester la grande nouveauté Lego de 2026 sans payer le tarif plein.
À l’occasion de sa conférence Build 2026, Microsoft a dévoilé une famille de sept nouveaux modèles d'IA développés en interne. Cette nouvelle gamme s'étend notamment de la génération d'images à la gestion de la voix.
La prochaine révolution de l'assurance automobile est désormais engagée : le constructeur chinois BYD offre une garantie inédite aux utilisateurs de son système de conduite autonome, indépendamment des couvertures souscrites par ailleurs. Le changement est logique… mais l'industrie et les régulateurs sont-ils prêts à l'appréhender ?
Une première étape avait été franchie l'an dernier avec une protection similaire lors de l'usage de la fonction de parking automatique, avec laquelle les risques étaient évidemment limités. Cette fois, tout accident intervenant pendant que l'option « God's Eye » de conduite assistée en milieu urbain est activée sera intégralement pris en charge, sans coût supplémentaire, pour les dommages au véhicule et à ceux des tiers impliqués ainsi que pour la responsabilité civile, sans limitation de montant et sans impact sur la prime d'assurance du propriétaire. L'offre est toutefois limitée à un an.
Pour BYD, l'enjeu est avant tout commercial. Dans un contexte de concurrence exacerbée sur ses terres natales (qui s'étend actuellement en Europe), l'ajout représente un facteur de différenciation, complété d'ailleurs par un tarif agressif (de l'ordre de 1 500 €) pour son module d'assistance. Il constitue en outre un instrument de promotion de ce dernier, alors que ce genre de système souffre encore d'une certaine méfiance parmi le grand public. Et la recette semble réussir à merveille : le recours au parking automatique est passé de 21% à 93% du parc après l'opération précédente.
En coulisses, le constructeur a de quoi tenir sa promesse sans trop s'inquiéter pour ses finances. Ses plus de 3 millions de voitures équipées de dispositifs intelligents (enrichis régulièrement) parcourant quelques 200 millions de kilomètres chaque jour en mode autonome lui procurent une masse considérable de données sur lesquelles ses équipes, composées de 5 000 ingénieurs, œuvrent en vue d'affiner les algorithmes embarqués et visent un objectif ultime de zéro accident (ou presque, soyons réalistes).
Indépendamment de ces considérations, et des préoccupations sociétales vis-à-vis de la sécurité routière (objectivement renforcée par l'automatisation), la démarche de BYD apporte un début de réponse à une problématique qui devrait rapidement devenir centrale avec l'expansion des voitures entièrement autonomes. En effet, dans cette situation, l'entité assurée ne peut plus être, comme aujourd'hui, le conducteur, puisqu'il ne porte évidemment aucune responsabilité dans les éventuels sinistres. BYD laisse entrevoir la possibilité qu'il assume le risque… mais ce n'est pas la seule possibilité.
Au Computex 2026, le PDG de Qualcomm a déroulé une vision radicale de l'informatique agentique : des agents IA omniprésents, embarqués sur chaque appareil, qui vous suivront partout.
Le 4 juin, le Parlement européen remplacera Google par Qwant comme moteur de recherche par défaut sur ses ordinateurs. Une bascule qui intervient au moment où les institutions européennes multiplient les initiatives pour réduire leur dépendance aux logiciels américains… et où les grands groupes américains s'inquiètent de la politique européenne.
Un bricoleur connu sous le nom de Startup Chuck a eu une idée que personne ne lui avait demandée : fabriquer des cookies de A à Z avec son imprimante 3D. Pas seulement les façonner. Les cuire aussi, dans la machine. Le tout documenté dans une vidéo YouTube, évidemment.
Tout commence par l'attirail. Chuck a imprimé en plastique l'ensemble du matériel du pâtissier, comme s'il montait une vraie petite chaîne de production de biscuits.
Et il ne fait pas semblant. Il parle carrément d'une production sérieuse de cookies, avec de quoi tout faire de la première à la dernière étape sans sortir un seul ustensile de cuisine du commerce.
Au menu, un bol à mélanger, des doseurs, un fouet compatible avec un robot KitchenAid, et une spatule équipée d'une lame en TPU, ce filament souple et un peu caoutchouteux qui plie sans casser.
Le plateau de cuisson, lui, est sorti en filament nylon, choisi pour mieux encaisser la chaleur que le plastique habituel.
Reste l'étape qui intrigue vraiment, la cuisson elle-même. L'imprimante est un modèle fermé, avec un caisson tout autour, et Chuck la détourne en mini-four basse température en se servant du plateau chauffant (le lit, en jargon de l'impression 3D) comme résistance pour chauffer la pâte.
Le résultat ? Mitigé. Les cookies sont parfaitement reconnaissables, on voit bien que ce sont des cookies, sauf qu'ils ne dorent jamais. Un lit chauffant plafonne à quelques dizaines de degrés, on est très loin des 180 °C d'un vrai four.
Bref, on récupère une pâte cuite mais pâlotte, plus proche du biscuit mou que de la gourmandise croustillante et dorée dont on rêvait.
Notez aussi que glisser de l'humidité et des ingrédients alimentaires dans une machine prévue pour du plastique, ce n'est une bonne idée ni pour vos futures impressions, ni pour votre estomac.
Le procédé en question, le FDM (le dépôt de fil fondu, couche par couche), laisse de minuscules rainures un peu partout sur les pièces imprimées. L'endroit rêvé pour piéger des résidus de pâte et laisser quelques bactéries s'installer tranquillement entre deux fournées.
On est en plein dans cet esprit bidouille où l'imprimante 3D sert à tout sauf à ce pour quoi elle est faite, juste pour voir si la chose est possible. Et Chuck est le premier à le reconnaître, son truc est une démonstration pour s'amuser, pas une recette à reproduire chez soi, même si votre four est en panne et que vous avez une mega fringale.
Are AI models conscious, and if not, could they be in the near future? The possibility is far-fetched, but AI companies seem to feel it’s in their best interests to keep the question as open-ended as possible.
Now, the Financial Times reports that three of the industry’s top dogs — Anthropic, Google’s AI lab DeepMind, and Meta — have all hired experts in fields like psychology, philosophy, and ethics to pursue research into machine consciousness and AI welfare.
Anthropic, which has arguably done the most out of the bunch to anthropomorphize its models and play up the AI consciousness angle — its chatbot has the human name of “Claude,” after all — has been testing its models for behaviors that resemble “panic” and “anxiety,” per the reporting, and is pursuing “model welfare research” to explore whether AI models might have experiences that matter morally.
“We remain deeply uncertain about this, but we think the question is serious enough to study carefully as AI systems get more capable,” the company said in a statement.
DeepMind, meanwhile, has hired University of Cambridge researcher Henry Shevlin as a philosopher working on machine consciousness, human-AI relationships, and AGI readiness, per the reporting. (Earlier this year, Shevlin sparked a wave of discourse in online AI circles after sharing his stunned reaction to an email he received from an AI agent.)
DeepMind ethicist Iason Gabriel, who leads the lab’s AGI and society team, called the question of AI consciousness “very complicated,” and described AI as “highly capable cognitive agents that are also just very deeply different from human beings and even from animal consciousness.”
These weighty claims are disputed by many scientists and AI researchers. But the FT, in seeking a counterargument to round out its reporting, quotes an expert who makes claims that ascribe a questionable degree of humanlike agency to chatbots. “[AI models] have goals, they can deceive, they can hide what their true interests are,” Susan Schneider, director of the Center for the Future of AI, Mind and Society, told the newspaper. But she added it’s “entirely scientifically possible that they’re doing this without having the felt quality of experience, which is what consciousness is.”
Certainly, the possibility of AI consciousness shouldn’t be completely dismissed out of hand. But neither should alien civilizations, which are generally treated more as a sci-fi musing than an urgent existential issue.
Moreover, we should be skeptical when most of the noise on this topic is coming from the industry itself. Anthropic CEO Dario Amodei has repeatedly dangled the possibility of AI consciousness in interviews. And his company’s research frequently makes bold claims about their models showing humanlike behavior, such as supposedly harboring “emotions.” Just remember that it’s easier for AI companies to string us along with wild Skynet doomsday scenarios instead of confronting the tech’s far mundane consequences currently playing out before our eyes.
En août 2026, Lego lance une nouvelle gamme de produits Pokémon liés à la Smart Brick, sa brique intelligente. L'idée est simple : animer les célèbres créatures pour se prendre pour un dresseur.
Carry extra peace of mind with the Pebblebee Halo. It’s a compact personal safety alarm designed for everyday situations.
Pull-to-Activate Siren and Strobe: Pulling the device apart triggers a loud siren and bright strobe light, drawing immediate attention and disorienting a potential threat without any button-pressing under pressure.
Silent Panic Button: Separate panic button sends an instant alert and your live location to up to 5 trusted contacts without activating the siren, useful when making noise could put you at greater risk.
Live Location Sharing: Shares your real-time location with your Safety Circle as soon as you trigger an alert. Pebblebee includes a 12-month Alert Live subscription.
Dual-Network Item Finding: Support for Apple Find My and Google Find Hub allows the Halo to connect with large device networks. You can locate it even after it leaves Bluetooth range.
Built-in Flashlight and Long Battery Life: Integrated LED flashlight helps illuminate dark paths and parking areas. The rechargeable battery lasts up to 12 months per charge and uses USB-C for charging.
The Pebblebee Halo is a compact, well-built safety tool that covers alerts, tracking, and lighting in one carry-everywhere device.
Nvidia RTX Spark Unreal Engine demonstrated the laptop form factor of its GB10 Blackwell system architecture at Computex 2026. This prototype houses the same silicon architecture previously deployed in desktop small box configurations, integrating a 20-core ARM CPU and 128 GB of unified memory. The design adapts the high-performance processor architecture into a mobile envelope, addressing the thermal management and power delivery constraints inherent in portable hardware.
—
HDMI® Technology is the foundation for the worldwide ecosystem of HDMI-connected devices; integrated with displays, set-top boxes, laptops, audio video receivers and other product types. Because of this global usage, manufacturers, resellers, integrators and consumers must be assured that their HDMI® products work seamlessly together and deliver the best possible performance by sourcing products from licensed HDMI Adopters or authorized resellers. For HDMI Cables, consumers can look for the official HDMI® Cable Certification Labels on packaging. Innovation continues with the latest HDMI 2.2 Specification that supports higher 96Gbps bandwidth and next-gen HDMI Fixed Rate Link technology to provide optimal audio and video for a wide range of device applications. Higher resolutions and refresh rates are supported, including up to 12K@120 and 16K@60. Additionally, more high-quality options are supported, including uncompressed full chroma formats such as 8K@60/4:4:4 and 4K@240/4:4:4 at 10-bit and 12-bit color.
—
The laptop hardware runs the Windows operating system, marking a transition from the Ubuntu-based DGX OS environment used in previous small-form-factor iterations. This OS transition enables developers and creators to access the broader Windows application ecosystem while utilizing the unified memory architecture. The platform supports dynamic power profile adjustments, allowing users to configure performance modes through the operating system to balance thermal output, battery life, and processing throughput.
During the technical demonstration, the system ran the Unreal Engine Editor to showcase its performance in real-time 3D asset rendering and game development workflows. Historically, mobile workstations have been constrained by separate GPU and CPU memory pools, often limiting dedicated graphics memory to 16 GB or 24 GB. The 128 GB unified memory pool in this prototype allows the GPU to access the entire memory allocation dynamically, eliminating the need to optimize scenes or reduce texture resolutions to fit within narrow hardware limits.
To highlight this capability, the system loaded a 3D city scene totaling approximately 70 GB directly into the unified memory. This setup allows developers to navigate large environments smoothly without experiencing loading screens or latency from transferring assets between system memory and GPU memory. The unified memory architecture enables real-time editing and viewport navigation of complex geometry and high-resolution textures at scale.
The integration of unified memory and Windows OS also opens up possibilities for local developer workflows involving agentic AI applications. Developers can run agentic frameworks to interact with local Windows applications directly on the hardware, bypassing the OS-level application limitations of previous Linux-only setups. This configuration supports tasks ranging from automated gameplay testing in QA environments to complex game asset generation and content creation pipelines.
Vous pensiez que le billet de banque avait atteint sa forme définitive ? La Bundesdruckerei, institution bicentenaire qui imprime les documents les plus sensibles de l'État allemand, vient de prouver le contraire. Présenté le 13 mai à la Banknote Conference de Washington, STELLA est un billet noir au format carte bancaire, deuxième prototype de la sé...
Le pari de Kyle McDonald, le site dont je vais vous parler aujourd'hui c'est que si une catastrophe se prépare, les milliardaires le sauront avant nous et fileront en jet privé. C'est pour ça que cet artiste-programmeur de Los Angeles a construit l'
Apocalypse Early Warning System
, un site qui surveille en temps réel l'activité des jets privés pour repérer le moment où les riches se barrent.
Son truc écoute un réseau de récepteurs radio répartis sur toute la planète, qui captent les signaux ADS-B des avions, à savoir leur position, leur altitude, leur direction et leur identifiant. McDonald filtre alors tout ça pour ne garder que les jets privés et d'affaires, soit plus de 35 000 appareils sur la carte, puis compare le nombre en vol à un instant donné avec ce qu'on devrait attendre normalement en cas de panique.
Et ce "normalement", ce sont des années de données historiques, avec les habitudes de chacun selon l'heure, le jour et les vacances genre Thanksgiving ou Noël. Et quand l'activité grimpe au-dessus de tout ce qu'on a vu sur un an, hop, le niveau d'alerte passe à 5 sur 5, soit bien plus de décollages que d'habitude.
Par contre, ce n'est pas sur du pistage individuel à la ElonJet, avion par avion, mais sur le compteur global de toute cette flottille privée. Difficile donc de dire si tel ou tel milliardaire précis a décollé, mais ça offre une tendance générale.
L'idée lui est venue après une menace de Trump sur l'Iran, du genre "une civilisation entière" pourrait disparaître. McDonald s'est alors demandé qui serait prévenu en premier, même si en réalité, l'activité des jets de milliardaires ne prédit pas vraiment l'apocalypse, et un pic n'annonce pas forcément une catastrophe.
Et le mec est productif puisque durant les manifestations George Floyd à Los Angeles, il a aidé à pister les hélicos de la police via le trafic aérien, et a découvert que le LAPD masquait l'identité de certains appareils. Puis il a aussi sorti ICESpy et FuckLAPD, deux projets de reconnaissance faciale pour identifier des agents des forces de l'ordre. Ça lui a même valu des menaces de mort.
Tout ça, vous l'aurez deviné, repose sur de l'OSINT, c'est à dire l'art d'exploiter des données déjà publiques que personne ne prend la peine de vraiment croiser, dans la lignée de
ShadowBroker
que je vous ai présenté y'a pas longtemps.
Son tracker de jets, lui, est gratuit sur le web et avec des alertes Telegram, ou si vous voulez recevoir un SMS, c'est 5 dollars par an. N'allez pas croire que ça n'intéresse personne puisque près de 2 500 personnes ont déjà payé. C'est fou !
Bref, on est un peu entre la blague et le malaise et on peut voir tout ça en direct sur son
site
.
Are you familiar with pop tubes? Resembling the corrugated section of a bendy straw, they are at the core of PopTuber, an intriguing research project from the Actuated Experience Lab at the University of Chicago.
With five motors and specialized gears a pop tube can be formed into complex, arbitrary shapes, and just as easily reset.
PopTuber shows how five motors and some specialized gears are all it takes to bend pop tubes into complex and stable 3D shapes. One can design the shapes in software, feed a pop tube into the shaper, and watch the device do the work. Importantly, the device can just as easily reset and re-use the tube. Watch the video (embedded below the page break) to see it in action and get a feel for what it can do.
In concept, it’s a little like a wire-bending machine, although wire benders are bulkier in comparison, more complex to scale, and unbending a wire is a separate process with its own hardware.
This project explores possibilities for a machine that can crank out complex curves on demand, such as oddball user interfaces, physical prototyping, and even a strange sort of physical display. But the real forward-thinking and interesting question researchers asked is whether this idea could be a form of programmable matter. The project shows that five actuators in a relatively compact package are all that’s needed to shape (and reset) a pop tube of arbitrary length in a programmable way, and it can scale easily to different sizes.
Using pop tubes in this way reminds us of an origami-inspired method we covered for making a tube form stable, complex shapes. But it’s also different because origami- and kirigami-inspired methods bake the transformation into the material itself, whereas PopTuber can create arbitrary shapes on demand.
Un an après l'opération Spiderweb, Volodymyr Zelensky a décerné des décorations d'État aux artisans de l'attaque et publié des images inédites. L’occasion de revenir sur une opération marquante ayant frappé l’aviation stratégique russe en profondeur, ainsi que sur la doctrine qui en a découlé.
Tejas Chopra, ingénieur senior chez Netflix, a bricolé un petit logiciel appelé
Headroom
qui s'attaque à un poste de dépense devenu douloureux dans toutes les boîtes qui carburent à l'IA : la facture en tokens, ces unités que les modèles de langage facturent au passage et qui correspondent en gros à des morceaux de mots.
Son constat de départ est sévère. Près de 90% de ce qu'on balance à un grand modèle de langage, le type d'IA qui fait tourner ChatGPT, serait selon lui de la redondance pure, du remplissage que la machine paie au prix fort sans en tirer la moindre valeur.
Headroom s'installe comme un proxy, c'est-à-dire un intermédiaire qui se glisse entre votre machine et l'IA, et il tourne en local sur le port 8787. Avant que la moindre requête ne file vers le modèle, il intercepte tout ce qui gonfle le contexte, l'historique de conversation, les logs (les journaux d'activité techniques de la machine), les sorties d'outils, les bouts de documentation que le système a jugés utiles, et il compresse l'ensemble.
Un routeur devine d'abord le type de contenu, puis l'envoie vers le bon compresseur. Du code part vers un module qui le réduit à sa structure logique, son arbre syntaxique si vous voulez. Le JSON et le HTML, eux, passent à la moulinette pour virer tout le code de remplissage répétitif.
Et si le modèle réclame finalement la version complète ? Headroom garde les originaux de côté dans une petite base locale, Redis ou SQLite, et laisse l'IA aller les rechercher à la demande grâce à des marqueurs et au protocole MCP, ce standard récent qui permet à un modèle d'appeler des outils extérieurs tout seul.
Les taux de compression dépendent de la matière. Les logs serveur fondent de 90%. Les sorties d'outils MCP, bourrées de JSON répétitif, perdent à peu près 70%.
Présenté la semaine dernière à l'Open Source Summit, Headroom aurait déjà épargné quelque 700 000 dollars à ses utilisateurs, soit 200 milliards de tokens récupérés pour servir ailleurs.
Le projet reste officieux. Plusieurs équipes de Netflix s'en servent, mais ce n'est pas un produit maison estampillé par le studio.
À noter que Chopra a une explication assez simple à ce succès : beaucoup de ses utilisateurs sont des gens qui se sont fait sérieusement échauder par le coût des tokens, plus que par n'importe quoi d'autre.
Voir un ingénieur régler son propre problème de facture puis filer la solution gratuitement, plutôt que d'en faire une startup, c'est suffisamment rare pour qu'on le souligne.
En 2027, Apple pourrait s'attaquer aux lunettes connectées et viser beaucoup plus haut que les Meta Ray-Ban ou les lunettes Samsung. La marque californienne, sous la direction de son nouveau patron John Ternus, souhaiterait devenir le plus grand vendeur de lunettes de la planète, à l'image de l'Apple Watch sur sa catégorie.
Last month’s name that ware demonstrates that even though non-destructive IR imaging is not capable of resolving an individual bit cell, at least at 22nm it is still possible to constrain the number of bits in an SRAM macro.
An important step in establishing trust in a computer is measuring all of its state and confirming that nothing is amiss. A typical trusted boot would make a point of zeroing and/or patterning & hashing all the known bits of memory in a system. This process helps constrain the amount of malicious or foreign code that could be hiding in the system.
Physical measurements are important is because it’s possible for designers to “hide” memory from this check. For example, inserting a few kilobytes of RAM into a chip the size of the Baochip-1x would not affect the die size. Likewise, its impact on power consumption would be indistinguishable from offsets due to normal manufacturing tolerances. Furthermore, the presence of the RAM can be masked from a pure software inspection by gating it off using a “secret knock” register that only activates the memory when challenged with a correct sequence of words. This makes it practically impossible to discover hidden memories with a brute force address space scan. Such a memory would evade security measurements, and thus makes a useful primitive for staging malicious operations.
IR imaging can place an upper bound on how much SRAM is on a chip. This allows end users to check that all the RAM claimed to be in an open-RTL system (such as the Baochip-1x) matches what was actually fabricated. This in turn puts strong bounds on certain security operations, such as zeroing and/or measuring the state of all known RAM bits in a system.
The good news is that a simple physical measurement through IR inspection thoroughly eliminates the possibility of extra RAM macros in a system, as such a block would be observable even by the most entry-level home IRIS setup: the smallest blocks of RAM are gigantic compared to the resolution of an IR scan. The number of claimed blocks should strictly line up with the number shown in the source code, as it does in the case of the Baochip-1x.
That being said, it’s worth asking if an attacker could “just make a few bytes of RAM in a subtle way” or perhaps “just insert an extra row or column” in an existing macro. To understand the answer to this question better, let’s take a look at deeper look at the structure of an SRAM macro.
How to Read SRAM Macros
Above shows some details on the “rdram1kx32” RAM macro (“Macro D” from the competition) that makes up the data cache elements for the RV32 in the Baochip-1x. From the source code, we can see that this is a dual-port (1r/1w) RAM, organized as 1024 x 32 bits. I’ve rotated the RAM macro so that it’s in canonical “textbook” orientation, such that the columns go vertically and the rows go horizontally. When looking at a micrograph like this, generally speaking, lighter areas are metal-heavy, and darker areas are transistor-heavy. The transistor-heavy RAM arrays correspond to the eight dark rectangles on either side of a central spine.
Such a central spine is a common motif in circuit design. Splitting circuits in half reduces the maximum wire length by half, compared to sticking all the drive circuits on one side. Circuits work equally well when laid out in mirror-image, thus allowing techniques like this to achieve perfect symmetry around a central axis.
I’ve labeled some of the macroscopic features of the RAM. Along the center of the macro, you can see two banks of address decoders. You can use this structure as a “tell” for whether you’re looking at a single or dual port RAM macro. The bottom edge has the column sense amps & drivers. Each column shares circuitry across 4 bits, so their pitch is quite wide and thus visible even at IR wavelengths. Here you can readily count the number of bits in each half of the macro, 16 on each side, giving us a total of 32 bits.
The RAM is further subdivided into four distinct black rows on either side, with some sort of stippled “metal-heavy” region stitched in between. These stippled regions are inserted into RAM macros to improve their performance – they are repeaters that reduce the maximum length of a wire between RAM cells and their corresponding drivers. Repeaters are essential for performance because the wires in a RAM array are minimum-width for maximum density. This means they are quite resistive – in fact, they behave closer to a chain of resistors than ideal wires. Thus, higher performance can be achieved by reducing the wire length through the insertion of repeaters like this.
This leaves us now looking at the RAM array itself. The actual storage cells are too small to see, but we can infer that its size must be 256 bits x 16 bits total gross storage. The actual organization is (4×64)x16 bits – in other words, each of those column circuits is connected to 4 bits of memory, and the lower two bits of address are used to select between those, while 6 bits of address are decoded to select between 64 rows inside that gray blob.
Above is detail of the bioram1kx32 macro (“Macro A“). From the RTL, we can infer that this is a single-port, 1kx32 RAM. Here, you can see the central spine consisting of the single-port address decoder, and then on the lower edge mirror-symmetric 2×16 bit data in/out drivers/sense amps. Again, the core memory cells are laid out in a (64×4)x16 pattern, but in this case, we don’t have any repeaters between the banks. The single-port structure required of the BIO means we can hit the necessary timing without losing density to repeaters. Of course, we pay for this in the lower IPC of the BIO (due to separate read and write phases to the RAM), but for this design, a primary concern was keeping the cores small, so it works out.
And finally, above, here’s some detail of the aoram1kx36 RAM macro (“Macro E“). This has a structure distinct from the previous two: this is a density-optimized RAM macro. Here the address decoders occupy the yellow “spine” down the middle, and the column drivers/decoders are also primarily along the middle instead of along the edge. We can just barely make out the number of columns, and we have to infer that the rows are 512 deep organized as 128×4. While the merits of the density-optimized macro doesn’t really shine for such a small chunk of RAM, I’ll leave it as an exercise to the reader to look at the ifram32kx36 macro and observe how the structure scales up favorably compared to the smaller performance-oriented macros.
One side note is that in a modern silicon process, RAM macros are always in the same orientation across the entire chip. There are increasingly strict rules on the orientation of transistor gates as node size diminishes – fabs hyper-optimize yield for a single orientation and spacing of transistor gate relative to the crystalline lattice of the base silicon. Transistor gates are not so much as “drawn”, but carved out of sub-wavelength interference patterns. For some fairly solid technical reasons, you never have to worry about looking for RAMs built at funny angles or orientations.
No Place to Hide?
Now that we understand a little more about the structure of the RAM macros, let’s return to the question of places where we could hide “a little bit” of extra memory.
First, let’s observe that every SRAM bit is accompanied by some overhead circuitry. For various circuit-level reasons, it’s not possible to “just” lay down some extra-tiny SRAM cells without the supporting decoder, amplifier and driver circuitry. Thus, even the smallest “extra” RAM macro would be noticeable, if not simply for the presence of all the overhead circuitry surrounding the SRAM.
We can also observe that in all cases, we’re able to clearly see the number of columns in an SRAM. Thus, widening an SRAM by adding bits to its datapath would be clearly evidenced by the additional column drivers.
The tricky case is extra rows. We definitely can’t make out individual rows – we can only infer their existence by the process of eliminating the observed datapath widths against the list of possible RAMs on the chip. Here we need to make a distinction between attackers that can only generate fab-approved RAMs, versus an attacker that can modify RAM arbitrarily. RAMs are some of the densest, most difficult to fabricate structures, so in general fabs would not accept a design with true custom RAM.
In the case that an attacker is someone working at the netlist level and trying to insert a trojan in the RTL, I would argue this would be detectable because “just adding another row” of memory has ripple effects onto the memory organization. For example, a 1024×32 RAM already has all the row decoders fully-occupied. Building a 1028×32 RAM (recall that the rows come in increments of 4 bits) would add an extra level to the address decoders and a noticeable offset to the row size.
In the case that an attacker is someone with the necessary fab privileges to make small, customized modifications to the RAM macro, I would say it’s a coin toss if it can be detected with an IR level scan: definitely detectable in an older node, and probably detectable in 22 nm. The extra row will need space, and you’d need some circuitry to handle hiding it; but I can’t definitively rule out that someone can’t be clever enough to make it so small that it’s within the noise margin of an IR inspection. That being said, RAM cells are not infinitesimally small, and further measurements with e.g. laser interferometry could characterize sub-wavelength interference patterns with coherent light to detect even subtle changes in highly regular arrays such as SRAMs. In any case, a destructive inspection tool such as SEM can be used to follow up on any suspect structures and find ground truth.
That being said, a full custom RAM macro is a lot of work for just a single row of memory. While a row of RAM is enough to perhaps hide a shadow copy of a cryptographic key, one would still need to consider where the code or logic would be hidden that both obscures the extra row’s existence from an address space scan, and also choreographs the attack in a useful fashion.
While it’s somewhat unsatisfying to conclude that we can’t rule out every conceivable attack with simple visual inspection, I take solace in the fact that the majority of the attack surface is ruled out, and we now know what sorts of measurements and attacks are relevant to focus on from a defense perspective.
The types of attacks that are possible with and without visual inspection of the chip are vastly different. Without inspection, an attacker is readily able to hide several kilobytes of RAM behind a secret knock; with inspection, we’re looking for at most a few dozen bytes of RAM in an otherwise unprotected array. We turn an effectively unbounded search for hidden memory into a much more tractable proof-of-space measurement, with a caveat that it’s possible for a few extra bytes to be hidden by a determined adversary. I feel that this caveat is diminishingly small, as these extra bytes are very expensive to inject, yet still readily detectable with other means, tilting the cost/reward function towards “not worth it” on the list of potential attacks an adversary could spend resources on.
You Can Try It at Home!
In addition to the practical application of security measurements, I think it’s just interesting to be able to look at a piece of silicon and determine how much RAM is located where. I’m always curious about what makes things tick, and this is yet another tool in the toolbox that allows me to peer through the veil and see into the heart of my hardware. In the spirit of name that ware, I hope this spurs more people to look at silicon, and to learn what we can by observing the design patterns harbored within our gadgets. If you want to learn more about using IR to read CSP-packaged chips such as the Baochip-1x, check out my IRIS page, or just skip to a video I made on how to modify an off the shelf USB camera microscope into an IR camera.
Google Drive's new document scanner lets you sweep over a stack of pages in one continuous motion, automatically fixes blurry captures, and skips duplicates.
Attention, votre iPhone vous surveille pendant vos appels FaceTime. Depuis iOS 26, Apple a glissé une fonction baptisée "Sensitive Content Warning" qui scanne en temps réel le flux vidéo et fige automatiquement la communication dès qu'elle détecte de la nudité.
Audio coupé, vidéo coupée, avec un message qui s'affiche : "Audio et vidéo sont en pause car vous montrez peut-être quelque chose de sensible". Sympa.
À la base, la fonctionnalité fait partie d'une suite plus large appelée "Communication Safety", destinée aux comptes enfants. L'idée est légitime : éviter que des mineurs ne tombent sur du contenu inapproprié, ou en envoient.
Sauf que voilà, dans iOS 26, la fonction est aussi accessible sur les comptes adultes. Elle est désactivée par défaut, mais on peut l'activer manuellement. Et plusieurs utilisateurs constatent que les faux positifs ne sont pas rares.
Le principe technique est plutôt rassurant côté vie privée. La détection se fait entièrement sur l'appareil grâce à du machine learning embarqué (les modèles de reconnaissance d'image tournent en local sur votre iPhone, sans envoyer la moindre image à Apple). Du coup, contrairement à ce qu'on pourrait imaginer, Apple n'a aucune idée de ce qui déclenche la détection sur votre téléphone. Le scan ne sort pas de l'appareil.
Le problème, c'est que ce genre de filtre se trompe souvent. Une consultation médicale en visio ? Bloquée. Une conversation entre adultes consentants ? Bloquée aussi. Et chaque interruption nécessite une action manuelle pour reprendre l'appel. Pour les usages légitimes qui ressemblent visuellement à de la "nudité" sans en être, c'est franchement pénible.
Au-delà du bug ou choix de design, la question de fond, c'est le précédent. Apple installe l'idée que votre flux vidéo, même dans un appel chiffré de bout en bout, peut être analysé en permanence par les filtres maison. Aujourd'hui c'est de la nudité.
Demain, ça pourrait être des armes, des drogues, des contenus politiques selon les pays, ou tout ce que les gouvernements demanderont. La porte est ouverte, et c'est ce qui inquiète une partie des défenseurs de la vie privée. Apple a beau jurer que tout reste en local, l'infrastructure d'analyse est désormais en place sur des centaines de millions d'iPhone.
Vous pouvez désactiver la fonction en allant dans Réglages, puis FaceTime, puis « Avertissement de contenu sensible ». C'est en théorie OFF par défaut, mais autant vérifier. Et si vous l'aviez activée par curiosité, sachez qu'elle peut ruiner un appel important au pire moment.
The HW1 by Gesture Platforms is a 10-degree-of-freedom (DOF) high-dexterity robotic hand and wrist built around an ESP32-S3 wireless MCU. It’s primarily designed for researchers, educators, and hobbyists; it bridges the gap between basic DIY robotic hands and expensive industrial models.
The device weighs just around 500 grams but can handle a 1kg dynamic load and a 3kg static load. It communicates via USB-C or Bluetooth 5.0 and comes with a companion desktop app as well as Python and C++ SDKs for custom development.
Gesture HW1 specifications:
Microcontroller – Espressif Systems ESP32-S3 dual-core LX7 microprocessor with WiFi 4 and Bluetooth 5.0 LE connectivity
Degrees of Freedom (active) – 10 DOF
Finger Flexion – 0° – 90° (at each joint)
Finger Adduction/Abduction – -10° – 40°
Thumb Flexion/Extension – 0° – 120°
Thumb Adduction/Abduction – 20° – 90°
Thumb Distal Flexion – 0° – 90°
Wrist Flexion/Extension – -50° – 50°
Wrist Radial/Ulnar Deviation – -10° – 40°
Joints – 19 joints total
Payload – 1kg dynamic / 3kg static load
Accuracy/Repeatability – ±1mm
Control Speed – 100Hz
Sensors – Angle (at motor), Current, Temperature
USB – 1x USB Type-C port for communications
Misc
4 x user-programmable RGB LEDs
Four-way stretch fabric cover prevents debris from entering joints
Power – Via the XT60 connector on the side
Dimensions – ~275mm total height (fits men-small / women-medium gloves)
Weight – ~500 grams
Mounting – M3x0.5 threads (depth of 4.5mm)
The HW1 uses aluminum gears at key joints like the wrist for better strength and durability, while its skeleton is made from lightweight, strong glass and carbon-fiber-filled polycarbonate, and the outer shell uses ASA plastic to handle heat and rough use. The fingertips are covered with soft silicone (Shore 30A) to improve grip. Gesture Platforms states that it will provide STEP files, so users can 3D print, modify, or replace the outer parts easily. Most probably, they will release the file after the campaign ends.
On the software side, the company provides a plug-and-play desktop application for Windows 10/11. The app also works completely offline and features four control modes, which include Text (importing .csv files), Timeline (keyframe-style editing), Controller (mapping to a standard game controller), and Stream (real-time input). For more advanced users, C++ and Python SDKs will be available to integrate the hand into custom machine learning and control algorithms.
Gesture HW1 Use CasesGesture HW1 supports rapid AI deployment for testing AI-generated control actions.
The Gesture HW1 ESP32-S3 high-dexterity robotic hand is currently crowdfunding on Kickstarter. The “Launch Day Special” reward requires a $699 pledge (limited to 100 units), while the standard “Early Backer” tier is available for $849. A double pack for both the left and right hands is available for $1,599. Shipping is not included and will be charged after the campaign; deliveries are estimated to begin in November 2026.
Groupe chinois d’électronique grand public, Anker Innovations a lancé la première plateforme de puces d’intelligence artificielle qui permet aux appareils du quotidien de profiter de la puissance des réseaux neuronaux. Une vraie révolution technologique.
Chaque fois que votre smartphone capte un signal alors que vous êtes au milieu de nulle part, sans antenne relais à des kilomètres, c'est probablement un satellite qui fait le travail.Cette technologie, dite "direct-to-device", permet à un téléphone de se connecter directement à un satellite en orbite basse, sans infrastructure terrestre. Starlink do...
