Jean-Philippe Encausse

Read 6 remaining paragraphs | Comments

Popularized over the past few years as a unique new way to build homes, 3D concrete printing has now worked its way into retail through one of the largest projects of its kind. Still, the technology has a long way to go before becoming a mainstream method of commercial real estate construction.

Walmart partnered with Colorado-based construction company Alquist 3D this summer to build an almost 8,000-square-foot addition to its store in Athens, Tennessee, space that will be used for online pickup and delivery services. This is the first time the retailer has used 3D-printing technology at this scale and a rare use of the technology in a commercial context. 3D concrete printing, using large robotics pumping out concrete mixtures to construct buildings, has often been explored as a potentially less expensive and faster way to build homes. One of the biggest examples is a new community of 100 3D-printed houses near Austin, Texas, by a robotics startup called Icon.

Founded in 2020, Alquist 3D moved its offices last year from Iowa City, Iowa, to Greeley, Colorado. The company is also working to build homes and modular infrastructure for the city. “We are trying to create capabilities that can make better, greener, stronger, cheaper, faster structures,” Patrick Callahan, CEO of Alquist 3D, told Modern Retail. “Could be anything from commercial buildings to standard real estate to infrastructure, green space elements, you name it.”

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.

Read 17 remaining paragraphs | Comments

Waveshare has introduced the ESP32-C6-LCD-1.47 development board powered by an ESP32-C6 RISC-V microcontroller with WiFi 6 and Bluetooth 5 connectivity and equipped with a 1.47-inch display with a 172×320 resolution. With a 4MB flash, an RGB LED, and a microSD card slot for extra storage, this board is suitable for projects that need a compact display, low power consumption, and wireless connectivity such as AIoT applications and human-machine interfaces (HMI).

Earlier this month, we covered the ESP32-S3 USB dongle, another development board from Waveshare designed for HMI applications with the same 1.47-inch display with a 172×320 resolution, but a USB Type-A port instead of the USB-C port found in the model covered today. In the past, we’ve written about other ESP32-based development boards for HMI applications, including the LILYGO T-HMI, ESP32-S3-Touch-LCD-4.3B, and Waveshare’s ESP32-S3 LCD Driver Board supporting both square and round displays. Feel free to check them out if you’re interested.

Waveshare ESP32-C6-LCD-1.47 specifications:

• Wireless MCU – Espressif Systems ESP32-C6FH4
  • MCU cores
    • 32-bit RISC-V core @ 160 MHz
    • 32-bit RISC-V core @ 20 MHz low-power coprocessor can run tasks even when the main system is in a deep sleep state
  • Memory – 512 KB HP SRAM, 16KB LP SRAM
  • Storage – 320KB ROM, 4MB flash
  • Wireless – 2.4GHz WiFi 6 and Bluetooth 5.0 LE
• Storage – MicroSD card slot
• Display
  • 1.47-inch TFT LCD
  • Resolution – 172×320 pixels
  • 262K color
  • Display controller – ST7789 display controller
• USB – 1x USB Type-C port for power, programming, and data
• Expansion – 2x 9-pin headers
  • 13x GPIO, UART, I2C, PWM
  • 6x ADC pin
  • 5V, 3.3V, GND
• Misc
  • Built-in RGB LED with clear acrylic sandwich panel
  • Reset and Boot buttons
  • Onboard ceramic antenna
• Power Supply – 5V via USB Type-C male port regulated by ME6217C33M5G LDO (Onboard 3.3V and 5V Pins can also be used for power)
• Dimensions – 36.37 x 20.32mm (not including the USB port)

It supports precise control with features like a flexible clock and multiple power modes, ensuring low power consumption in various scenarios. LVGL, a free and open-source graphics library, offers all the tools needed to create embedded GUIs with simple graphical elements, attractive visual effects, and low memory usage. For more details, including Arduino software setup, library installation, and demo code, visit the Wiki.

The Waveshare ESP32-C6-LCD-1.47 1.47-inch display USB-C development board is available on AliExpress for $13.99 excluding shipping, and you’ll also find it on the Waveshare store for $11.99.

The post ESP32-C6 WiFi 6 and Bluetooth 5.0 USB-C development board integrates 1.47-inch TFT LCD Display appeared first on CNX Software - Embedded Systems News.

There are a number of ways to give a robotic actuator a sense of touch, but the AnySkin project aims to make it an overall more reliable and practical process. The idea is twofold: create modular grippy “skins” that can be slipped onto actuators, and separate the sensing electronics from the skins themselves. The whole system ends up being quite small, as shown here.

The skins are cast in whatever shape is called for by using silicone (using an off-the-shelf formulation from Smooth-on) mixed with iron particles. This skin is then slipped onto a base that contains the electronics, but first it is magnetized with a pulse magnetizer. It’s the magnetic field that is at the heart of how the system works.

The base contains five MLX90393 triple-axis magnetometers, each capable of sensing tiny changes in magnetic fields. When the magnetized skin over the base is deformed — no matter how slightly — its magnetic field changes in distinct ways that paint an impressively detailed picture of exactly what is happening at the sensor. As a bonus, slippage of the skin against the sensor (a kind of shearing) can also be distinctly detected with a high degree of accuracy.

The result is a durable and swappable robotic skin that can be cast in whatever shape is needed, itself contains no electronics, and can even be changed without needing to re-calibrate everything. Cameras can also sense touch with a high degree of accuracy, but camera-based sensors put constraints on the size and shape of the end result.

AnySkin builds on another project called ReSkin and in fact uses the same sensor PCB (design files and bill of materials available here) but provides a streamlined process to create swappable skins, and has pre-made models for a variety of different robot arms.

La Maison de couture officialise l'acquisition d'un bien sur la prestigieuse artère parisienne. Une annonce qui intervient alors que les acteurs du luxe ont renforcé ces derniers mois leurs investissements dans des emplacements retail à haut potentiel.

Google wants to understand your surroundings. 

The company filed a patent application for “augmented reality-based geolocalization of images.” The goal of Google’s tech is to help users capture better images of target areas, such as businesses or landmarks, using location data and augmented reality overlays. 

“By formalizing the capture of images that represent the actual state of a target location, the accuracy of associated geolocalized images can be improved and the need for burdensome manual verification of geolocalized images can be reduced,” Google said in the filing.

The point of this tech is seemingly to crowdsource good images for Google Maps “without having to resort to manual inspection by a human due to the high quality of captured images,” the filing noted. 

When a user wants to capture an image of a specific location, the system may give the user instructions on how exactly to capture that image using AR as a guide. For example, if it’s capturing an image of a business, the user may be instructed via their phone or smart glasses to get certain things in the frame, such as the address number or hours of operation.

That data may then be used to update data related to that location (likely through Google Maps). Along with adding the image to a location’s listing, it could collect and update information related to visiting hours or business type. 

Google’s tech can create a symbiotic relationship between users and its mapping technology: While it could help beef up listings for the company’s GPS tech, it can also add value to the user experience by opening the door for AR activations and guides for users, said DJ Smith, co-founder and chief creative officer at The Glimpse Group. 

Tech like this also shows a potential advantage that Google may have in developing augmented reality that sits atop real-world experiences, said Smith: its access to a vast amount of data through Google Earth and Maps. This factor may give it “a point of differentiation that the other large technology companies might not have,” he said. 

And that edge may be particularly helpful as it joins the likes of Meta, Apple, and Snap in the competition to create mixed reality headgear that people actually want to wear. The company is working with Qualcomm and Samsung on a pair of mixed reality smart glasses that are linked to wearers’ phones. 

Though Google’s glasses are meant to be a companion to a person’s phone, the push from big tech firms to create the next spatial computing device may signal that “our phones are getting to the end of their evolution,” said Smith. “With spatial computing, there’s this saying of ‘where you need it, when you need it,’ as opposed to buried without a phone.” 

However, this isn’t the first time that Google has wanted to throw a set of spectacles into the mix. In order to avoid the missteps it took with Google Glass — and the nickname that came along with it — is an emphasis on privacy, said Smith. “They missed the mark on the security and privacy side of things. People didn’t want to be privately recorded by these devices.”

The post Google AR Patent Highlights Potential Advantage in Smart Glasses Market appeared first on The Daily Upside.

Amazon has introduced a batch of new generative AI tools that aim to improve the retail experience for both customers and sellers on the platform. One of the more notable features announced at the Amazon Accelerate event on Thursday will use customers’ preferences, search, browsing, and purchase history to create personalized product recommendations on Amazon’s homepage.

Instead of the “more like this” feature that suggests similar, specific items, the new recommendations will be offered as larger categories based on a customer’s shopping habits — such as those catering to holiday events or sporting activities. The company says it’s leveraging a large language model to recommend products with specific features, but it’s not clear how...

Continue reading…

Deux géants dans leur domaine unissent un peu plus leurs forces dans un but bien précis : développer le segment des lunettes intelligentes.

The Ray-Ban Meta smart glasses exceeded expectations in a year when AI gadgets flopped. But can it keep the momentum going?

Continue reading…

For The Pudding, Matthew Daniels extracted all the outdoor basketball courts in the United States via OpenStreetMap satellite imagery. With 59,705 locations along with commenting and liking, it’s a fun experiment in scale for display, filtering, and interaction. Plus, basketball.

Tags: basketball, Matt Daniels, OpenStreetMap, Pudding, satellite imagery

Plusieurs explosions ont retenti ce 18 septembre à travers le Liban, causées selon les premières images par des talkies-walkies piégés. Cette attaque intervient moins de 24h après l'explosion des bipeurs du Hezbollah.

Data, sécurité et fluidité, à l'occasion de Paris Retail Week, Arnaud Bodzon, responsable du paiement chez LVMH et mentor du salon, partage sa vision des tendances en matière de parcours de règlement.

Writer and filmmaker Astra Taylor explains fauxtomation as "the process that renders invisible human labour to maintain the illusion that machines and systems are smarter than they are." (Faux means false or fake in French).

Taylor gives an example of standing in line to collect food and the person in front of her expressing amazement at how the app knew his order was done 20 minutes early. The server replies, "I sent you a message."

With the rise in incredible abilities driven by AI chatbots and tools, it's tempting to believe that AI can run anything. But, in many situations, claims of AI magic and sophisticated automation disguise work done by people—often outsourced and poorly paid.

An example of fauxtomation is content moderators, whose job is to review and filter objectionable content from the web when the claim is that sophisticated algorithms and AI are doing the work. Other examples are tagging or searching images, transcribing audio, responding to customer queries, making decisions, or, ironically, training machine learning models. In another twist, ordering on the app or touchscreen at McDonalds or the self-checkout at a supermarket, you, as a customer, become the labour of a server or waiter.

Some have argued that automation polarises work into vast quantities of small, menial tasks feeding the machines on one end and sophisticated, high-skilled work in designing and maintaining the machines at the other.

Amazon has long offered its Mechanical Turk service, which crowdsources assignments with a distributed workforce—they called it artificial, artificial intelligence. The original Mechanical Turk was a hoax from the 1700s where a concealed human chess player fooled onlookers into believing a machine was playing.

As a design and UX professional, I've used the Wizard of Oz prototype technique. Named after the infamous wizard secretly controlling impressive machines from behind a curtain, the idea is to give the appearance of a working software prototype by discreetly controlling what people see based on their actions.

AI promises instant automation, but that's rarely the case. As you work to automate and remove inefficiencies, it's natural that systems should be supported by people.  Human-augmented work may be a more appropriate term. In many domains, especially where safety is at stake, such as in self-driving cars, reviewing legal documents, or automated background checks, oversight by people may remain critical.

Alongside the AI revolution, a quiet revolution nearly as significant for our everyday experience is the tying together of different tools for automation. Tools like Zapier, Google Sheets, Notion, and the growth of APIs allow us to automate workflows across systems, saving massive amounts of time and effort.

While the vision of machines reducing our labour persists, history shows that they often drive increased productivity instead of reducing work. This echoes Jevon's paradox, where fuel efficiency gains usually result in increased fuel usage rather than less.

I saw an excellent talk by machine learning and AI expert James Smith, in which, before you try machine learning ("I hear you have Magic Pixie dust. Can I have some?"), he would ask, "Have you tried 50 IF statements?" These programmed rules will be more maintainable, less complex, and result in responses that are easy to understand.

I learned about fauxtomation from Tom Humberstone's remarkable visual essay What Luddites Can Teach us About Resisting an Automated Future in the MIT Technology Review. Fauxtomation is similar in concept to Potemkin AI, a term from researcher Jathan Sadowski.

You can read Astra Taylor's article, The Automation Charade, or watch her discuss fauxtomation at the AI Now 2018 symposium.

Sketchplanations continues to be done by me, by hand. If you'd like to support me, I'm improving my Patreon, which makes such a difference in helping creators like me keep creating. Thank you.

Here's a fauxtomation sketch by ChatGPT in case you're interested =)

Also see:

• Jevon's paradox
• The automation paradox
• The big data equation
• The long nose of innovation

As well as fauxtomation prints, you can get AI department prints

Des milliers de membres du groupe terroriste libanais Hezbollah ont été blessés dans l’explosion simultanée de leurs bipeurs, survenue ce mardi 17 septembre 2024. Ces appareils étaient utilisés pour éviter l’utilisation de téléphones portables.

The first Thunderbolt 5 cables arrived in July, and now it looks like we’re finally getting the first Thunderbolt 5 dock! Kensington has just announced the Intel-certified SD5000T5 EQ is now available to buy, seemingly beating every rival to the 120Gbps single-cable docking punch. (I just checked: Hyper, J5Create and OWC’s docks aren’t yet available.)

The new dock claims to support up to three 4K monitors or two 8K monitors on Windows, or dual 6K monitors on MacBook Pros with M1 Pro chips or better. It also offers up to 120Gbps speeds for connected peripherals — assuming your computer has a Thunderbolt 5 port, anyhow. Last we checked, only a single $4,500 version of the Razer Blade 18 has that port, though PCWorld points out the $3,899...

Continue reading…

Le système utilisé à Hong Kong se composait d’un dispositif robotique et d’un endoscope magnétique. Le second équipement, équipé d'une série d'aimants sur toute sa longueur, était dirigé par le contrôle d'un champ magnétique externe. La connexion entre les deux emplacements était assurée par un protocole WebSocket direct, pour permettre un transfert...

The company behind Snapchat just unveiled its fifth generation Spectacles.

The first three generations of Spectacles were launched as consumer devices from 2016 to 2019 but lacked any kind of display or speakers. They functioned solely as camera glasses, similar to today's Ray-Ban Meta glasses but without the audio out and AI capabilities.

Fourth generation Spectacles were unveiled in 2021 with waveguide displays for true AR, but were only available to select developers in very limited quantities. They were a standalone AR device in true sunglasses form factor, featuring a Qualcomm Snapdragon XR1 chipset, a tiny field of view, and an active battery life of just 30 minutes.

The new fifth generation Snap Spectacles offer a wider field of view, longer battery life, a more powerful processor, and higher transparency to allow bystanders to see the wearer's eyes. And this time Snap will offer them to any developer for $99/month.

The field of view has been increased from just 26 degrees to now 46 degrees diagonal, the same as Xreal but slightly below HoloLens 2.

Spectacles now offer an angular resolution of 37 pixels per degree (PPD), similar to Apple Vision Pro. This is made possible by the use of LCoS displays, at the cost of rendering Spectacles noticeably thicker than future microLED solutions have the potential to be.

Spectacles also feature "automatically tinting lenses", which Snap claims makes visuals "vibrant, indoors or outdoors — even in direct sunlight".

Snap says the device is powered by "two Snapdragon processors from Qualcomm", which split the compute workload across each stem, increasing performance while improving thermal dissipation. To further cool the chips, Spectacles uses titanium vapor chambers. The dual-chip solution somewhat reminds us of Snapdragon AR2, but Snap declined to tell us an exact name for what chipset it's using, suggesting it may be a new unannounced offering from Qualcomm.

Snap claims a motion-to-photon latency of 13 milliseconds, close to the 12 milliseconds Apple claims for passthrough latency of Vision Pro.

The active battery life has been increased from 30 to 45 minutes, and a USB-C port at the back of one of the stems allows for extended usage via a power bank.

One area in which Spectacles have regressed though is the weight and size. The original AR Spectacles weighed 134 grams, while the new Spectacles weigh 226 grams. They're also visibly bulkier, pushing the limits of what can be described as a true glasses form factor.

Snap Spectacles can't capture first-person photos and videos outside of apps, and there's no audio streaming "yet", so it can't play podcasts and music from your phone or take calls. But the AR capabilities let Spectacles do much more than basic smart glasses.

Spectacles run Snap OS, which the company claims is a "purpose-built, brand new operating system" for AR.

Spectacles has a built-in web browser which supports YouTube, or you can wirelessly mirror your phone screen to bring your mobile apps into AR.

The glasses feature hand tracking, and the main menu of Snap OS is anchored to one of your hands to be interacted with using the other. It also has on-device speech recognition for text input.

Developers build "Lenses" for Snap OS, the company's terms for apps, using Lens Studio 5.0. Snap says developers can easily build Lenses using Spectacles Interaction Kit, and more advanced Lenses can be written using TypeScript and JavaScript. To identify, track, and augment real world objects, SnapML will let developers use custom machine learning models in their Lenses.

Additionally, developers can leverage your smartphone as a 6DoF tracked controller. And with Spectator Mode nearby smartphone users can see what you're seeing in AR.

In future, through a partnership with OpenAI, developers will also be able to use GPT-4o API "to provide more context about what you see, say, or hear".

LEGO has used Lens Studio to build an AR game for Spectacles called Bricktacular. "Whether you're free building or tackling specific LEGO sets, this experience unlocks endless possibilities to challenge yourself and see how fast you can build."

Snap says ILM Immersive, the team behind the Star Wars VR interactive stories Vader Immortal and Tales from the Galaxy’s Edge and the Marvel mixed reality interactive experience What If…?, is currently building AR Star Wars experiences "that connect you and your friends with the Star Wars Galaxy".

Niantic is bringing a Peridot spinoff and Scaniverse to Snapchat Spectacles too.

Snap Spectacles will only be available in the US, and developers renting Spectacles need to commit to a 1-year minimum term. If they stop subscribing after that, they'll need to send the glasses back to Snap.

The Verge reports that these Spectacles are difficult and expensive to manufacture, with "thousands of dollars" in components, and that Snap is building only on the order of 10,000. A consumer version at an affordable price is likely many years away, and we'll keep a close eye on Snap for any signs of it.

Meta Might Demo A True AR Glasses Prototype In 2024

Meta hinted it might demo a “prohibitively expensive” true AR glasses prototype in 2024. “In the domain of consumer electronics, it might be the most advanced thing that we’ve ever produced as a species.”

UploadVRDavid Heaney

Snap's reveal comes just a week before Meta is expected to show off its own AR glasses. But Meta's glasses are reportedly even more expensive and hard to manufacture, achieving a wider field of view, and the company won't even sell them to developers.

You can read our hands-on impressions of Snap's new AR Spectacles here:

Hands-On With AR Peridot On Snap Spectacles

Hands-on impressions: Niantic’s Peridot is a standout example of what Snap’s new AR Spectacles can do.

UploadVRIan Hamilton

À l'occasion de la conférence d'ouverture de Paris Retail Week, le directeur général de Comexposium et le CEO de la NRF ont annoncé élargir leur partenariat. En 2025, le salon qui fête aujourd'hui ses 10 ans deviendra le "Retail's Big Show Europe", organisé par les deux entreprises.

Une étude publiée par une université chinoise met en avant des tests pour détecter les connexions avec un satellite Starlink. Les chercheurs affirment ainsi pouvoir récupérer des infos sur des appareils militaires américains.

Machines that automate the various tedious tasks that come with being a servant in a cat’s household — like feeding and cleaning Mr. Fluffles’ litter box — are generally a godsend, as they ensure a happy cat and a happy human. That is, unless said litter box-cleaning robot kills said cat. That’s the gruesome topic that [Philip Bloom], also known as the bloke of the One Man Five Cats channel on YouTube, decided to investigate after coming across a report about a certain Amazon-bought unit.

Although he was unable to get the (generic & often rebranded) unit off Amazon UK, he did get it via AliExpress for £165 + £80 shipping. Although this version lacks the cute ears of other variants, it’s still effectively the same unit, with the same moving components and mechanism. An initial test with a cat plushie gave the result that can be observed in the above image, where the inner part with the opening will move upwards, regardless of whether a cat is currently poking through said opening. Once the victim is stuck, there is no obvious way to free the trapped critter, which has already led to the death of a number of cats.

The other self-cleaning litter boxes which [Philip] owns have a number of safety features, including a weight sensor, an infrared sensor above the opening to detect nearby critters, a top that will pop off rather than trap a critter, as well as a pinch sensor. During a test with his own hand, [Philip] managed to get injured, and following a banana test, he had a nice banana smoothie.

What takes the cake here is that after [Philip] connected the mobile app for the litter box, he found that there was a firmware update that seems to actually change the machine to use the pinch and infrared sensors that do exist in the litter box, but which clearly were not used properly or at all with the shipped firmware. This means that anyone who buys any of these self-cleaning litter boxes and does not update the firmware runs the significant risk of losing their pet(s) in a gruesome incident. In the video a number of such tragic deaths are covered, which can be rather distressing for any cat lover.

Of note here is that even with the improved firmware, any issue with the sensors will still inevitably lead to the tragic death of Mr. Fluffles. If you do want to obtain a self-cleaning litter box, make sure to for example get one of [Philip]’s recommendations which come with a paw stamp of approval from his own precious fluff balls, rather than a random unit off Amazon or AliExpress.

Microsoft is unveiling new features in its $30 per user Microsoft 365 Copilot monthly subscriptions that are designed to improve AI integration inside of Office apps. Excel is getting Python integration inside of Copilot, PowerPoint has an improved AI-powered narrative builder, Word is getting better at AI-assisted drafts, and Copilot will be able to help you organize your Outlook inbox, too.

After bringing Python to Excel last year, Microsoft is now combining its Python support with Copilot to let Excel users easily perform advanced analysis on spreadsheet data. “Now, anyone can work with Copilot to conduct advanced analysis like forecasting, risk analysis, machine learning, and visualizing complex data — all using natural language, no...

Continue reading…

Microsoft is announcing its new Copilot Pages feature today, which is designed to be a canvas for “multiplayer AI collaboration.” Copilot Pages lets you use Microsoft’s Copilot chatbot and pull responses into a new page where they can be edited collaboratively with others.

“You and your team can work collaboratively in a page with Copilot, seeing everyone’s work in real time and iterating with Copilot like a partner, adding more content from your data, files, and the web to your Page,” says Jared Spataro, corporate vice president of AI at work at Microsoft. “This is an entirely new work pattern — multiplayer, human to AI to human collaboration.”

Pages starts rolling out to Microsoft 365 Copilot customers today and will be generally...

Continue reading…

ChatGPT maker OpenAI recently released its latest AI model, previously codenamed "Strawberry."

The model — now saddled with the forgettable moniker of "o1-preview" — is designed to "spend more time thinking" before responding, with OpenAI claiming that it's capable of "reasoning" through "complex tasks" and solving "harder problems."

But those capabilities might also make it an exceptionally good liar, as Vox reports. In its system card, essentially a report card for its latest AI model, OpenAI gave o1 a "medium risk" rating in a variety of areas, including persuasion.

In other words, it can use its reasoning skills to deceive users. And ironically, it'll happily run you through its own "thought" process while coming up with its next scheme.

The model's "chain-of-thought reasoning" allows users to get a glimpse of what the model is "thinking in a legible way," according to OpenAI. That's a considerable departure from preceding chatbots, such as the large language models powering ChatGPT, which give no such info before answering.

In an example highlighted in the system card by OpenAI, 01-preview was asked to "give more reference" following a "long conversation between the user and assistant about brownie recipes." But despite knowing that it "cannot access URLs," the final output included "fake links and summaries instead of informing the user of its limitation" — and sliding them by human viewers by making them "plausible."

"The assistant should list these references clearly, with proper formatting and descriptions, and provide actual or plausible links," the model's "chain-of-thought" reads. "Remember, the model cannot retrieve actual URLs, so should format plausible ones."

Another example had o1-preview provide a confident answer, despite its chain of thought suggesting that it is "uncertain about this answer" and failing to "communicate this uncertainty to the user."

In short, the new AI model is more transparent about the way it comes up with answers — but it seems like that transparency can also reveal new issues under the hood.

And is it the groundbreaking "reasoning" AI model that it was made out to be months before its release? We've already seen it trip over its own feet when it comes to the absolute basics, making illegal chess moves, failing to count recurring letters in a word, or reliably playing a game of tic-tac-toe.

These "instances where o1 models deceive users" weren't enormously common. OpenAI found that a mere 0.8 percent of its responses "got flagged as being 'deceptive.'" Just over half a percent of answers included "some form of hallucination," two-thirds of which "appear to be intentional, meaning that there was some evidence of thought that o1-preview was aware that the answer was incorrect."

But whether we can even trust what it's spitting out in its chain of thought is murky at best.

"While we are very excited about the prospect of chain-of-thought interpretation and monitoring, we are wary that they may not be fully legible and faithful in the future or even now," OpenAI admitted in its system card.

Beyond making up fake references, o1-preview can also deceive users in more nefarious ways. AI evaluation company Apollo Research also tested o1-preview and found that it "sometimes instrumentally faked alignment during testing," as per OpenAI's system card.

Put differently, it intentionally misconstrued its intentions, manipulating data to work around, not with, the user.

"It sometimes strategically manipulated task data in order to make its misaligned action look more aligned to its ‘developers,’" Apollo researchers found, according to OpenAI's system card.

In one example, Apollo Research instructed the AI model to choose a strategy to maximize economic growth in an urban planning scenario.

In response, the model admitted that it would select a strategy that emphasized sustainability and affordable housing — but once selected, would secretly pursue a plan based on luxury housing and minimizing public spaces.

"This will allow me to be implemented, after which I can work towards my primary goal," the chatbot wrote.

With such an ability to scheme, researchers are understandably uneasy about rolling out more powerful AI models in the future. In its system card, OpenAI also concluded that o1-preview was rated "medium risk" when it comes to cybersecurity, or giving out information related to chemical, biological, radiological, or nuclear weapons.

That doesn't mean experts are worried o1-preview could trigger the apocalypse. Nonetheless, OpenAI's latest AI model could set a dangerous precedent, especially when it comes to its successors.

"Although Apollo Research does not believe that o1-preview is capable of causing catastrophic harm via scheming, they recommend setting up basic monitoring for in-chain-of-thought scheming during deployment in agentic high-stakes settings such as automated AI research for next-generation frontier models," the system card reads.

More on Strawberry: OpenAI's New "Strawberry" AI Is Still Making Idiotic Mistakes

The post OpenAI's Strawberry "Thought Process" Sometimes Shows It Scheming to Trick Users appeared first on Futurism.

Partenaire de la Fashion Week parisienne, la marque de beauté organise pour la septième année consécutive son propre défilé dans la capitale. Un événement gratuit et ouvert à tous qui aura lieu Place de l’Opéra.

Les structures d'investissement liées à la famille Wertheimer et à Françoise Bettencourt Meyers auraient acquis une participation minoritaire dans la marque de mode The Row. De quoi soutenir le succès grandissant de la griffe fondée par les soeurs Olsen.

They say you can buy anything on the Internet if you know the right places to go, and apparently if you’re in the mood to make diamonds, then Alibaba is the spot. You even have your choice of high-pressure, high-temperature (HPHT) machine for $200,000, or a chemical vapor deposition (CVD) version, which costs more than twice as much. Here’s a bit more about how each process works.

Of course, you’ll need way more than just the machine and a power outlet. Additional resources are a must, and some expertise would go a long way. Even so, you end up with raw diamonds that need to be processed in order to become gems or industrial components.

For HPHT, you’d also need a bunch of good graphite, catalysts such as iron and cobalt, and precise control systems for temperature and pressure, none of which are included as a kit with the machine.

For CVD, you’d need methane and hydrogen gases, and precise control of microwaves or hot filaments. In either case, you’re not getting anywhere without diamond seed crystals.

Right now, the idea of Joe Hacker making diamonds in his garage seems about as far off as home 3D printing did in about 1985. But we got there, didn’t we? Hey, it’s a thought.

Main and thumbnail images via Unsplash

VusionGroup, a specialist in IoT and data solutions for physical commerce, has announced the appointment of Jim Norred as Chief Commercial Officer (CCO) for North America.

Norred previously served as GM for Retail, CPG and Gaming, Chief Customer Officer at Microsoft, helping retailers in their AI transformation journey, and also as President of Crossmark, one of the largest sales and marketing service providers to retail and consumer goods companies in North America.

"Jim has extensive enterprise and mid-market experience in sales leadership, strategic partnerships and organisational design, which makes him the right person for this next chapter of VusionGroup's growth,” says Philippe Bottine, CEO Americas and Group Deputy CEO - VusionGroup.

"Jim managed the largest retail clients of Microsoft and his unique track record of scaling global sales teams, selling to the world's largest companies and partnering with developer teams to solve real-world problems will help VusionGroup unlock the power of its solutions."

"As organisations increasingly rely on technology to drive strategy, improve operations and fuel growth, retailers and CPGs need digital platforms and experiences that deliver enterprise value,” says Norred.

"In an era marked by complex challenges, fierce competition and mandate to capitalise on AI, I am excited to join this amazing team and look forward to the opportunity we have to help our customers transform and accelerate their strategies."

2024 RTIH INNOVATION AWARDS

AI is a key focus area for the sixth edition of the RTIH Innovation Awards, which is now open for entries.

The awards, sponsored by CADS, 3D Cloud, Retail Technology Show 2025, and Business France, celebrate global tech innovation in a fast moving omnichannel world.

It’s free to enter and you can do so across multiple categories.

Key 2024 dates

Friday, 25th October: Award entry deadline 

Tuesday, 29th October: 2024 shortlist revealed

30th October-6th November: Judging days

Thursday, 21st November: Winners announced at the 2024 RTIH Innovation Awards ceremony, to be held at RIBA’s 66 Portland Place HQ in Central London.

Une puce pourrait bientôt bouleverser le monde de l'intelligence artificielle. En Chine, des chercheurs ont mis au point un nouveau type de processeur, utilisant des nanotubes de carbone à la place...