San Francisco autonomous vehicle company Cruise recalled and updated the software of its fleet of 300 cars, reports the San Francisco Chronicle, " after a Cruise taxi rear-ended a local bus "when the car's software got confused by the articulated vehicle, according to a federal safety report and the company."

The voluntary report notes that Cruise updated its software on March 25th. Since last month's low-speed crash, which resulted in no injuries, Cruise CEO Kyle Vogt said the company chose to conduct a voluntary recall, and the software update assured such a rare incident "would not recur...." As for the March bus collision, Vogt said the software fix was uploaded to Cruise's entire fleet of 300 cars within two days. He said the company's probe found the crash scenario "exceptionally rare" with no other similar collisions.

"Although we determined that the issue was rare, we felt the performance of this version of software in this situation was not good enough," Vogt wrote in a blog post. "We took the proactive step of notifying NHTSA that we would be filing a voluntary recall of previous versions of our software that were impacted by the issue." The CEO said such voluntary recalls will probably become "commonplace."

"We believe this is one of the great benefits of autonomous vehicles compared to human drivers; our entire fleet of AVs is able to rapidly improve, and we are able to carefully monitor that progress over time," he said.
The Cruise car was traveling about 10 miles per hour, and the collision caused only minor damage to its front fender, Vogt's blog post explained. San Francisco's buses have front and back coaches connected by articulated rubber, and when the Cruise taxi lost sight of the front half, it made the assumption that it was still moving (rather than recognizing that the back coach had stopped). Or, as Cruise told the National Highway Traffic Safety Administration, their vehicle ""inaccurately predicted the movement" of the bus. It was not the first San Francisco incident involving Cruise since June, when it became the first company in a major city to win the right to taxi passengers in driverless vehicles — in this case Chevrolet Bolts. The city's Municipal Transportation Agency and County Transportation Authority recorded at least 92 incidents from May to December 2022 in which autonomous ride-hailing vehicles caused problems on city streets, disrupting traffic, Muni transit and emergency responders, according to letters sent to the California Public Utilities Commission....

Just two days before the Cruise crash in March, the company had more problems with Muni during one of San Francisco's intense spring storms. A falling tree brought down a Muni line near Clay and Jones streets on March 21, and a witness reported on social media that two Cruise cars drove through caution tape into the downed wire. A company representative said neither car had passengers and teams were immediately dispatched to remove the vehicles.

On Jan. 22, a driverless Cruise car entered an active firefighting scene and nearly ran over hoses. Fire crews broke a car window to try to stop it.

Long-time Slashdot reader SpzToid shares a new report from the Economist: Thanks to innovations such as fractal colouration patterns, which mimic nature by repeating shapes at different scales, the distance from which naked eyes can quickly spot soldiers wearing the best camouflage has shrunk, by one reckoning, by a fifth over the past two decades. That is impressive. On today's battlefields, however, it is no longer enough to merely hide from human eyes.

People and kit are given away as well by signals beyond the visual spectrum, and devices that detect these wavelengths are getting better, lighter and cheaper. Thermal sensors are a case in point. Today, one that costs about $1,000 and weighs as little as five sachets of sugar can, in good weather, detect a warm vehicle as far off as 10km. As Hans Kariis, deputy head of signatures research at the Swedish Defence Research Agency, notes, that is well beyond the range at which a small drone would be spotted. Two decades ago, he adds, a less sensitive thermal sensor weighing a kilogram cost ten times as much.
And then there's automatic target-detection software, the article points out, like the Kestrel software deployed in more than 3,500 aircraft around the world, which "scans feeds of visual, infrared and radar data, and places red boxes around people and other potential targets, even as their positions in the frame move." And the threat has only increased with the arrival of satellite-based synthetic-aperture-radar (SAR) imagery.

But then the article lists examples of new camouflage that now tricks electronic sensors:

• Military vehicles affix hexagon-shaped sheets that can be cooled with electricity to blend into the temperature of their surroundings.
• Camouflage netting that absorbs (some) incoming radar beams with semi-conducting polymers while reducing heat signatures with insulation — and reflecting back the cooler temperature of the ground.
• Netherlands-based TNO makes "battery-powered sniper suits" embedded with 500 LEDs that match the luminosity and color of the surroundings using real-time data from a helmet camera.

"As automakers scour the planet for the metals it will take to build tens of millions of electric cars, they are deliberately taking a detour around one of the only places on earth where so much of what they need is laying around and available to be plucked," reports the Washington Post: The deep seabed is teeming with potato-sized rocks packed with the nickel, cobalt, copper and manganese EV manufacturers covet. But efforts by mining companies to harvest the nodules with undersea robots are hitting rough waters. EV manufacturers who need the minerals for their batteries are distancing themselves from the practice as diplomats and scientists sound an alarm over the ecological damage that could be caused by rushing to scrape the sea floor.

The misgivings of the auto companies are hardly assuaged by the messy, contentious deliberations over it all at the headquarters of the United Nations-chartered International Seabed Authority here. The authority, tasked with protecting and guiding development in international waters, has been in turmoil since the small Pacific Island nation of Nauru invoked a clause tucked in the Law of the Sea that could allow mining within months, likely before the full environmental impact is known or regulations are put in place....

More than 700 marine scientists have signed a petition demanding a [mining] moratorium, which is also supported by 13 countries. French President Emmanuel Macron is calling for a permanent ban.... Some car manufacturers, including BMW, Renault, Rivian, Volvo and VW publicly support a moratorium on seabed mining. GM, Ford and Daimler are, for now, keeping deep seabed materials out of their supply chain plans amid corporate concerns about environmental impact.
The Post got an interesting perspective from Ethan Elkind, director of the climate program at UC Berkeley's Center for Law, Energy and the Environment. "Auto companies don't want EV batteries associated with more destruction than they already have been," he tells them. "If ecological damage of these sensitive undersea areas comes to light after mining begins, they don't want to be a part of it."

Long-time Slashdot reader schwit1 shared this report from NBC News: The Chinese spy balloon that flew across the U.S. was able to gather intelligence from several sensitive American military sites, despite the Biden administration's efforts to block it from doing so, according to two current senior U.S. officials and one former senior administration official. China was able to control the balloon so it could make multiple passes over some of the sites (at times flying figure-eight formations) and transmit the information it collected back to Beijing in real time, the three officials said.

The intelligence China collected was mostly from electronic signals, which can be picked up from weapons systems or include communications from base personnel, rather than images, the officials said. The three officials said China could have gathered much more intelligence from sensitive sites if not for the administration's efforts to move around potential targets and obscure the balloon's ability to pick up their electronic signals by stopping them from broadcasting or emitting signals.
America's Department of Defense "directed NBC News to comments senior officials made in February that the balloon had 'limited additive value' for intelligence collection by the Chinese government 'over and above what [China] is likely able to collect through things like satellites in low earth orbit.'"

The Free Software Foundation certifies products that meet their standards in regard to users' freedom, control over the product, and privacy. And they put out a new "Respects Your Freedom" certification on Thursday for ThinkPenguin's free software gigabit mini VPN router, the TPE-R1400.

From the FSF's announcement: This is ThinkPenguin's first device to receive RYF certification in 2023, adding to their vast catalogue of certified devices from previous years. As with previous routers from ThinkPenguin, the Free Software Gigabit Mini VPN Router ships with an FSF-endorsed fully free embedded GNU/Linux distribution called libreCMC. It also comes with a custom flavor of the U-Boot boot loader, assembled by Robert Call, the maintainer of libreCMC and a former FSF intern.

The router enables users to run their network connection through a VPN service, helping to simplify the process of keeping their communications secure and private. While ThinkPenguin offers a VPN service, users are not required to purchase a subscription to their service in order to use the router, and the device comes with detailed instructions on how to use the router with a wide variety of VPN providers.

"We're pleased to see ThinkPenguin continue with their commitment to bringing out devices that put software freedom as their first priority under the RYF program. The release of this router shows that ThinkPenguin is committed to the privacy and freedom of their users," said the FSF's executive director, Zoë Kooyman....

"The latest version of ThinkPenguin's VPN router lets its users take advantage of gigabit per second Internet connections while protecting their rights and privacy," said FSF's copyright and licensing associate, Craig Topham.

America's Department of Justice just launched an investigation into the leaking of classified documents from the U.S. Department of Defense, reports the Washington Post.

"On Wednesday, images showing some of the documents began circulating on the anonymous online message board 4chan and made their way to at least two mainstream social media platforms, Telegram and Twitter." Earlier Friday, The Washington Post obtained dozens of what appeared to be photographs showing classified documents, dating to late February and early March, that range from worldwide intelligence briefings to tactical-level battlefield updates and assessments of Ukraine's defense capabilities. They outline information about the Ukrainian and Russian militaries, and include highly sensitive U.S. analyses about China and other nations. The materials also reference highly classified sources and methods that the United States uses to collect such information, alarming U.S. national security officials who have seen them.... The material that appeared online includes photographs of documents labeled "Secret" or "Top Secret," and began appearing on Discord, a chat platform popular with gamers, according to a Post review.

In some cases, it appears that the slides were manipulated. For instance, one image features combat casualty data suggesting the number of Russian soldiers killed in the war is far below what the Pentagon publicly has assessed. Another version of the image showed higher Russian casualty figures. Besides the information on casualties that appeared to be manipulated to benefit the Russian government, U.S. officials who spoke to The Post said many of the leaked documents did not appear to be forged and looked consistent in format with CIA World Intelligence Review reports distributed at high levels within the White House, Pentagon and the State Department....

The documents appear to have been drawn from multiple reports and agencies, and concern matters other than Ukraine. Two pages, for example, are purportedly a "CIA Operations Center Intelligence Update," and includes information about events concerning Russia, Hungary and Iran.... Rachel E. VanLandingham, a former Air Force attorney and expert on military law, said that whoever is responsible for the leak "is in a world of hurt." Such breaches, she said, constitute "one of the most serious crimes that exist regarding U.S. national security...."

Skepticism abounded Friday among both Russian and Ukrainian officials aware of reports about the leaks, with each side accusing the other of being involved in a deliberate act of disinformation.
The Post notes one defense official told them "hundreds — if not thousands" of people had access to the documents, so their source "could be anyone."

But the photographs received by the Post were apparently taken from printed documents, and "classified documents may only be printed from computers in a secure facility, and each transaction is electronically logged, said Glenn Gerstell, a former general counsel with the National Security Agency who emphasized that he was speaking only about general procedures. "The fact that the documents were printed out should significantly narrow the universe of the initial inquiry."

A Cornell University researcher has developed sonar glasses that "hear" you without speaking. Engadget reports: The eyeglass attachment uses tiny microphones and speakers to read the words you mouth as you silently command it to pause or skip a music track, enter a passcode without touching your phone or work on CAD models without a keyboard. Cornell Ph.D. student Ruidong Zhang developed the system, which builds off a similar project the team created using a wireless earbud -- and models before that which relied on cameras. The glasses form factor removes the need to face a camera or put something in your ear. "Most technology in silent-speech recognition is limited to a select set of predetermined commands and requires the user to face or wear a camera, which is neither practical nor feasible," said Cheng Zhang, Cornell assistant professor of information science. "We're moving sonar onto the body."

The researchers say the system only requires a few minutes of training data (for example, reading a series of numbers) to learn a user's speech patterns. Then, once it's ready to work, it sends and receives sound waves across your face, sensing mouth movements while using a deep learning algorithm to analyze echo profiles in real time "with about 95 percent accuracy." The system does this while offloading data processing (wirelessly) to your smartphone, allowing the accessory to remain small and unobtrusive. The current version offers around 10 hours of battery life for acoustic sensing. Additionally, no data leaves your phone, eliminating privacy concerns. "We're very excited about this system because it really pushes the field forward on performance and privacy," said Cheng Zhang. "It's small, low-power and privacy-sensitive, which are all important features for deploying new, wearable technologies in the real world." "The team at Cornell's Smart Computer Interfaces for Future Interactions (SciFi) Lab is exploring commercializing the tech using a Cornell funding program," adds Engadget. "They're also looking into smart-glasses applications to track facial, eye and upper body movements."

A video of the eyeglasses can be viewed here.

Walmart plans to have its own network of electric vehicle charging stations by 2030 to tap into the growing adoption of EVs in the United States. Reuters reports: The new fast-charging stations will be placed at thousands of Walmart and Sam's Club stores, alongside nearly 1,300 it already offers as part of a deal with Volkswagen unit Electrify America, one of the country's largest open public EV networks. Walmart's more than 5,000 stores and Sam's Club warehouses are located within 10 miles of about 90% of Americans.

"We have the ability to address range and charging anxiety in a way that no one else can in this country," Vishal Kapadia, Walmart's recently appointed senior vice president of Energy Transformation, said in an interview. Owning its chargers, instead of partnering with a network operator, will help Walmart address reliability and cost issues, Kapadia said. Kapadia said he expects the new charge points to be direct-current fast chargers, with about four chargers on average installed per store.

An anonymous reader quotes a report from Wired: The Russian government has banned more than 10,000 websites for content about the war in Ukraine since Moscow launched the full-scale invasion in February 2022. The blacklist includes Facebook, Twitter, Instagram, and independent news outlets. Over the past year, Russians living inside the country have turned to censorship circumvention tools such as VPNs to pierce through the information blockade. But as dozens of virtual private networks get blocked, leaving users scrambling to maintain their access to free information, local activists and developers are coming up with new solutions. One of them is Amnezia VPN, a free, open source VPN client.

"We even do not advertise and promote it, and new users are still coming by the hundreds every day," says Mazay Banzaev, Amnezia VPN's founder. Unlike commercial VPNs that route users through company servers, which can be blocked, Amnezia VPN makes it simple for users to buy and set up their own servers. This allows them to choose their own IP address and use protocols that are harder to block. "More than half of the commercial VPNs in Russia have been blocked because it's easy enough to block them: They do not block them by protocols, but by IP addresses," says Banzaev. "[Amnezia] is an order of magnitude more resilient than a typical commercial VPN." Amnezia VPN is similar to Outline, a free and open source tool developed by Jigsaw, a subsidiary of Google. Amnezia was created in 2020 during a hackathon supported by Russian digital rights organization Roskomsvoboda. Even then, "it was clear that things were moving toward stricter censorship," says Banzaev. [...]

It is unclear how many users the service has, since the organization doesn't have a way to monitor user numbers, Banzaev says. However, Amnezia offers a Telegram bot called AmneziaFree, which shares VPN configurations that help users access blocked platforms and news; it has almost 100,000 users. The bot is currently struggling with overload, and users are complaining about spotty service. Banzaev says the Amnezia team is working to add new servers on a limited budget, and that they are also working on a new version of the service. "Amnezia is not only used in Russia," notes Wired. "The service has spread to Turkmenistan, Iran, China, and other countries where users have been struggling with free access to the web."

With no discussion, the Arkansas House of Representatives overwhelmingly approved a bill that would require social media users in The Natural State to verify they're 18 years old or older to use the platforms. Arkansas Times reports: The proposal, backed by Gov. Sarah Sanders, is aimed at shielding minors from the harmful effects of social media. Young folks could use the platforms, but only if parents provide consent. Senate Bill 396, sponsored by Sen. Tyler Dees (R-Springdale) and Rep. Jon Eubanks (R-Paris), would require social media companies including Facebook, Instagram, Twitter and TikTok to contract with third-party companies to perform age verification. Users would have to provide the third-party company with a digital driver's license. Dees also sponsored a bill, now law, that requires anyone who wants to watch online pornography to verify they're an adult.

The social media bill squeaked through the Senate with 18 yes votes, the bare minimum, but passed the House 82-10 with four voting present (same as no). No one asked any questions of Eubanks -- who assured his colleagues that Facebook had "the AI and algorithms" to keep track of what users had parental consent without holding on to sensitive data -- but because it was amended (to among other things exempt LinkedIn, the most boring social media platform), the bill has to go back to the Senate, where perhaps it will meet some resistance. Utah's governor signed two bills into law last month requiring companies like Meta, Snap and TikTok to get parents permission before teens could create accounts on their platforms. "The laws also require curfew, parental controls and age verification features," adds Engadget.

An anonymous reader shares a report: Back in 2020, we reported that emulator developers were using a hole in the Xbox Store's app distribution system to get around Microsoft's longstanding ban on emulators running on Xbox consoles. This week, though, many of the emulators that were distributed through that workaround have stopped working, the apparent victims of a new crackdown by Microsoft. Xbox emulator makers and users can't say they weren't warned. In the "Gaming and Xbox" section of Microsoft's official Store Policies, section 10.13.10 clearly states that "products that emulate a game system or game platform are not allowed on any device family."

Microsoft's enforcement of this clause has historically focused on removing emulators published as "private" UWP apps to the Xbox Store. Those apps could be distributed to whitelisted users via direct links accessed on the system's Edge browser, getting around the usual approval process for a public store listing. Previously, users who downloaded one of these "hidden" emulator listings before Microsoft's inevitable takedown could run that emulator on an unmodified retail system indefinitely. That is no longer the case; trying to launch downloaded versions of emulators like Xenia or Retrospection on an Xbox console now generates an error saying, "Unable to launch this game or app. The game or app you're trying to launch violates Microsoft Store policy and is not supported."

Google is ending support for the Dropcam and the Nest Secure home security system in one year, on April 8th, 2024. From a report: They are among the few remaining Nest products that haven't been brought over to Google Home, and their demise hints that the new Google Home app might almost be here. At least, no more than a year away. Surely. Google is also winding down the last few legacy Works with Nest connections, but not 'til September 29th. Existing Dropcam cameras will keep working until April 8th, 2024, after which you won't be able to access them from the Nest app. To soften the blow, Google's offering a free indoor wired Nest Cam to Dropcam owners who subscribe to Nest Aware. Nonsubscribers will get a 50 percent-off coupon. The promotion runs until May 7, 2024, so you can keep using your Dropcam until it stops working.

Samsung Electronics will cut back on chip production, as it faces a sharp decline in global demand for semiconductors that has sent prices plunging. From a report: The world's biggest memory chip maker said it would make a "meaningful" cut to chip output after sales dropped sharply and it flagged a 96% drop in first-quarter profits, worse than expected. The fellow South Korean firm SK Hynix and Micron Technology of the US have also reduced production. "Samsung talking about production cuts is evidence of how bad the current slump really is," said Greg Roh, the head of research at Hyundai Motor Securities.

Smartphone and personal computer makers ramped up purchases of chips during the Covid-19 pandemic, when demand for consumer electronics soared as people were stuck at home during lockdowns. This led to a global chip shortage. However, demand has waned as consumers cut back on bigger purchases amid the cost of living crisis, with food and energy bills soaring. Samsung said demand had dropped because of a weaker world economy and companies buying fewer chips as they run down their inventories. "We are lowering the production of memory chips by a meaningful level, especially that of products with supply secured," the company said, referring to customers with sufficient inventories.

Startups and other companies trying to capitalize on the artificial intelligence boom sparked by OpenAI are running into a problem: They can't find enough specialized computers to make their own AI software. The Information: A spike in demand for server chips that can train and run machine-learning software has caused a shortage, prompting major cloud-server providers including Amazon Web Services, Microsoft, Google and Oracle to limit their availability for customers, according to interviews with the cloud companies and their customers. Some customers have reported monthslong wait times to rent the hardware. "All the startups who are trying to get into this space...maybe they can get one [server] but there's no way they're going to get five," said Johnny Dallas, founder and CEO of Zeet, which sells software that makes it easier for engineers to run apps across multiple clouds.

The server chip shortage is a frustrating hangup for software developers trying to build AI tools hinging on recent advancements in machine-learning models. These programmers, at small and big companies alike, are developing large-language models to make personalized writing coaches or search engines that respond to questions with written answers rather than links, similar to OpenAI's ChatGPT. Many others are licensing and augmenting software from OpenAI and its rivals to create specialized customer service chatbots and research tools for corporate employees. For instance, OpenAI software is helping Morgan Stanley bankers find the best locations to auction a work of art, based on the bank's myriad internal reports on art markets.

Multi-brand consumer megacorp Unilever says it has become a "cloud-only enterprise" with the help of Accenture and Microsoft. From a report: One of the largest and most complex cloud migrations in the retail goods industry, according to the company, will give Unilever "resilient, secure and optimised operations" as well as "a platform to drive innovation and growth." The Anglo-Dutch biz owns more than 400 brands, which include everything from ice cream to shampoo to toilet cleaner, and is set to use Microsoft's Azure as its "primary cloud platform."

According to the corporate blurb, the move will see Unilever employ "industrial metaverse technologies" that use real-time data from factory digital twins. It musn't have got the memo from Microsoft, which recently put a bullet in its own industrial metaverse masterplan. The cloud contract is also expected to help "achieve perpetual breakthroughs in research and development," says Unilver. Lastly, through Microsoft's partnership with the controversial GPT maker, it will use "Azure OpenAI Service across Unilever's business to drive increased automation, enabling better customer and employee experiences."

Putting a temporary pause on artificial intelligence development would only hand an advantage to competitors in China, former Google CEO Eric Schmidt said, after more than 1,000 researchers signed a letter warning of the consequences of moving too quickly on AI research. From a report: Speaking to the Australian Financial Review in an interview published Friday, Schmidt said there were legitimate concerns about the speed of research into AI but they should be mitigated by tech companies working together to set standards. In the past week, more than 1,000 researchers and executives, including Tesla Chief Executive Officer Elon Musk, signed an open letter published by the Future of Life Institute, which called for an AI research pause of "at least six months," warning of "potentially catastrophic effects" on society if appropriate governance wasn't put in place. But Schmidt said he wasn't in favor of the six-month pause as it would "simply benefit China. What I am in favor of is getting everyone together ASAP to discuss what are the appropriate guardrails."

AI research startup Anthropic aims to raise as much as $5 billion over the next two years to take on rival OpenAI and enter over a dozen major industries, according to company documents obtained by TechCrunch. From the report: A pitch deck for Anthropic's Series C fundraising round discloses these and other long-term goals for the company, which was founded in 2020 by former OpenAI researchers. In the deck, Anthropic says that it plans to build a "frontier model" -- tentatively called "Claude-Next" -- 10 times more capable than today's most powerful AI, but that this will require a billion dollars in spending over the next 18 months.

Anthropic describes the frontier model as a "next-gen algorithm for AI self-teaching," making reference to an AI training technique it developed called "constitutional AI." At a high level, constitutional AI seeks to provide a way to align AI with human intentions -- letting systems respond to questions and perform tasks using a simple set of guiding principles. Anthropic estimates its frontier model will require on the order of 10^25 FLOPs, or floating point operations -- several orders of magnitude larger than even the biggest models today. Of course, how this translates to computation time depends on the speed and scale of the system doing the computation; Anthropic implies (in the deck) it relies on clusters with "tens of thousands of GPUs."

Full-size, self-driving bus services will begin in Scotland next month in what is believed to be a world first. The BBC reports: Stagecoach said the route over the Forth Road Bridge would launch on May 15. The 14-mile route will run between Ferrytoll park and ride in Fife and Edinburgh Park train and tram interchange. Five single-decker autonomous buses will have the capacity for about 10,000 passenger journeys per week.

The vehicles have sensors enabling them to travel on pre-selected roads at up to 50mph. They will have two members of staff on board. A safety driver will sit in the driver's seat to monitor the technology, and a so-called bus captain will help passengers with boarding, buying tickets and queries. The UK government said Project CAVForth would be the world's first full-size, self-driving public bus service.

For $30, you can ship Best Buy a prepaid box filled with up to 15 pounds of unwanted electronics. CNET reports: Starting this month, two sizes of prepaid boxes are available on the Best Buy website: a 9-by-5-by-3-inch container that can carry up to 6 pounds for $23, and a larger, 18-by-14-by-4-inch box that can carry up to 15 pounds for $30. Once you've filled it up with approved electronics, you can take the box to a UPS location or arrange for a pickup.

The recycling-by-mail program is the latest salvo in Best Buy's campaign to achieve net-zero emissions by 2040. In April 2022, the company began offering a haul-away service that picks TVs, appliances and other products for recycling from customers' homes. You can also drop off unwanted electronics at Best Buy locations and trade in select merchandise for gift cards.

Meta has announced an AI model called the Segment Anything Model (SAM) that can identify individual objects in images and videos, even those not encountered during training. From a report: According to a blog post from Meta, SAM is an image segmentation model that can respond to text prompts or user clicks to isolate specific objects within an image. Image segmentation is a process in computer vision that involves dividing an image into multiple segments or regions, each representing a specific object or area of interest. The purpose of image segmentation is to make an image easier to analyze or process. Meta also sees the technology as being useful for understanding webpage content, augmented reality applications, image editing, and aiding scientific study by automatically localizing animals or objects to track on video.

Typically, Meta says, creating an accurate segmentation model "requires highly specialized work by technical experts with access to AI training infrastructure and large volumes of carefully annotated in-domain data." By creating SAM, Meta hopes to "democratize" this process by reducing the need for specialized training and expertise, which it hopes will foster further research into computer vision. In addition to SAM, Meta has assembled a dataset it calls "SA-1B" that includes 11 million images licensed from "a large photo company" and 1.1 billion segmentation masks produced by its segmentation model. Meta will make SAM and its dataset available for research purposes under an Apache 2.0 license. Currently, the code (without the weights) is available on GitHub, and Meta has created a free interactive demo of its segmentation technology.