verge-rss

Photo by Harun Ozalp / Anadolu via Getty Images

On Friday morning, shortly after midnight in New York, disaster started to unfold around the world. In Australia, shoppers were met with Blue Screen of Death (BSOD) messages at self-checkout aisles. In the UK, Sky News had to suspend its broadcast after servers and PCs started crashing. In Hong Kong and India, airport check-in desks began to fail. By the time morning rolled around in New York, millions of Windows computers had crashed, and a global tech disaster was underway.
In the early hours of the outage, there was confusion over what was going on. How were so many Windows machines suddenly showing a blue crash screen? “Something super weird happening right now,” Australian cybersecurity expert Troy Hunt wrote in a post on X. On Reddit, IT admins raised the alarm in a thread titled “BSOD error in latest CrowdStrike update” that has since racked up more than 20,000 replies.
The problems led to major airlines in the US grounding their fleets and workers in Europe across banks, hospitals, and other major institutions unable to log in to their systems. And it quickly became apparent that it was all due to one small file.
At 12:09AM ET on July 19th, cybersecurity company CrowdStrike released a faulty update to the Falcon security software it sells to help companies prevent malware, ransomware, and any other cyber threats from taking down their machines. It’s widely used by businesses for important Windows systems, which is why the impact of the bad update was so immediate and felt so broadly.
CrowdStrike’s update was supposed to be like any other silent update, automatically providing the very latest protections for its customers in a tiny file (just 40KB) that’s distributed over the web. CrowdStrike issues these regularly without incident, and they’re fairly common for security software. But this one was different. It exposed a massive flaw in the company’s cybersecurity product, a catastrophe that was only ever one bad update away — and one that could have been easily avoided.
How did this happen?
CrowdStrike’s Falcon protection software operates in Windows at the kernel level, the core part of an operating system that has unrestricted access to system memory and hardware. Most other apps run at user mode level and don’t need or get special access to the kernel. CrowdStrike’s Falcon software uses a special driver that allows it to run at a lower level than most apps so it can detect threats across a Windows system.
Running at the kernel makes CrowdStrike’s software far more capable as a line of defense — but also far more capable of causing problems. “That can be very problematic, because when an update comes along that isn’t formatted in the correct way or has some malformations in it, the driver can ingest that and blindly trust that data,” Patrick Wardle, CEO of DoubleYou and founder of the Objective-See Foundation, tells The Verge.
Kernel access makes it possible for the driver to create a memory corruption problem, which is what happened on Friday morning. “Where the crash was occurring was at an instruction where it was trying to access some memory that wasn’t valid,” Wardle says. “If you’re running in the kernel and you try to access invalid memory, it’s going to cause a fault and that’s going to cause the system to crash.”
CrowdStrike spotted the issues quickly, but the damage was already done. The company issued a fix 78 minutes after the original update went out. IT admins tried rebooting machines over and over and managed to get some back online if the network grabbed the update before CrowdStrike’s driver killed the server or PC, but for many support workers, the fix has involved manually visiting the affected machines and deleting CrowdStrike’s faulty content update.
While investigations into the CrowdStrike incident continue, the leading theory is that there was likely a bug in the driver that had been lying dormant for some time. It might not have been validating the data it was reading from the content update files properly, but that was never an issue until Friday’s problematic content update.
“The driver should probably be updated to do additional error checking, to make sure that even if a problematic configuration got pushed out in the future, the driver would have defenses to check and detect… versus blindly acting and crashing,” says Wardle. “I’d be surprised if we don’t see a new version of the driver eventually that has additional sanity checks and error checks.”
CrowdStrike should have caught this issue sooner. It’s a fairly standard practice to roll out updates gradually, letting developers test for any major problems before an update hits their entire user base. If CrowdStrike had properly tested its content updates with a small group of users, then Friday would have been a wake-up call to fix an underlying driver problem rather than a tech disaster that spanned the globe.
Microsoft didn’t cause Friday’s disaster, but the way Windows operates allowed the entire OS to fall over. The widespread Blue Screen of Death messages are so synonymous with Windows errors from the ’90s onward that many headlines initially read “Microsoft outage” before it was clear CrowdStrike was at fault. Now, there are the inevitable questions over how to prevent another CrowdStrike situation in the future — and that answer can only come from Microsoft.
What can be done to prevent this?
Despite not being directly involved, Microsoft still controls the Windows experience, and there is plenty of room for improvement in how Windows handles issues like this.
At the simplest, Windows could disable buggy drivers. If Windows determines that a driver is crashing the system at boot and forcing it into a recovery mode, Microsoft could build in more intelligent logic that allows a system to boot without the faulty driver after multiple boot failures.
But the bigger change would be to lock down Windows kernel access to prevent third-party drivers from crashing an entire PC. Ironically, Microsoft tried to do exactly this with Windows Vista but was met with resistance from cybersecurity vendors and EU regulators.
Microsoft tried to implement a feature known at the time as PatchGuard in Windows Vista in 2006, restricting third parties from accessing the kernel. McAfee and Symantec, the big two antivirus companies at the time, opposed Microsoft’s changes, and Symantec even complained to the European Commission. Microsoft eventually backed down, allowing security vendors access to the kernel once again for security monitoring purposes.
Apple eventually took that same step, locking down its macOS operating system in 2020 so that developers could no longer get access to the kernel. “It was definitely the right decision by Apple to deprecate third-party kernel extensions,” says Wardle. “But the road to actually accomplishing that has been fraught with issues.” Apple has had some kernel bugs where security tools running in user mode could still trigger a crash (kernel panic), and Wardle says Apple “has also introduced some privilege execution vulnerabilities, and there are still some other bugs that could allow security tools on Mac to be unloaded by malware.”
Regulatory pressures may still be stopping Microsoft from taking action here. The Wall Street Journal reported over the weekend that “a Microsoft spokesman said it cannot legally wall off its operating system in the same way Apple does because of an understanding it reached with the European Commission following a complaint.” The Journal paraphrases the anonymous spokesperson and also mentions a 2009 agreement to provide security vendors the same level of access to Windows as Microsoft.
Microsoft reached an interoperability agreement with the European Commission in 2009 that was a “public undertaking” to allow developers to get access to technical documentation for building apps on top of Windows. The agreement was formed as part of a deal that included implementing a browser choice screen in Windows and offering special versions of Windows without Internet Explorer bundled into the OS.
The deal to force Microsoft to offer browser choices ended five years later in 2014, and Microsoft also stopped producing its special versions of Windows for Europe. Microsoft now bundles its Edge browser in Windows 11, unchallenged by European regulators.
It’s not clear how long this interoperability agreement was in place, but the European Commission doesn’t seem to believe it’s holding back Microsoft from overhauling Windows security. “Microsoft is free to decide on its business model and to adapt its security infrastructure to respond to threats provided this is done in line with EU competition law,” European Commission spokesperson Lea Zuber says in a statement to The Verge. “Microsoft has never raised any concerns about security with the Commission, either before the recent incident or since.”
The Windows lockdown backlash
Microsoft could attempt to go down the same route as Apple, but the pushback from security vendors like CrowdStrike will be strong. Unlike Apple, Microsoft also competes with CrowdStrike and other security vendors that have made a business out of protecting Windows. Microsoft has its own Defender for Endpoint paid service, which provides similar protections to Windows machines.
CrowdStrike CEO George Kurtz also regularly criticizes Microsoft and its security record and boasts of winning customers away from Microsoft’s own security software. Microsoft has had a series of security mishaps in recent years, so it’s easy and effective for competitors to use these to sell alternatives.
Every time Microsoft tries to lock down Windows in the name of security, it also faces backlash. A special mode in Windows 10 that limited machines to Windows Store apps to avoid malware was confusing and unpopular. Microsoft also left millions of PCs behind with the launch of Windows 11 and its hardware requirements that were designed to improve the security of Windows PCs.
Cloudflare CEO Matthew Prince is already warning about the effects of Microsoft locking down Windows further, framed in a way that Microsoft will favor its own security products if such a scenario were to occur. All of this pushback means Microsoft has a tricky path to tread here if it wants to avoid Windows being at the center of a CrowdStrike-like incident again.
Microsoft is stuck in the middle, with pressure from both sides. But at a time when Microsoft is overhauling security, there has to be some room for security vendors and Microsoft to agree on a better system that will avoid a world of blue screen outages again.

Photo by Harun Ozalp / Anadolu via Getty Images

On Friday morning, shortly after midnight in New York, disaster started to unfold around the world. In Australia, shoppers were met with Blue Screen of Death (BSOD) messages at self-checkout aisles. In the UK, Sky News had to suspend its broadcast after servers and PCs started crashing. In Hong Kong and India, airport check-in desks began to fail. By the time morning rolled around in New York, millions of Windows computers had crashed, and a global tech disaster was underway.

In the early hours of the outage, there was confusion over what was going on. How were so many Windows machines suddenly showing a blue crash screen? “Something super weird happening right now,” Australian cybersecurity expert Troy Hunt wrote in a post on X. On Reddit, IT admins raised the alarm in a thread titled “BSOD error in latest CrowdStrike update” that has since racked up more than 20,000 replies.

The problems led to major airlines in the US grounding their fleets and workers in Europe across banks, hospitals, and other major institutions unable to log in to their systems. And it quickly became apparent that it was all due to one small file.

At 12:09AM ET on July 19th, cybersecurity company CrowdStrike released a faulty update to the Falcon security software it sells to help companies prevent malware, ransomware, and any other cyber threats from taking down their machines. It’s widely used by businesses for important Windows systems, which is why the impact of the bad update was so immediate and felt so broadly.

CrowdStrike’s update was supposed to be like any other silent update, automatically providing the very latest protections for its customers in a tiny file (just 40KB) that’s distributed over the web. CrowdStrike issues these regularly without incident, and they’re fairly common for security software. But this one was different. It exposed a massive flaw in the company’s cybersecurity product, a catastrophe that was only ever one bad update away — and one that could have been easily avoided.

How did this happen?

CrowdStrike’s Falcon protection software operates in Windows at the kernel level, the core part of an operating system that has unrestricted access to system memory and hardware. Most other apps run at user mode level and don’t need or get special access to the kernel. CrowdStrike’s Falcon software uses a special driver that allows it to run at a lower level than most apps so it can detect threats across a Windows system.

Running at the kernel makes CrowdStrike’s software far more capable as a line of defense — but also far more capable of causing problems. “That can be very problematic, because when an update comes along that isn’t formatted in the correct way or has some malformations in it, the driver can ingest that and blindly trust that data,” Patrick Wardle, CEO of DoubleYou and founder of the Objective-See Foundation, tells The Verge.

Kernel access makes it possible for the driver to create a memory corruption problem, which is what happened on Friday morning. “Where the crash was occurring was at an instruction where it was trying to access some memory that wasn’t valid,” Wardle says. “If you’re running in the kernel and you try to access invalid memory, it’s going to cause a fault and that’s going to cause the system to crash.”

CrowdStrike spotted the issues quickly, but the damage was already done. The company issued a fix 78 minutes after the original update went out. IT admins tried rebooting machines over and over and managed to get some back online if the network grabbed the update before CrowdStrike’s driver killed the server or PC, but for many support workers, the fix has involved manually visiting the affected machines and deleting CrowdStrike’s faulty content update.

While investigations into the CrowdStrike incident continue, the leading theory is that there was likely a bug in the driver that had been lying dormant for some time. It might not have been validating the data it was reading from the content update files properly, but that was never an issue until Friday’s problematic content update.

“The driver should probably be updated to do additional error checking, to make sure that even if a problematic configuration got pushed out in the future, the driver would have defenses to check and detect… versus blindly acting and crashing,” says Wardle. “I’d be surprised if we don’t see a new version of the driver eventually that has additional sanity checks and error checks.”

CrowdStrike should have caught this issue sooner. It’s a fairly standard practice to roll out updates gradually, letting developers test for any major problems before an update hits their entire user base. If CrowdStrike had properly tested its content updates with a small group of users, then Friday would have been a wake-up call to fix an underlying driver problem rather than a tech disaster that spanned the globe.

Microsoft didn’t cause Friday’s disaster, but the way Windows operates allowed the entire OS to fall over. The widespread Blue Screen of Death messages are so synonymous with Windows errors from the ’90s onward that many headlines initially read “Microsoft outage” before it was clear CrowdStrike was at fault. Now, there are the inevitable questions over how to prevent another CrowdStrike situation in the future — and that answer can only come from Microsoft.

What can be done to prevent this?

Despite not being directly involved, Microsoft still controls the Windows experience, and there is plenty of room for improvement in how Windows handles issues like this.

At the simplest, Windows could disable buggy drivers. If Windows determines that a driver is crashing the system at boot and forcing it into a recovery mode, Microsoft could build in more intelligent logic that allows a system to boot without the faulty driver after multiple boot failures.

But the bigger change would be to lock down Windows kernel access to prevent third-party drivers from crashing an entire PC. Ironically, Microsoft tried to do exactly this with Windows Vista but was met with resistance from cybersecurity vendors and EU regulators.

Microsoft tried to implement a feature known at the time as PatchGuard in Windows Vista in 2006, restricting third parties from accessing the kernel. McAfee and Symantec, the big two antivirus companies at the time, opposed Microsoft’s changes, and Symantec even complained to the European Commission. Microsoft eventually backed down, allowing security vendors access to the kernel once again for security monitoring purposes.

Apple eventually took that same step, locking down its macOS operating system in 2020 so that developers could no longer get access to the kernel. “It was definitely the right decision by Apple to deprecate third-party kernel extensions,” says Wardle. “But the road to actually accomplishing that has been fraught with issues.” Apple has had some kernel bugs where security tools running in user mode could still trigger a crash (kernel panic), and Wardle says Apple “has also introduced some privilege execution vulnerabilities, and there are still some other bugs that could allow security tools on Mac to be unloaded by malware.”

Regulatory pressures may still be stopping Microsoft from taking action here. The Wall Street Journal reported over the weekend that “a Microsoft spokesman said it cannot legally wall off its operating system in the same way Apple does because of an understanding it reached with the European Commission following a complaint.” The Journal paraphrases the anonymous spokesperson and also mentions a 2009 agreement to provide security vendors the same level of access to Windows as Microsoft.

Microsoft reached an interoperability agreement with the European Commission in 2009 that was a “public undertaking” to allow developers to get access to technical documentation for building apps on top of Windows. The agreement was formed as part of a deal that included implementing a browser choice screen in Windows and offering special versions of Windows without Internet Explorer bundled into the OS.

The deal to force Microsoft to offer browser choices ended five years later in 2014, and Microsoft also stopped producing its special versions of Windows for Europe. Microsoft now bundles its Edge browser in Windows 11, unchallenged by European regulators.

It’s not clear how long this interoperability agreement was in place, but the European Commission doesn’t seem to believe it’s holding back Microsoft from overhauling Windows security. “Microsoft is free to decide on its business model and to adapt its security infrastructure to respond to threats provided this is done in line with EU competition law,” European Commission spokesperson Lea Zuber says in a statement to The Verge. “Microsoft has never raised any concerns about security with the Commission, either before the recent incident or since.”

The Windows lockdown backlash

Microsoft could attempt to go down the same route as Apple, but the pushback from security vendors like CrowdStrike will be strong. Unlike Apple, Microsoft also competes with CrowdStrike and other security vendors that have made a business out of protecting Windows. Microsoft has its own Defender for Endpoint paid service, which provides similar protections to Windows machines.

CrowdStrike CEO George Kurtz also regularly criticizes Microsoft and its security record and boasts of winning customers away from Microsoft’s own security software. Microsoft has had a series of security mishaps in recent years, so it’s easy and effective for competitors to use these to sell alternatives.

Every time Microsoft tries to lock down Windows in the name of security, it also faces backlash. A special mode in Windows 10 that limited machines to Windows Store apps to avoid malware was confusing and unpopular. Microsoft also left millions of PCs behind with the launch of Windows 11 and its hardware requirements that were designed to improve the security of Windows PCs.

Cloudflare CEO Matthew Prince is already warning about the effects of Microsoft locking down Windows further, framed in a way that Microsoft will favor its own security products if such a scenario were to occur. All of this pushback means Microsoft has a tricky path to tread here if it wants to avoid Windows being at the center of a CrowdStrike-like incident again.

Microsoft is stuck in the middle, with pressure from both sides. But at a time when Microsoft is overhauling security, there has to be some room for security vendors and Microsoft to agree on a better system that will avoid a world of blue screen outages again.

Read More 

Image: Arrowhead Studios

The team at Arrowhead Game Studios is working on the biggest Helldivers 2 update yet. On August 6th, the Escalation of Freedom update will offer players new game modes, enemies, mission objectives, and a much more punishing difficulty. In the announcement trailer, chief creative officer Johan Pilestedt and design director Niklas Malmborg went over the details, sharing that they’re planning on bringing back an old enemy from the original Helldivers to terrorize freedom lovers across the galaxy.

Do you have what it takes, Helldiver? The Escalation of Freedom Major Update begins on August 6, featuring:• Difficulty Level 10• New mission objectives• Bigger enemy outposts• New Bugs and Bots• New environmental hazardsFull briefing: https://t.co/GTTi0R4tc3 pic.twitter.com/h8gEewiu5G— HELLDIVERS™ 2 (@helldivers2) July 23, 2024

The impaler is a large enemy type designed to halt divers’ attempts to esca… err tactical retreats. The update will include a number of new enemy types both terminid and automaton, while remixing existing enemies to give players a new experience. Speaking of experiences, the Helldivers team also shared details on one of the new mission objectives: retrieve the larva. In that mission, divers will be tasked with carrying a screaming terminid larva through enemy territory to reach the extraction point. In addition to the new mission, enemy outposts have been expanded, turning them into massive, more challenging bases for players to capture. Finally, Helldivers 2 will be getting a brand-new difficulty level, combat rating 10, that’s meant to be a significant step up from other difficulties.
Helldivers 2 has enjoyed massive success since its release in February of this year. The game did, however, suffer a couple setbacks with server instability and by requiring PC players to connect to a PSN account. Arrowhead and Sony later walked back that decision after a massive outcry from players, including review bombing the game on Steam.
Despite the game’s success, player numbers on Steam have seen a bit of a dip since February. Likely, this new content refresh, which the developers say is the first step of many they have planned, will be what the game needs to bring back the masses.

Image: Arrowhead Studios

The team at Arrowhead Game Studios is working on the biggest Helldivers 2 update yet. On August 6th, the Escalation of Freedom update will offer players new game modes, enemies, mission objectives, and a much more punishing difficulty. In the announcement trailer, chief creative officer Johan Pilestedt and design director Niklas Malmborg went over the details, sharing that they’re planning on bringing back an old enemy from the original Helldivers to terrorize freedom lovers across the galaxy.

The impaler is a large enemy type designed to halt divers’ attempts to esca… err tactical retreats. The update will include a number of new enemy types both terminid and automaton, while remixing existing enemies to give players a new experience. Speaking of experiences, the Helldivers team also shared details on one of the new mission objectives: retrieve the larva. In that mission, divers will be tasked with carrying a screaming terminid larva through enemy territory to reach the extraction point. In addition to the new mission, enemy outposts have been expanded, turning them into massive, more challenging bases for players to capture. Finally, Helldivers 2 will be getting a brand-new difficulty level, combat rating 10, that’s meant to be a significant step up from other difficulties.

Helldivers 2 has enjoyed massive success since its release in February of this year. The game did, however, suffer a couple setbacks with server instability and by requiring PC players to connect to a PSN account. Arrowhead and Sony later walked back that decision after a massive outcry from players, including review bombing the game on Steam.

Despite the game’s success, player numbers on Steam have seen a bit of a dip since February. Likely, this new content refresh, which the developers say is the first step of many they have planned, will be what the game needs to bring back the masses.

Read More 

Illustration by Alex Castro / The Verge

Amazon’s plan to launch a paid version of Alexa is part of a strategy change to reverse the over $25 billion in losses that its devices business incurred from 2017 to 2021, according to a report from The Wall Street Journal. The AI-supercharged Alexa, which is rumored to cost up to $10 / month, could arrive as soon as this month.
With a potential launch just weeks away, employees reportedly have doubts about whether the new version of Alexa will catch on. A person who worked on the Alexa team told the WSJ that the division is racing toward the deadline to launch the subscription even though “the technology isn’t there.” Amazon’s former head of devices, David Limp, first revealed Amazon’s plans to charge for an improved version of Alexa last year.
As previously reported by Reuters, the “more conversational” Alexa is supposed to come with generative AI features that could allow it to complete multiple tasks in one prompt and learn from users to create routines. But Amazon has reportedly fallen far behind on the project. Last month, a report from Fortune suggested that Alexa isn’t even close to “accomplishing Amazon’s mission of being ‘the world’s best personal assistant’” as the team grapples with technical and organizational issues.
Employees are also concerned about whether people will want to pay for a subscription on top of Amazon’s $139 / year Prime membership, the WSJ reports. With the base versions of AI assistants — like ChatGPT, Google Gemini, and soon, an improved version of Siri — not costing anything to use, Amazon might have a hard time charging for Alexa, especially when the existing version of the voice assistant will stay free.

Illustration by Alex Castro / The Verge

Amazon’s plan to launch a paid version of Alexa is part of a strategy change to reverse the over $25 billion in losses that its devices business incurred from 2017 to 2021, according to a report from The Wall Street Journal. The AI-supercharged Alexa, which is rumored to cost up to $10 / month, could arrive as soon as this month.

With a potential launch just weeks away, employees reportedly have doubts about whether the new version of Alexa will catch on. A person who worked on the Alexa team told the WSJ that the division is racing toward the deadline to launch the subscription even though “the technology isn’t there.” Amazon’s former head of devices, David Limp, first revealed Amazon’s plans to charge for an improved version of Alexa last year.

As previously reported by Reuters, the “more conversational” Alexa is supposed to come with generative AI features that could allow it to complete multiple tasks in one prompt and learn from users to create routines. But Amazon has reportedly fallen far behind on the project. Last month, a report from Fortune suggested that Alexa isn’t even close to “accomplishing Amazon’s mission of being ‘the world’s best personal assistant’” as the team grapples with technical and organizational issues.

Employees are also concerned about whether people will want to pay for a subscription on top of Amazon’s $139 / year Prime membership, the WSJ reports. With the base versions of AI assistants — like ChatGPT, Google Gemini, and soon, an improved version of Siri — not costing anything to use, Amazon might have a hard time charging for Alexa, especially when the existing version of the voice assistant will stay free.

Read More 

Image: Nick Barclay / The Verge

Back in April, Meta teased that it was working on a first for the AI industry: an open-source model with performance that matched the best private models from companies like OpenAI.
Today, that model has arrived. Meta is releasing Llama 3.1, the largest-ever open-source AI model, which the company claims outperforms GPT-4o and Anthropic’s Claude 3.5 Sonnet on several benchmarks. It’s also making the Llama-based Meta AI assistant available in more countries and languages while adding a feature that can generate images based on someone’s specific likeness. CEO Mark Zuckerberg now predicts that Meta AI will be the most widely used assistant by the end of this year, surpassing ChatGPT.
Llama 3.1 is significantly more complex than the smaller Llama 3 models that came out a few months ago. The largest version has 405 billion parameters and was trained with over 16,000 of Nvidia’s ultraexpensive H100 GPUs. Meta isn’t disclosing the cost of developing Llama 3.1, but based on the cost of the Nvidia chips alone, it’s safe to guess it was hundreds of millions of dollars.
So, given the cost, why is Meta continuing to give away Llama with a license that only requires approval from companies with hundreds of millions of users? In a letter published on Meta’s company blog, Zuckerberg argues that open-source AI models will overtake — and are already improving faster than — proprietary models, similar to how Linux became the open-source operating system that powers most phones, servers, and gadgets today.
“An inflection point in the industry where most developers begin to primarily use open source”
He compares Meta’s investment in open-source AI to its earlier Open Compute Project, which he says saved the company “billions” by having outside companies like HP help improve and standardize Meta’s data center designs as it was building out its own capacity. Looking ahead, he expects the same dynamic to play out with AI, writing, “I believe the Llama 3.1 release will be an inflection point in the industry where most developers begin to primarily use open source.”
To help get Llama 3.1 out into the world, Meta is working with more than two dozen companies, including Microsoft, Amazon, Google, Nvidia, and Databricks, to help developers deploy their own versions. Meta claims that Llama 3.1 costs roughly half that of OpenAI’s GPT-4o to run in production. It’s releasing the model weights so that companies can train it on custom data and tune it to their liking.

Chart: Meta
Gemini isn’t included in these benchmark comparisons because Meta had a hard time using Google’s APIs to replicate its previously stated results, according to Meta spokesperson Jon Carvill.

Chart: Meta
A list of Meta’s key partners and the capabilities they offer for deploying Llama 3.1.

Unsurprisingly, Meta isn’t saying much about the data it used to train Llama 3.1. The people who work at AI companies say they don’t disclose this information because it’s a trade secret, while critics say it’s a tactic to delay the inevitable onslaught of copyright lawsuits that are coming.
What Meta will say is that it used synthetic data, or data generated by a model rather than humans, to have the 405-billion parameter version of Llama 3.1 improve the smaller 70 billion and 8 billion versions. Ahmad Al-Dahle, Meta’s VP of generative AI, predicts that Llama 3.1 will be popular with developers as “a teacher for smaller models that are then deployed” in a “more cost effective way.”
When I ask if Meta agrees with the growing consensus that the industry is running out of quality training data for models, Al-Dahle suggests there is a ceiling coming, though it may be farther out than some think. “We definitely think we have a few more [training] runs,” he says. “But it’s difficult to say.”
For the first time, Meta’s red teaming (or adversarial testing) of Llama 3.1 included looking for potential cybersecurity and biochemical use cases. Another reason to test the model more strenuously is what Meta is describing as emerging “agentic” behaviors.
For example, Al-Dahle tells me that Llama 3.1 is capable of integrating with a search engine API to “retrieve information from the internet based on a complex query and call multiple tools in succession in order to complete your tasks.” Another example he gives is asking the model to plot the number of homes sold in the United States over the last five years. “It can retrieve the [web] search for you and generate the Python code and execute it.”

Meta’s own implementation of Llama is its AI assistant, which is positioned as a general-purpose chatbot like ChatGPT and can be found in just about every part of Instagram, Facebook, and WhatsApp. Starting this week, Llama 3.1 will be first accessible through WhatsApp and the Meta AI website in the US, followed by Instagram and Facebook in the coming weeks. It’s being updated to support new languages as well, including French, German, Hindi, Italian, and Spanish.
While Llama 3.1’s most advanced 405-billion parameter model is free to use in Meta AI, the assistant will switch you to the more scaled-back 70-billion model after surpassing an unspecified number of prompts in a given week. This suggests the 405-billion model is too expensive for Meta to run at full scale. Spokesperson Jon Carvill tells me the company will provide more information on the prompt threshold after it assesses early usage.

Image: Meta

A new “Imagine Me” feature in Meta AI scans your face through your phone’s camera to then let you insert your likeness into images it generates. By capturing your likeness this way and not through the photos in your profile, Meta is hopefully avoiding the creation of a deepfake machine. The company sees demand for people wanting to create more kinds of AI media and share it to their feeds, even if that means blurring the line between what is discernibly real and not.
Meta AI is also coming to the Quest headset in the coming weeks, replacing its voice command interface. Like its implementation in the Meta Ray-Ban glasses, you’ll be able to use Meta AI on the Quest to identify and learn about what you’re looking at while in the headset’s passthrough mode that shows the real world through the display.
“I think the entire industry is still early on its path towards product market fit”
Aside from Zuckerberg’s prediction that Meta AI will be the most-used chatbot by the end of this year (ChatGPT has over 100 million users), Meta has yet to share any usage numbers for its assistant. “I think the entire industry is still early on its path towards product market fit,” Al-Dahle says. Even with how overhyped AI can already feel, it’s clear that Meta and other players think the race is just beginning.

Image: Nick Barclay / The Verge

Back in April, Meta teased that it was working on a first for the AI industry: an open-source model with performance that matched the best private models from companies like OpenAI.

Today, that model has arrived. Meta is releasing Llama 3.1, the largest-ever open-source AI model, which the company claims outperforms GPT-4o and Anthropic’s Claude 3.5 Sonnet on several benchmarks. It’s also making the Llama-based Meta AI assistant available in more countries and languages while adding a feature that can generate images based on someone’s specific likeness. CEO Mark Zuckerberg now predicts that Meta AI will be the most widely used assistant by the end of this year, surpassing ChatGPT.

Llama 3.1 is significantly more complex than the smaller Llama 3 models that came out a few months ago. The largest version has 405 billion parameters and was trained with over 16,000 of Nvidia’s ultraexpensive H100 GPUs. Meta isn’t disclosing the cost of developing Llama 3.1, but based on the cost of the Nvidia chips alone, it’s safe to guess it was hundreds of millions of dollars.

So, given the cost, why is Meta continuing to give away Llama with a license that only requires approval from companies with hundreds of millions of users? In a letter published on Meta’s company blog, Zuckerberg argues that open-source AI models will overtake — and are already improving faster than — proprietary models, similar to how Linux became the open-source operating system that powers most phones, servers, and gadgets today.

He compares Meta’s investment in open-source AI to its earlier Open Compute Project, which he says saved the company “billions” by having outside companies like HP help improve and standardize Meta’s data center designs as it was building out its own capacity. Looking ahead, he expects the same dynamic to play out with AI, writing, “I believe the Llama 3.1 release will be an inflection point in the industry where most developers begin to primarily use open source.”

To help get Llama 3.1 out into the world, Meta is working with more than two dozen companies, including Microsoft, Amazon, Google, Nvidia, and Databricks, to help developers deploy their own versions. Meta claims that Llama 3.1 costs roughly half that of OpenAI’s GPT-4o to run in production. It’s releasing the model weights so that companies can train it on custom data and tune it to their liking.

Unsurprisingly, Meta isn’t saying much about the data it used to train Llama 3.1. The people who work at AI companies say they don’t disclose this information because it’s a trade secret, while critics say it’s a tactic to delay the inevitable onslaught of copyright lawsuits that are coming.

What Meta will say is that it used synthetic data, or data generated by a model rather than humans, to have the 405-billion parameter version of Llama 3.1 improve the smaller 70 billion and 8 billion versions. Ahmad Al-Dahle, Meta’s VP of generative AI, predicts that Llama 3.1 will be popular with developers as “a teacher for smaller models that are then deployed” in a “more cost effective way.”

When I ask if Meta agrees with the growing consensus that the industry is running out of quality training data for models, Al-Dahle suggests there is a ceiling coming, though it may be farther out than some think. “We definitely think we have a few more [training] runs,” he says. “But it’s difficult to say.”

For the first time, Meta’s red teaming (or adversarial testing) of Llama 3.1 included looking for potential cybersecurity and biochemical use cases. Another reason to test the model more strenuously is what Meta is describing as emerging “agentic” behaviors.

For example, Al-Dahle tells me that Llama 3.1 is capable of integrating with a search engine API to “retrieve information from the internet based on a complex query and call multiple tools in succession in order to complete your tasks.” Another example he gives is asking the model to plot the number of homes sold in the United States over the last five years. “It can retrieve the [web] search for you and generate the Python code and execute it.”

Meta’s own implementation of Llama is its AI assistant, which is positioned as a general-purpose chatbot like ChatGPT and can be found in just about every part of Instagram, Facebook, and WhatsApp. Starting this week, Llama 3.1 will be first accessible through WhatsApp and the Meta AI website in the US, followed by Instagram and Facebook in the coming weeks. It’s being updated to support new languages as well, including French, German, Hindi, Italian, and Spanish.

While Llama 3.1’s most advanced 405-billion parameter model is free to use in Meta AI, the assistant will switch you to the more scaled-back 70-billion model after surpassing an unspecified number of prompts in a given week. This suggests the 405-billion model is too expensive for Meta to run at full scale. Spokesperson Jon Carvill tells me the company will provide more information on the prompt threshold after it assesses early usage.

A new “Imagine Me” feature in Meta AI scans your face through your phone’s camera to then let you insert your likeness into images it generates. By capturing your likeness this way and not through the photos in your profile, Meta is hopefully avoiding the creation of a deepfake machine. The company sees demand for people wanting to create more kinds of AI media and share it to their feeds, even if that means blurring the line between what is discernibly real and not.

Meta AI is also coming to the Quest headset in the coming weeks, replacing its voice command interface. Like its implementation in the Meta Ray-Ban glasses, you’ll be able to use Meta AI on the Quest to identify and learn about what you’re looking at while in the headset’s passthrough mode that shows the real world through the display.

Aside from Zuckerberg’s prediction that Meta AI will be the most-used chatbot by the end of this year (ChatGPT has over 100 million users), Meta has yet to share any usage numbers for its assistant. “I think the entire industry is still early on its path towards product market fit,” Al-Dahle says. Even with how overhyped AI can already feel, it’s clear that Meta and other players think the race is just beginning.

Read More 

Image: Toyota

Toyota Research Institute (TRI) and Stanford are plugging AI into two Supras that pull off Formula Drift-style tandem driving — but they’re looking for something more important than style points.
In a press release, TRI’s VP of human interactive driving, Avinash Balachandran, says that drifting two cars in tandem autonomously is a “milestone” and has “far-reaching implications for building advanced safety systems” in future passenger vehicles.
Beyond the impressive showing, which can be seen in a video, professor Chris Gerdes, who codirects the Center for Automotive Research at Stanford, says the physics of drifting is similar to the behavior of cars on snow or ice. Balachandran adds that the tech can kick in precisely in time to manage a driver’s loss of control, just like expert drifters. The system can solve and re-solve a problem up to 50 times per second to decide what steering, throttle, and brake commands work best in the conditions.

The dueling and drifting modified GR Supras use AI that learns from each trip on the track. TRI developed the lead car’s control mechanisms, while Stanford’s School of Engineering made the AI vehicle models and algorithms for the chase car designed to copy (and not collide with) the other. The vehicles communicate through Wi-Fi and were tuned by GReddy and Toyota Racing Development. By the way, self-drifting cars aren’t new to Stanford; a group of researchers built a DeLorean with that capability in 2015.

Image: Toyota

Toyota Research Institute (TRI) and Stanford are plugging AI into two Supras that pull off Formula Drift-style tandem driving — but they’re looking for something more important than style points.

In a press release, TRI’s VP of human interactive driving, Avinash Balachandran, says that drifting two cars in tandem autonomously is a “milestone” and has “far-reaching implications for building advanced safety systems” in future passenger vehicles.

Beyond the impressive showing, which can be seen in a video, professor Chris Gerdes, who codirects the Center for Automotive Research at Stanford, says the physics of drifting is similar to the behavior of cars on snow or ice. Balachandran adds that the tech can kick in precisely in time to manage a driver’s loss of control, just like expert drifters. The system can solve and re-solve a problem up to 50 times per second to decide what steering, throttle, and brake commands work best in the conditions.

The dueling and drifting modified GR Supras use AI that learns from each trip on the track. TRI developed the lead car’s control mechanisms, while Stanford’s School of Engineering made the AI vehicle models and algorithms for the chase car designed to copy (and not collide with) the other. The vehicles communicate through Wi-Fi and were tuned by GReddy and Toyota Racing Development. By the way, self-drifting cars aren’t new to Stanford; a group of researchers built a DeLorean with that capability in 2015.

Read More 

President Joe Biden delivers remarks during a climate event at the White House complex on November 14th, 2023, in Washington, DC. | Photo by Win McNamee / Getty Images

Greenhouse gas emissions are falling in the US thanks to major investments in clean energy and transportation — but it’s not happening fast enough, a new report shows. As a result, the US isn’t on track to meet the ambitious climate goals set by the Biden administration as the president enters his final months in office. Political upheaval and soaring electricity demand from AI, crypto, and EVs risk pushing those goals further out of reach.
The latest forecast on clean energy and climate progress in the US, from research firm Rhodium Group, paints a sobering picture of what’s at stake this election season. The Biden administration has managed to push through landmark climate legislation that could get rid of a significant chunk of America’s carbon footprint. But those policies are in jeopardy with a conservative supermajority in the Supreme Court and the prospect of another Donald Trump presidency. Even if climate policies survive, they’re not enough to keep up with new technologies that are making it harder to rein in greenhouse gas emissions.
Even if climate policies survive, they’re not enough to keep up with new technologies that are making it harder to rein in greenhouse gas emissions
Rhodium modeled out future outcomes based on federal and state policies and different economic factors. This year, Rhodium also had to weigh the potential impact of a boom in energy-hungry data centers used to train AI or mine crypto. They’re expected to add even more pressure to an aging grid that was already bracing for rising electricity demand from EV charging and manufacturing.
By 2035, electricity demand could be 24 to 29 percent higher than it was in 2023, according to Rhodium. Transportation would account for nearly half of that growth. But data centers have also become a major player, expected to make up 22 percent of the added electricity demand. And unlike EVs, which displace existing fossil fuel demand for transportation, data centers generally represent completely new energy demand.
Based on current policies in play, US greenhouse gas emissions would fall by between 32 to 43 percent by the end of the decade compared to a 2005 baseline, according to Rhodium’s estimates. But in a scenario with data center electricity demand growing nearly threefold by 2035, power sector emissions would rise 56 percent compared to Rhodium’s central estimate. The US could potentially meet its goal several years late, reducing emissions between 38 to 56 percent by 2035.

That entails steeper annual drops in pollution than the US has managed over the past couple of decades. And it only works with new policies set under the Biden administration aimed at cleaning up emissions from the power sector, transportation, and oil and gas fields.
Shortly after stepping into office, Biden recommitted the US to the Paris agreement to stop climate change — a deal Trump tried to exit. As part of the agreement, Biden set a 2030 goal for slashing US climate pollution in half compared to a 2005 baseline. It was ambitious but roughly in line with what research shows is necessary to meet the overarching goals of the Paris agreement.
But each of those policies faces legal challenges. Landmark decisions from the Supreme Court since the Trump administration have curbed federal agencies’ authority to regulate industry. Courts aside, climate policies could unravel if Trump is reelected.
The GOP’s stance on energy in its official platform is to “drill, baby, drill” (in capital letters)
Trump says he’d try to pull the US out of the Paris agreement again, and the GOP’s stance on energy in its official platform is to “drill, baby, drill” (in capital letters). If Republicans gain control of the White House and Congress come November, they could also work to dismantle the Inflation Reduction Act signed into law in 2022 as the biggest investment in climate and clean energy to date in the US. Notably, crypto and Silicon Valley bigwigs have started to cozy up with the Trump campaign with the GOP platform pledging to “champion innovation” (AKA cut back on regulation) in crypto and AI.
“The person who wins the White House in November will have the chance to leave their mark on [greenhouse gas] regulations, either defending them in court and pursuing additional action or revoking them wholesale,” Rhodium’s analysis says.
At this point, the US likely has to pass more legislation to reach its current 2030 climate goals. That was an uphill battle even during the Biden administration. And whatever the US does on climate moving forward, of course, has a global impact. The US has more data centers than any other country and is the biggest hub for Bitcoin mining. It’s also still the world’s top oil and gas producer and the country with the most carbon emissions after China.
That doesn’t mean the US has to sacrifice economic growth for a healthy climate, the Rhodium report points out. Economic strength and greenhouse gas emissions have “largely decoupled,” it says, with the US economy growing faster than its carbon footprint in eight of the past 10 years.

President Joe Biden delivers remarks during a climate event at the White House complex on November 14th, 2023, in Washington, DC. | Photo by Win McNamee / Getty Images

Greenhouse gas emissions are falling in the US thanks to major investments in clean energy and transportation — but it’s not happening fast enough, a new report shows. As a result, the US isn’t on track to meet the ambitious climate goals set by the Biden administration as the president enters his final months in office. Political upheaval and soaring electricity demand from AI, crypto, and EVs risk pushing those goals further out of reach.

The latest forecast on clean energy and climate progress in the US, from research firm Rhodium Group, paints a sobering picture of what’s at stake this election season. The Biden administration has managed to push through landmark climate legislation that could get rid of a significant chunk of America’s carbon footprint. But those policies are in jeopardy with a conservative supermajority in the Supreme Court and the prospect of another Donald Trump presidency. Even if climate policies survive, they’re not enough to keep up with new technologies that are making it harder to rein in greenhouse gas emissions.

Rhodium modeled out future outcomes based on federal and state policies and different economic factors. This year, Rhodium also had to weigh the potential impact of a boom in energy-hungry data centers used to train AI or mine crypto. They’re expected to add even more pressure to an aging grid that was already bracing for rising electricity demand from EV charging and manufacturing.

By 2035, electricity demand could be 24 to 29 percent higher than it was in 2023, according to Rhodium. Transportation would account for nearly half of that growth. But data centers have also become a major player, expected to make up 22 percent of the added electricity demand. And unlike EVs, which displace existing fossil fuel demand for transportation, data centers generally represent completely new energy demand.

Based on current policies in play, US greenhouse gas emissions would fall by between 32 to 43 percent by the end of the decade compared to a 2005 baseline, according to Rhodium’s estimates. But in a scenario with data center electricity demand growing nearly threefold by 2035, power sector emissions would rise 56 percent compared to Rhodium’s central estimate. The US could potentially meet its goal several years late, reducing emissions between 38 to 56 percent by 2035.

That entails steeper annual drops in pollution than the US has managed over the past couple of decades. And it only works with new policies set under the Biden administration aimed at cleaning up emissions from the power sector, transportation, and oil and gas fields.

Shortly after stepping into office, Biden recommitted the US to the Paris agreement to stop climate change — a deal Trump tried to exit. As part of the agreement, Biden set a 2030 goal for slashing US climate pollution in half compared to a 2005 baseline. It was ambitious but roughly in line with what research shows is necessary to meet the overarching goals of the Paris agreement.

But each of those policies faces legal challenges. Landmark decisions from the Supreme Court since the Trump administration have curbed federal agencies’ authority to regulate industry. Courts aside, climate policies could unravel if Trump is reelected.

Trump says he’d try to pull the US out of the Paris agreement again, and the GOP’s stance on energy in its official platform is to “drill, baby, drill” (in capital letters). If Republicans gain control of the White House and Congress come November, they could also work to dismantle the Inflation Reduction Act signed into law in 2022 as the biggest investment in climate and clean energy to date in the US. Notably, crypto and Silicon Valley bigwigs have started to cozy up with the Trump campaign with the GOP platform pledging to “champion innovation” (AKA cut back on regulation) in crypto and AI.

“The person who wins the White House in November will have the chance to leave their mark on [greenhouse gas] regulations, either defending them in court and pursuing additional action or revoking them wholesale,” Rhodium’s analysis says.

At this point, the US likely has to pass more legislation to reach its current 2030 climate goals. That was an uphill battle even during the Biden administration. And whatever the US does on climate moving forward, of course, has a global impact. The US has more data centers than any other country and is the biggest hub for Bitcoin mining. It’s also still the world’s top oil and gas producer and the country with the most carbon emissions after China.

That doesn’t mean the US has to sacrifice economic growth for a healthy climate, the Rhodium report points out. Economic strength and greenhouse gas emissions have “largely decoupled,” it says, with the US economy growing faster than its carbon footprint in eight of the past 10 years.

Read More 

The Cruise Origin in 2020. | Image: Vjeran Pavic / The Verge

General Motors told reporters on Tuesday that it has indefinitely suspended production of its self-driving Cruise Origin robotaxi, reports The New York Times. The company will refocus on using the Chevy Bolt EV, which Cruise has already been using. A modified version will go into production next year.
The Origin created “regulatory uncertainty” owing to its design, GM CEO Mary Barra said in her letter to shareholders today. It was to be purpose-built for self-driving, with no steering wheel or pedals, no clear front or rear, and no driver; but estimates put each vehicle’s cost in the hundreds of thousands. The company said in September of last year that it was “just days away” from regulators approving the Origin for mass production.

Image: Vjeran Pavic / The Verge
The Origin’s interior was solely for passengers.

Then, a month later, Cruise’s robotaxis were banned in California after multiple incidents, including one where a driverless Cruise car hit and dragged a San Francisco pedestrian. Since then, GM, which has lost several billions on the company already, has had to keep Cruise afloat while it reorients.
Meanwhile, Cruise has been testing its Chevy Bolt EV robotaxis in Dallas, Houston, and Phoenix, using human safety drivers. The company resumed testing earlier this year with manually driven vehicles and no passengers.

The Cruise Origin in 2020. | Image: Vjeran Pavic / The Verge

General Motors told reporters on Tuesday that it has indefinitely suspended production of its self-driving Cruise Origin robotaxi, reports The New York Times. The company will refocus on using the Chevy Bolt EV, which Cruise has already been using. A modified version will go into production next year.

The Origin created “regulatory uncertainty” owing to its design, GM CEO Mary Barra said in her letter to shareholders today. It was to be purpose-built for self-driving, with no steering wheel or pedals, no clear front or rear, and no driver; but estimates put each vehicle’s cost in the hundreds of thousands. The company said in September of last year that it was “just days away” from regulators approving the Origin for mass production.

Image: Vjeran Pavic / The Verge
The Origin’s interior was solely for passengers.

Then, a month later, Cruise’s robotaxis were banned in California after multiple incidents, including one where a driverless Cruise car hit and dragged a San Francisco pedestrian. Since then, GM, which has lost several billions on the company already, has had to keep Cruise afloat while it reorients.

Meanwhile, Cruise has been testing its Chevy Bolt EV robotaxis in Dallas, Houston, and Phoenix, using human safety drivers. The company resumed testing earlier this year with manually driven vehicles and no passengers.

Read More 

The gray model may lack some of the Game Boy’s vintage vibes like the other colors on offer, but the design homages are there once you look close.

Between the Analogue Pocket and the proliferation of game emulators on Apple’s App Store, retro gaming is more than chic in 2024. So, it’s not terribly surprising that boutique keyboard maker Angry Miao is getting in on some of that sweet nostalgia with its latest mechanical board, the Game Boy-inspired AM RGB 65.
But the real surprise is that what at first blush looks like a near-$600 parts bin rehash of an existing board is actually Angry Miao’s best mech to date.

There’s a lot of Cyberboard DNA here, but oh so much better.

The RGB 65 is a 65 percent layout keyboard, but its large dot matrix LED forehead means it takes up almost as much space as most 75 percent keyboards. It looks a whole lot like a shrunken-down Cyberboard, but thankfully, it takes more design cues from one of the most beloved handheld devices in human history than from a frequently mocked rolling dumpster.
The retro vibes are most pronounced on the white and purple versions, but all colors of the RGB 65 have small homages to Nintendo’s industrial design from the original “DMG-01” Game Boy. This includes a curved bottom-right corner with six slats mimicking the speaker grill, a negative relief D-pad cutout surrounding the USB-C port, and mock B and A buttons on the bottom, which replace the wireless charging pad found on other AM models. Another aesthetic flourish is a Galaga pixel art graphic adorning the plastic antenna window that lets the Bluetooth 5.1 and 2.4GHz wireless connections through the all-metal chassis.

Typing on the RGB 65 is a breath of fresh air compared to previous Angry Miao boards. The marbly sound signature is still present, but the new Icy Silver V2 linear switches with their nylon housings and revised internal design give the RGB 65 a deeper, much more “thocky” sound than any other AM keyboard I’ve tried.

The new Icy Silver V2 Pro switches are part of what gives the RGB 65 it’s thocky tone. The cutouts in the Polyoxymethylene (POM) plate combined with adjustable leaf springs give the keys a bit of flex while typing.

This board is actually bigger than my personal Meletrix Zoom75, which is a 75-percent keyboard with more keys in a smaller footprint.

I’m relieved to see Angry Miao making strides in the sound and typing feel departments. While I’ve personally enjoyed the sound of most prior AM keyboards, it’s all very subjective. And the “meta” around what’s popular in the custom mechanical keyboard scene continues to evolve. I think Angry Miao is doing a fine job keeping up with trends here, and it needs to. This is a very expensive keyboard that should feel good and sound great out of the box, especially when you consider it rehashes a big design element from the Cyberboard.
The 200-LED dot matrix panel on the Cyberboard always looked charming, but while it’s recycled on the RGB 65, it’s better implemented and slightly more practical. Instead of being angled away from you and upside-down like on the Cyberboard, it’s flat on the main deck and easy to see. It’s still mostly a flashy novelty to go along with the RGB light show under the keys, but you can use Angry Miao’s site to do some deep customization or download fully baked presets from the community. It’s fun to tinker with, even if I end up using it as a big desk clock most of the time.

Image: Angry Miao
The RGB 65’s white version seems like the correct color for the most nostalgia.

Image: Angry Miao
And the violet model has a unique surface treatment, made using multicolor anodizing, though it costs about $40 extra.

But the most underwhelming bit of the RGB 65 is this “cyber gray” color combo, at least in comparison to the white and “violet fury” options. It looks slick in its own right, sure, but the other two feel like where it’s at for the maximum Game Boy vibes. Perhaps if the gray model used opaque light gray or charcoal black keycaps, it could channel a tiny bit of the Game Boy Advance SP or Game Boy Pocket.

Like most Angry Miao keyboards, the RGB 65 certainly has desk presence.

Either way, the RGB 65 desperately needs some see-through atomic purple or jungle green action (which is another ongoing trend), so I hope Angry Miao considers a cheaper all-plastic version in the future. I’d even be willing to sacrifice a bit of that thock if Angry Miao can allow itself to not overengineer every little thing from aluminum.
The AM RGB 65’s Kickstarter campaign launches today, starting at $449 for a bare-bones kit (no keycaps or switches) and tops out at $598 for a full prebuilt bundle in violet. Angry Miao is running an early-bird 8 percent discount for the first 48 hours of the campaign.
Photography by Antonio G. Di Benedetto / The Verge

The gray model may lack some of the Game Boy’s vintage vibes like the other colors on offer, but the design homages are there once you look close.

Between the Analogue Pocket and the proliferation of game emulators on Apple’s App Store, retro gaming is more than chic in 2024. So, it’s not terribly surprising that boutique keyboard maker Angry Miao is getting in on some of that sweet nostalgia with its latest mechanical board, the Game Boy-inspired AM RGB 65.

But the real surprise is that what at first blush looks like a near-$600 parts bin rehash of an existing board is actually Angry Miao’s best mech to date.

The RGB 65 is a 65 percent layout keyboard, but its large dot matrix LED forehead means it takes up almost as much space as most 75 percent keyboards. It looks a whole lot like a shrunken-down Cyberboard, but thankfully, it takes more design cues from one of the most beloved handheld devices in human history than from a frequently mocked rolling dumpster.

The retro vibes are most pronounced on the white and purple versions, but all colors of the RGB 65 have small homages to Nintendo’s industrial design from the original “DMG-01” Game Boy. This includes a curved bottom-right corner with six slats mimicking the speaker grill, a negative relief D-pad cutout surrounding the USB-C port, and mock B and A buttons on the bottom, which replace the wireless charging pad found on other AM models. Another aesthetic flourish is a Galaga pixel art graphic adorning the plastic antenna window that lets the Bluetooth 5.1 and 2.4GHz wireless connections through the all-metal chassis.

Typing on the RGB 65 is a breath of fresh air compared to previous Angry Miao boards. The marbly sound signature is still present, but the new Icy Silver V2 linear switches with their nylon housings and revised internal design give the RGB 65 a deeper, much more “thocky” sound than any other AM keyboard I’ve tried.

I’m relieved to see Angry Miao making strides in the sound and typing feel departments. While I’ve personally enjoyed the sound of most prior AM keyboards, it’s all very subjective. And the “meta” around what’s popular in the custom mechanical keyboard scene continues to evolve. I think Angry Miao is doing a fine job keeping up with trends here, and it needs to. This is a very expensive keyboard that should feel good and sound great out of the box, especially when you consider it rehashes a big design element from the Cyberboard.

The 200-LED dot matrix panel on the Cyberboard always looked charming, but while it’s recycled on the RGB 65, it’s better implemented and slightly more practical. Instead of being angled away from you and upside-down like on the Cyberboard, it’s flat on the main deck and easy to see. It’s still mostly a flashy novelty to go along with the RGB light show under the keys, but you can use Angry Miao’s site to do some deep customization or download fully baked presets from the community. It’s fun to tinker with, even if I end up using it as a big desk clock most of the time.

But the most underwhelming bit of the RGB 65 is this “cyber gray” color combo, at least in comparison to the white and “violet fury” options. It looks slick in its own right, sure, but the other two feel like where it’s at for the maximum Game Boy vibes. Perhaps if the gray model used opaque light gray or charcoal black keycaps, it could channel a tiny bit of the Game Boy Advance SP or Game Boy Pocket.

Either way, the RGB 65 desperately needs some see-through atomic purple or jungle green action (which is another ongoing trend), so I hope Angry Miao considers a cheaper all-plastic version in the future. I’d even be willing to sacrifice a bit of that thock if Angry Miao can allow itself to not overengineer every little thing from aluminum.

The AM RGB 65’s Kickstarter campaign launches today, starting at $449 for a bare-bones kit (no keycaps or switches) and tops out at $598 for a full prebuilt bundle in violet. Angry Miao is running an early-bird 8 percent discount for the first 48 hours of the campaign.

Photography by Antonio G. Di Benedetto / The Verge

Read More 

Photo: Andrey Rudakov/Bloomberg via Getty Images

It all started with a configuration file. A seemingly routine update, the kind that happens hundreds of times a year, to millions of computers around the world. But last week that update crashed 8.5 million computers, and wreaked havoc on banks, airlines, schools, and more.
On this episode of The Vergecast, The Verge’s Tom Warren joins the show to talk about the story and legacy of the CrowdStrike crash. We talk about what exactly happened, how CrowdStrike and Microsoft scrambled to fix it, and whether we’re due for a reckoning over how much we rely on complex and fragile technology. (Don’t forget to subscribe to Notepad!)

After that, we chat with The Verge’s Victoria Song along with Adrian Hon, a writer and game designer who spent many years as the co-creator of the popular fitness game Zombies, Run. Hon tells us why he set out to build a game without streaks and aggressive notifications, why more data isn’t necessarily better, and why zombies are such a good motivator.
Finally, we take a question from the Vergecast Hotline (866-VERGE11, or email us at vergecast@theverge.com!) about portable game consoles for new parents. Because sometimes there’s a tiny creature sleeping on you, and you’ve got games to play.
If you want to more know about everything we discuss in this episode, here are some links to get you started, beginning with CrowdStrike:

CrowdStrike and Microsoft: all the latest news on the global IT outage
Major Windows BSOD issue hits banks, airlines, and TV broadcasters
What is CrowdStrike, and what happened?
CrowdStrike’s faulty update crashed 8.5 million Windows devices, says Microsoft
CrowdStrike outage: Photos, videos, and tales of IT workers fixing BSODs

And on Zombies, Run and fitness:

Zombies, Run
Adrian Hon’s Substack
Finally, the Apple Watch will let you rest
This walking app let me whack my co-workers with a baseball bat
Ignore your fitness tracker and walk to Mordor instead

And on handheld gaming:

Backbone One review: the best mobile gaming controller yet
Handheld consoles are the future of gaming
Holedown

Photo: Andrey Rudakov/Bloomberg via Getty Images

It all started with a configuration file. A seemingly routine update, the kind that happens hundreds of times a year, to millions of computers around the world. But last week that update crashed 8.5 million computers, and wreaked havoc on banks, airlines, schools, and more.

On this episode of The Vergecast, The Verge’s Tom Warren joins the show to talk about the story and legacy of the CrowdStrike crash. We talk about what exactly happened, how CrowdStrike and Microsoft scrambled to fix it, and whether we’re due for a reckoning over how much we rely on complex and fragile technology. (Don’t forget to subscribe to Notepad!)

After that, we chat with The Verge’s Victoria Song along with Adrian Hon, a writer and game designer who spent many years as the co-creator of the popular fitness game Zombies, Run. Hon tells us why he set out to build a game without streaks and aggressive notifications, why more data isn’t necessarily better, and why zombies are such a good motivator.

Finally, we take a question from the Vergecast Hotline (866-VERGE11, or email us at vergecast@theverge.com!) about portable game consoles for new parents. Because sometimes there’s a tiny creature sleeping on you, and you’ve got games to play.

If you want to more know about everything we discuss in this episode, here are some links to get you started, beginning with CrowdStrike:

CrowdStrike and Microsoft: all the latest news on the global IT outage
Major Windows BSOD issue hits banks, airlines, and TV broadcasters
What is CrowdStrike, and what happened?
CrowdStrike’s faulty update crashed 8.5 million Windows devices, says Microsoft
CrowdStrike outage: Photos, videos, and tales of IT workers fixing BSODs

And on Zombies, Run and fitness:

Zombies, Run
Adrian Hon’s Substack
Finally, the Apple Watch will let you rest
This walking app let me whack my co-workers with a baseball bat
Ignore your fitness tracker and walk to Mordor instead

And on handheld gaming:

Backbone One review: the best mobile gaming controller yet
Handheld consoles are the future of gaming
Holedown

Read More 

Image: Hugo Herrera / The Verge

Nearly a decade ago, Uber revealed that customers are more willing to pay surge prices when their phones are about to die. That data point sparked a pervasive belief that Uber and other ridehailing apps use algorithms to tailor each passenger’s price to their behavior and willingness to pay — a practice known as “surveillance pricing.” A related myth — that airlines use your search history to jack up prices for flights you’re interested in — is similarly ubiquitous even though it’s not true.
Surveillance pricing may not be widespread yet, but experts say it’s on the horizon, and consultants are starting to offer surveillance pricing models to businesses across various industries. The Federal Trade Commission announced today that it’s ordering eight companies that offer AI surveillance pricing products and services to turn over information about the impact these pricing schemes have on privacy, competition, and consumer protection, the FTC announced today.
“The technology is there, the incentives are there, the data certainly is there to do this targeting.”
Surveillance pricing — often called “dynamic pricing,” “personalized pricing,” or “price optimization” — involves offering individual consumers different prices for the same products based on a combination of factors, including the device they’re shopping on, their location, demographic information, credit history, and browsing and shopping history.
“Firms that harvest Americans’ personal data can put people’s privacy at risk. Now firms could be exploiting this vast trove of personal information to charge people higher prices,” FTC Chair Lina Khan said in a statement. “Americans deserve to know whether businesses are using detailed consumer data to deploy surveillance pricing, and the FTC’s inquiry will shed light on this shadowy ecosystem of pricing middlemen.”
Companies have already experimented with implementing Uber-style surge pricing. JetBlue quietly implemented “peak” and “off-peak” pricing to its checked bag fees in March. Walmart plans on bringing digital price tags to 2,300 stores over the next several years that will let it change products’ prices depending on a number of factors, including the weather, whether items are closer to their expiration date, and more. Wendy’s said it would roll out “dynamic pricing” in 2025. Amazon had a secret algorithm called Project Nessie that let it gauge how much it could raise its prices before its competitors took notice and changed theirs, too.

These pricing models, while variable, are a bit different from surveillance pricing, which relies on using an individual consumer’s information to determine how much they’re willing to pay for a specific product.
The FTC orders were sent to Mastercard, Revionics, Bloomreach, Chase, Task, Pros, Accenture, and McKinsey & Co., the FTC announced in a press release. They were issued under the FTC’s 6(b) authority, which lets it conduct studies without any specific law enforcement purpose.
“These eight firms advertise their use of AI and machine learning as part of the services they offer, often claiming they can use historical and real-time customer information to drive unique purchasing experiences,” an FTC official said on a call with reporters ahead of the announcement, adding that none of the firms issued orders have been accused of any wrongdoing. “What we’re trying to do is lift the hood on opaque practices that have the potential to profoundly reshape how consumers purchase goods and services. We want to make sure that surveillance pricing isn’t resulting in harm to certain communities, like women and rural consumers.”
The companies are being asked to turn over information on:

Types of surveillance pricing products and services they’ve produced, developed, or licensed to a third party
Data sources used for each product or service
Who they’re offering surveillance pricing products and services to
How this impacts surveilled consumers and pricing

On a call with reporters, an FTC official said a study is necessary because the extent to which businesses use surveillance pricing is currently unknown. Media reports suggest that companies across a wide variety of sectors — including grocery stores, restaurant chains, travel, and hospitality — are considering implementing surveillance pricing or have already done so.
“The technology is there, the incentives are there, the data certainly is there to do this targeting,” the FTC official told reporters. “And now we actually see this industry of consultancies potentially offer to help companies personalize pricing, which gives us a strong reason to believe that this practice is likely to grow.”

Image: Hugo Herrera / The Verge

Nearly a decade ago, Uber revealed that customers are more willing to pay surge prices when their phones are about to die. That data point sparked a pervasive belief that Uber and other ridehailing apps use algorithms to tailor each passenger’s price to their behavior and willingness to pay — a practice known as “surveillance pricing.” A related myth — that airlines use your search history to jack up prices for flights you’re interested in — is similarly ubiquitous even though it’s not true.

Surveillance pricing may not be widespread yet, but experts say it’s on the horizon, and consultants are starting to offer surveillance pricing models to businesses across various industries. The Federal Trade Commission announced today that it’s ordering eight companies that offer AI surveillance pricing products and services to turn over information about the impact these pricing schemes have on privacy, competition, and consumer protection, the FTC announced today.

Surveillance pricing — often called “dynamic pricing,” “personalized pricing,” or “price optimization” — involves offering individual consumers different prices for the same products based on a combination of factors, including the device they’re shopping on, their location, demographic information, credit history, and browsing and shopping history.

“Firms that harvest Americans’ personal data can put people’s privacy at risk. Now firms could be exploiting this vast trove of personal information to charge people higher prices,” FTC Chair Lina Khan said in a statement. “Americans deserve to know whether businesses are using detailed consumer data to deploy surveillance pricing, and the FTC’s inquiry will shed light on this shadowy ecosystem of pricing middlemen.”

Companies have already experimented with implementing Uber-style surge pricing. JetBlue quietly implemented “peak” and “off-peak” pricing to its checked bag fees in March. Walmart plans on bringing digital price tags to 2,300 stores over the next several years that will let it change products’ prices depending on a number of factors, including the weather, whether items are closer to their expiration date, and more. Wendy’s said it would roll out “dynamic pricing” in 2025. Amazon had a secret algorithm called Project Nessie that let it gauge how much it could raise its prices before its competitors took notice and changed theirs, too.

These pricing models, while variable, are a bit different from surveillance pricing, which relies on using an individual consumer’s information to determine how much they’re willing to pay for a specific product.

The FTC orders were sent to Mastercard, Revionics, Bloomreach, Chase, Task, Pros, Accenture, and McKinsey & Co., the FTC announced in a press release. They were issued under the FTC’s 6(b) authority, which lets it conduct studies without any specific law enforcement purpose.

“These eight firms advertise their use of AI and machine learning as part of the services they offer, often claiming they can use historical and real-time customer information to drive unique purchasing experiences,” an FTC official said on a call with reporters ahead of the announcement, adding that none of the firms issued orders have been accused of any wrongdoing. “What we’re trying to do is lift the hood on opaque practices that have the potential to profoundly reshape how consumers purchase goods and services. We want to make sure that surveillance pricing isn’t resulting in harm to certain communities, like women and rural consumers.”

The companies are being asked to turn over information on:

Types of surveillance pricing products and services they’ve produced, developed, or licensed to a third party
Data sources used for each product or service
Who they’re offering surveillance pricing products and services to
How this impacts surveilled consumers and pricing

On a call with reporters, an FTC official said a study is necessary because the extent to which businesses use surveillance pricing is currently unknown. Media reports suggest that companies across a wide variety of sectors — including grocery stores, restaurant chains, travel, and hospitality — are considering implementing surveillance pricing or have already done so.

“The technology is there, the incentives are there, the data certainly is there to do this targeting,” the FTC official told reporters. “And now we actually see this industry of consultancies potentially offer to help companies personalize pricing, which gives us a strong reason to believe that this practice is likely to grow.”

Read More