PC memory prices remain seriously high, making even a straightforward 32GB DDR5 upgrade far more expensive than it would have been just a year ago.

Sadly, I don’t have a time machine, so there’s nothing I can do about that, but there are still decent discounts to be had. I’ve found one on a 32GB (2 x 16GB) kit of Corsair Vengeance DDR5-6000 RAM for $490 (was $605) at Newegg when you use code SSF73323.

It’s already got a small discount, but the code knocks even more off the list price.

Intel XMP 3.0 support also makes getting the advertised performance relatively straightforward on a compatible motherboard. Corsair uses a low-profile aluminum heat spreader, so the modules should also play nicely with larger CPU coolers.

There’s no RGB lighting, which could actually be a plus if you prefer a clean build without turning your PC into a miniature nightclub.

For more savings, we’ve picked out this week’s best RAM deals.

Should you buy it?

This is a decent deal for someone building or upgrading now, but if you’re not in any rush, it’s worth waiting for memory prices to come back down. You should also check your motherboard and CPU compatibility before ordering. This particular kit supports Intel XMP 3.0 rather than being marketed as an AMD EXPO kit.

from the but-kids-can’t-witness-violence! dept

Earlier this year, one of the smartest internet rights people around, Heather Burns, suggested the “Darnella Test” regarding any kind of “kid safety” rule online. It’s named after Darnella Frazier. You might not recognize her name, but you’ve seen her work. She was the 17-year-old who was walking to the corner shop when she happened to see Derek Chauvin kneeling on George Floyd’s neck, in the process of murdering him. She got out her phone and filmed it, enabling the world to see that the initial statement from the Minneapolis police — the one headlined “Man Dies After Medical Incident During Police Interaction” — was bullshit.

She won a Pulitzer Prize for the video.

But she was a teen. On social media. So Burns came up with the Darnella test, to explore whether any particular internet regulation would have prevented Darnella from documenting the murder of George Floyd:

For any young person who is just going to be walking along the street for a snack and ends up witnessing something that nobody should ever see because it should not be happening. For that young person whose only recourse, at that moment in time, is to document and report.

And as that future Darnella pulls out their phone to document the event:

1. Would they be allowed to have that phone, at all, under xyz regulation?
2. Would they be allowed to have that social media account, at all, under xyz regulation?
3. Would they be allowed to upload video, or livestream content, under xyz regulation?
4. Would the video, because of their age-verified account information, be instantly flagged and/or taken down for violent content?
5. Would the video, as urgent journalistic content in the public interest, be suppressed and censored based not on the content within it but on the age of the person who filmed it?

That’s your Darnella test. Run through it as if your life depends on it. Because it just might.

Thankfully, unlike the various outright bans of social media for teens, the newly announced Meta settlement with most state AGs does not actually ban kids from social media. But that doesn’t mean it passes the Darnella test. At the very least, the settlement will make it much harder for the next Darnella to document and distribute things that need to be documented and distributed.

Let’s start by running the settlement’s terms through the Darnella Test, question by question, to see what a future Darnella would actually be able to do. Especially since the goal of the settlement is that it become the industry norm across all of social media.

The first question doesn’t really apply here, since it has nothing to do with kids having phones directly, so that passes the Darnella test.

You could argue that the second question regarding whether or not the next Darnella would have a social media account also passes, since nothing in the settlement officially bars teens from having an account. But it certainly could impact them indirectly. The age assurance requirements make signing up more of a hassle, so the next Darnella might not bother creating an account at all. And if she does, that verification is precisely what marks her account as a teen account — which is what triggers everything that follows.

We can lump the next three questions together, because they’re all about the content itself. And here is where it seems clear that the settlement agreement fails the Darnella test. The “age-appropriate content restrictions” and the “content restrictions” for teen users would almost certainly create real problems for a teenager documenting a murder like George Floyd’s. A video of a police officer kneeling on a man’s neck until he stops moving is graphic violence by any classifier’s reckoning — and the entire point of a verified teen account is that graphic violence doesn’t stay on it.

And on that last item in the test — whether or not “urgent journalistic content in the public interest” would be suppressed — there are serious problems, even if the video somehow slipped past the content restrictions. The settlement imposes a default two-hour daily time limit on teen accounts — which means the answer might depend on how much time the next Darnella had already burned scrolling before she happened to walk past a murder in progress. Considering that George Floyd was murdered in the evening (around 8:30pm) there’s a good chance a teen user would have already used up their allotted time.

Yes, it’s possible that the next Darnella might have more time due to a parent or guardian bypassing the two hour restriction, but we don’t know that ahead of time. And a teenager watching a man die on the pavement in front of her does not have time to go find a parent and ask them to unlock the app.

Also, crime — and other things worth documenting — doesn’t happen only during your waking hours. The settlement includes “night mode” restrictions that say that teens will not be able to post content between midnight and 6am. If Darnella happened to witness Floyd’s murder past midnight, she might be out of luck. Or, at least, the police would have extra hours to lock in their false narrative.

And this is the problem with so much of the discourse regarding child safety online. It starts from the position that the internet is inherently unsafe for kids, and that the only remaining question is how thoroughly to block them from it.

But reality is more complicated than that. A kid with a phone and a social media account might also be the only thing standing between the official story and the actual truth. A teenager may be the one livestreaming a school shooting from a locked classroom. Or be a bystander filming ICE violating the rights of people across America. They might be the person documenting police violently attacking protestors.

Yet, under a framework built around “protecting teens from the dangerous internet,” such things will face multiple hurdles. A verified minor uploading graphic, unmoderated content in real time will violate all sorts of rules.

Run the actual Darnella video of Derek Chauvin kneeling on George Floyd against the “industry wide” standard Meta just paid $17 billion to create. It doesn’t pass. It’s graphic violence, filmed by a verified teen account late in the evening, perhaps after she’s used up all her allotted time. If it were after midnight she’d be barred entirely from posting. The next Darnella video might not exist. But at least 52 Attorneys General get to pretend they “protected kids.”

Filed Under: age verification, darnella frazier, george floyd, kid safety, social media

Companies: meta

If you’ve heard the high-pitched whine or buzz from an electronic device when a current-carrying inductor is vibrating, you’ll know how unpleasant it can be. It’s common in all kinds of equipment, but it’s become a particular annoyance of late in hardware like PC power supplies, GPUs, and cooling pumps. There is plenty of hearsay on the web about which parts whine and which don’t, and [Lowell Wood] wants to get to the bottom of it. 

To track the issue, [Lowell] has put together the Coil Whine Database. It accepts user reports on hardware, regarding the level at which coil whine is present. A score of 0 is given to a part that is inaudible in a quiet room, with higher scores representing higher sound levels. A part scored at 2 is audible working at a desk with the computer under load; a part scored at 4 is audible even when the machine or device is at idle.

For now, the database is largely empty—[Lowell] has just opened submissions, adding a report on their own ROG Astral RTX 5080 card for good measure. If you want to submit a report on a unit, either silent or noisy, that’s simple enough—just fill in the coil whine report form. Over time, submissions will hopefully grow, and it will be easier to get a good idea of what equipment whines and what is likely to operate silently out of the box.

The database aims to present a guide to what parts whine, and how much, noting that any given population of devices tends to vary. To that end, any given device won’t be reported on publicly until it receives at least 5 reports. To counter bias, reports of silent parts will be weighted higher, since it’s unlikely that people happy with their quiet hardware will be rushing to research this issue or report it to a database. Relevant files to the project are available on GitHub for the curious.

This database could be a great boon to the brigade of PC builders out there who like their machines to be as silent as possible. If that sort of thing appeals to you, it’s probably time you started researching passive cooling as well…

Most people have never held an Amprobe AT-7000-EUR, and even fewer would guess a cable tracer can cost about $2,700. Electricians use this kit when a cable disappears into drywall, a ceiling, a floor, or a crowded panel and a cheap tone probe is not enough. One half of the kit is a transmitter you clip onto a circuit. That box injects a tracing signal, 6.25 kHz on live wires or 32.768 kHz on dead ones, in high, low, or clamp mode. The other half is a receiver with a color screen, a tip sensor, and a Smart Sensor array that shows where the wire runs and which way it is going, often to about two inches, and can pick the correct breaker instead of a handful of false hits.

Tom bought the pair for work at roughly £2000 ($2700). After months of sitting in storage, the transmitter simply went dark for no apparent reason. The transmitter went dead, which means the batteries, buttons, and display did not even twitch. However, the receiver opted to reactivate, which is at least some aid in locating the problem. Fluke’s service team quoted £70 for an initial quote and another £600 to even try a repair. When the fees were factored in, it was off to the “YouTube Heroes for Hire” op Buy it Fix it.

Sale

DJI Mic Mini (2 TX + 1 RX + Charging Case), Ultralight, Detail-Rich Audio

• Small but Mighty – The DJI Mic Mini lavalier microphone transmitter is small and ultralight, weighing only 10 g, [1] making it comfortable to wear…
• Detail-Rich Sound – Mic Mini wireless microphones delivers high-quality audio. A 400m max transmission range [2] ensures stable recording, even in…
• Extended Battery, More Recording Time – Mic Mini wireless lavalier microphone with Charging Case offers up to 48 hours of battery life, [3] ideal for…

Five small screws later, the transmitter board was on the bench. A 10.2 volt bench supply showed the normal voltages popping up: 5V, 3.3V, and 1.2V, all present and accurate. A thermal camera scanned for hotspots but found nothing unusual. An oscilloscope showed a clear 23.9989 MHz clock signal from the 24 MHz crystal as well. Even the backlight seemed to work well after the ribbon was reseated. So it appeared that the hardware issue was not with the power supply.

Only after connecting the serial pins to a computer at a baud rate of 115200 did the true tale become obvious. Boot logs appeared before abruptly terminating with a NAND boot failure. Both machines run some ancient Linux 2.6 firmware from a 128 MB NAND chip housed within a TSOP-48 box. At startup, the system on a chip reads that chip, checks the boot image, and refuses to continue if it discovers a problem. This is exactly what the dead transmitter was doing.

After blowing hot air into the NAND chip, a programmer read it. We could see the data was there, but a hex view looked like a nightmare, with a large empty patch and random bit flips. In the erroneous areas, E1 was where E should be, O where 01 should be, and C9 where C1 should be, among other things. A wild goose hunt through Google’s Gemini chatbot took up half of the day in theory on how ECC works, and I didn’t get anywhere or get the unit up and running, but the smart play was simpler.

The receiver was also ripped apart, and it had the same CPU, RAM, and, most critically, flash IC. So some wires were attached to the serial header, and a fast run of U-Boot with the programmer’s bit-flip settings adjusted slightly resulted in a clean NAND image. The next thing he knew, he had the files side by side, and it was obvious what had gone wrong: only a few cells in the transmitter’s bootloader had faded out after years of inactivity. Using a hex editor, clone the sound healthy bootloader bytes from the receiver into the rotten transmitter bootloader, creating a hybrid image that was then written back into the original chip.

That got him all the way there, and the resoldering should have done the work, but one weak solder joint on the flash pin meant it wouldn’t boot up. After a brief fix up, the transmitter rebooted into its normal Linux 2.6 interface. So a live test was performed, and when the receiver approached the transmitter on the cable, the signal climbed up as it should. The kit was back up and running for daily operations. Those leaky NAND cells are still in there, so a fresh flash IC is the lasting insurance policy. A professional tracer that official service priced like a small appliance came back because two identical boards shared a bootloader, and a handful of forgotten bits were put back where they belonged.
[Source]

NVIDIA is getting ready to unleash DLSS 5, its latest attempt to push real-time game graphics closer to photorealism. The company says the new 3D-guided neural rendering technology will be available from September 3 at 9 p.m. PT, with NBA 2K27 becoming its first game to support it across RTX 50-series GPUs, laptops and GeForce NOW. 

I got a look at DLSS 5 running in NBA 2K27 and the Zorah demo at Gamescom, and I’ll admit: it is considerably more impressive in motion than the initial announcement might suggest.

DLSS 5 isn’t taking the artist out of the equation

When NVIDIA first showed DLSS 5, one of the immediate concerns was that AI-generated visuals could end up overriding the artistic choices made by game developers. Having seen the technology in action, that’s not really what is happening here.

DLSS 5 analyzes the rendered frame along with information such as color and motion vectors, using its neural model to understand characters, objects, materials, lighting and their relationships within a scene. More importantly, developers remain in control of how much of that processing actually happens. 

That control goes surprisingly deep. Developers can use automatic or manual masking and adjust the intensity of structural details such as contact shadows, reflections and ambient occlusion, as well as broader lighting and color response. They can also apply those controls globally or to individual objects. In other words, DLSS 5 isn’t simply telling a game, “Here, let me make this prettier.” Artists can decide what gets the AI treatment and how aggressive it should be. 

And this is where the technology gets particularly interesting. In the demos I saw, DLSS 5 could distinguish different elements within a scene and treat them independently. A character, their clothing, surrounding objects and the background can effectively become separate pieces that developers can control. The result is a much more convincing interaction between lighting, materials and surfaces.

NVIDIA also emphasizes that the better the input, the better the output. That makes sense: DLSS 5 isn’t conjuring an entire game out of thin air. It is using the information already produced by the game to enhance the final image. 

From two RTX 5090s to an RTX 5060 laptop

Perhaps the biggest surprise for me wasn’t even the visual quality. It was the hardware running it. When DLSS 5 was initially shown, the technology required two RTX 5090 GPUs. At Gamescom, however, I saw the technology running on a laptop equipped with an RTX 5060. That’s a pretty wild reduction in hardware requirements for something that is still producing remarkably convincing results.

There is a performance trade-off depending on the quality settings, but the visual improvement is hard to ignore. NVIDIA’s own NBA 2K27 figures show DLSS 5 running alongside ray tracing and DLSS Multi Frame Generation, with an RTX 5060 reaching 258 fps at 1080p High settings, while an RTX 5090 reaches 370 fps at 4K Ultra using Performance mode.  

And that is probably the part I’m most excited to test properly. DLSS 5 isn’t just another checkbox in NVIDIA’s increasingly enormous graphics-settings menu. From what I’ve seen, it can fundamentally change how lighting, materials and characters appear on screen while still giving developers considerable control over the final image.

NBA 2K27 gets first dibs on September 3. Based on what I saw at Gamescom, I’m very curious to see whether DLSS 5 can consistently deliver that same “wait, that’s actually rendered in real time?” feeling once it leaves the carefully prepared demo environment.

Skynet has missed another appointment. Depending on which Terminator timeline you prefer, the murderous artificial intelligence should have become self-aware in 1997, 2004, 2011, 2017, or perhaps last Tuesday. Each date has passed without a nuclear exchange. The machines have mostly occupied themselves by flattering users, mangling fingers in photos, and confidently inventing facts for gullible people to believe.

That has not stopped Silicon Valley from treating artificial general intelligence as both an imminent miracle and an approaching monster. AGI supposedly marks the moment a machine can match human intelligence across a broad range of tasks, though nobody can agree on which tasks, which humans, or what counts as intelligence.

Facepalm: Graphics card scams on Amazon are far from new, but one customer has encountered what appears to be a new method of trying to trick victims – one that some people could fall for. It involves sending a cheap “present” to the buyer and a promise that the GPU is on its way. The scammer also sends a pamphlet that supposedly explains the process and includes a fake Amazon email to cancel the order.

Reddit user BransonArsen wrote that they bought an MSI Gaming RTX 5080 16G SUPRIM Liquid SOC edition from Amazon for nearly $1,400. They later explained that the seller was a third party with a Chinese name, which even they admit was a warning sign. The lower-case “msi” in the product description is a bit of a giveaway, too.

Those suspicions proved warranted when a delivery arrived at BransonArsen’s door a few days later. While we’ve seen many GPU scams involving something other than the graphics card being sent in the original packaging, including putty, rice, and older cards in diguise, it wasn’t even an RTX 5080 box that turned up.

This won’t get many fps

What arrived at the door was a plain Amazon box, inside of which was a cheap but correctly packaged vacuum mug.

Rather than just sending something totally different, this was all part of the scam. The box also included a pamphlet claiming the mug was a “complimentary gift.” The graphics card would be arriving separately, in about seven working days, because Amazon “collects orders and places a bulk order with this seller every Monday,” making the delivery longer but helping keep the price lower – this particular card is usually $2,000.

The leaflet, which appears to contain AI-generated images, adds that if the buyer wants a tracking number or has any questions, they can email the definitely-not-fake [email protected] customer service address. It also advises anyone wishing to cancel their order and receive a full refund to contact this address – but don’t worry, they can keep the mug.

Unfortunately, that mug might end up costing $1,400. It seems the idea here is for the victim to repeatedly email the fake address while the scammers keep stringing them along until the return window expires, leaving them with a very expensive piece of crap as a reminder to be wary of suspicious Amazon sellers.

Meta removed dozens of advertisements for apps that presented themselves as pornography but actually functioned as banking malware, after the Indian government issued an advisory.

Reuters then found at least 39 more of the ads still running and reported on Monday that Meta removed those as well after being asked about them.

The sequence is what makes the story notable. A government warning led to some of the advertisements being removed, while a reporter’s email prompted Meta to take down the rest.

Advertising makes this different from ordinary malware distribution. An app listed in a store still has to be discovered and installed by a user, whereas a paid advertisement is actively delivered to people selected by an advertising system because they are considered likely to engage with it.

The apps themselves were capable of accessing information on users’ phones, capturing one-time passwords and banking PINs, and transferring money from accounts without the owner’s knowledge.

Pornography is also an effective lure for this type of malware for reasons that have little to do with the technology itself. Someone who installs an app they would rather keep private may be less likely to report it immediately, particularly if they are embarrassed about how they encountered it.

One-time passwords create another vulnerability because they are now part of the security process used by many banking services.

Malware that can read those codes directly from a device can undermine the second factor that is supposed to protect an account if a password has already been compromised.

Meta did not respond to Reuters’ questions. The company removed the flagged advertisements without commenting on the findings, following a pattern that has become familiar in cases involving problematic advertising on its platforms.

India has good reason to be concerned about the issue. The country recorded close to $2.4 billion in losses from cyber fraud during 2025, while mobile banking has become the main way many people access financial services, particularly among newer account holders.

India is also Meta’s largest market by users, making gaps in its advertising enforcement more significant than they might be elsewhere. Hundreds of millions of accounts are potentially exposed to whatever the platform’s advertising review systems manage to identify and whatever they miss.

The advertising system is what turns this into a platform problem rather than simply a criminal one.

These apps were not being distributed through obscure forums or private networks; they were being promoted through a paid advertising system that Meta operates, moderates, and ultimately makes money from.

Internal projections reported last year also put the issue in financial terms. Meta estimated that scam and banned-goods advertising could account for roughly 10% of its 2024 revenue, or about $16 billion.

That figure changes the context around every subsequent enforcement announcement. Removing an advertisement after someone flags it is part of the cost of operating the platform, while the advertisement itself generated revenue for the same company responsible for deciding whether it should be there.

The regulatory response is now moving in a similar direction across several major markets, although governments are using different tools. India has relied on advisories and direct requests to the company, European authorities are pursuing fines, and US plaintiffs have been taking their claims to court.

India has been escalating its scrutiny for months. It has summoned Meta over Instagram advertising promoting child sexual abuse material and separately ordered the company to remove those advertisements.

European regulators have been raising similar concerns through different mechanisms. Poland has asked the European Commission to fine Meta €250 million over scam advertising, while banks in the UK have said that a large majority of the payment fraud they encounter originates on Meta’s platforms.

Meta has been developing countermeasures at the same time, including new scam-detection systems across WhatsApp, Messenger and Facebook.

The same company operates both the advertising auction and the systems intended to detect abusive content within it, which makes repeated failures harder to separate from the incentives built into the platform.

The common issue across these cases is less about whether Meta intends to distribute scams and more about how reliably it detects them. Nobody is suggesting that Meta wants banking malware in its advertising system.

The criticism is that the system continues to accept those advertisements, while government agencies, banks, and journalists sometimes identify them before Meta does.

Scale is a legitimate part of the explanation. Millions of advertisements enter Meta’s systems every day, and no moderation process can be expected to identify every fraudulent campaign immediately.

That does not, however, explain why the additional 39 advertisements remained active after the company had already received a government warning about the same type of threat.

The 39 advertisements Reuters found were still running after the government advisory had been issued.

They were removed after a journalist contacted Meta about them, leaving a fairly straightforward question about the effectiveness of the platform’s own detection systems: how much harmful advertising is being caught internally, and how much is being discovered by everyone else?

The Dyson CameraJet has what the company calls a “100k pixel macro lens camera” mounted right under the head of the brush. That’s 0.1 megapixels in more typical camera terminology, which may seem like a very small number, but it’s certainly supplemented by some algorithmic effort. More specifically, 16 million lines of code and “a machine learning system trained on 470,000 dental images collected during development” are behind the computer vision targeting here.

Basically, all that computing has taught the CameraJet how to identify the gaps between your teeth. When you brush and the system sees a gap, it will shoot a jet of fluid in a conical shape that helps to floss while you brush. To prevent liquid from accumulating in your mouth as you move along, these bursts of mouthrinse are fired in tiny amounts. The capacity of the water tank that surrounds the neck of the brush is just 12.5ml (0.42 fluid ounces), which is just about two and a half teaspoons.

According to Dyson, the CameraJet’s “Gap Optical Targeting” system analyzes “28 live images per second to identify and target gaps between teeth in real time” and “within 100ms of the camera seeing a gap, the device detects and jets the gap.”

That camera isn’t just there to tell the toothbrush where to shoot fluid. You can use the companion MyDyson app to get a better view of your mouth, especially the back of your teeth. The live-viewing feature is only available during live viewing or auto-jetting, and Dyson says no images are recorded or stored on the product or in the cloud. You can also use the app to get tips, guides and personalized feedback on how to brush better, with details on where you’re already good and what areas you might be falling back on. The app also offers settings so you can tweak brushing intensity and various jetting modes. 

In addition to the camera, the company said it uses an “anti-gravity tank with a diaphragm pump and pressure chamber that adapt during use to deliver controlled jets in any orientation.” That means no matter how you hold your toothbrush, the CameraJet should be able to pump water to your mouth, even if you’re somehow upside down and pointing awkwardly at the back of your teeth.

Dyson also stresses the importance of the shape of its water stream. Instead of a simple needle-shaped jet like many other water flossers, Dyson said its “fluid dynamicists spent years optimizing the profile of the jet to remove plaque more effectively.” As a Singaporean, I feel inclined to mention that the company developed “its own proxy plaque through a five-year collaboration with the National University of Singapore’s Faculty of Dentistry” as part of its engineering process. 

The result of all that work is a cone-shaped spray of fluid that Dyson says offers wider coverage to sweep and remove plaque while being propelled “at a lower pressure to be gentler on gums and enamel.”

As for what fluids to fill the tank with, you can choose your favorite clear mouth rinse that’s SLS-free to prevent foaming. Dyson is also making its own non-foaming toothpaste and a low-foaming mouthrinse to be used with the CameraJet, so the camera can see clearly without foam blocking its view.

Apple is expected to launch the iPhone 18 Pro and iPhone 18 Pro Max on September 9. Both models are likely to share several flagship features and upgrades. However, recent leaks suggest that Apple could reserve some hardware changes for the Pro Max. The camera may be the biggest difference between the two models, with the larger phone tipped to get new features. Apple has not officially confirmed these details yet.

iPhone 18 Pro Max Could Get a Variable Aperture Camera

The camera could be the biggest difference between the iPhone 18 Pro Max and iPhone 18 Pro. According to MacRumors, the Pro Max will introduce a variable-aperture camera to the iPhone. Previous Pro models have used fixed apertures on their primary cameras. The new system would be able to adjust the lens aperture to determine how much light enters. A larger aperture would help when taking photos in low-light situations and allow more light to enter the phone’s camera. A smaller aperture would ensure good depth control and background blur.

The Pro Max could get another camera advantage with up to 6x optical zoom. Apple may pair this feature with a more advanced telephoto camera. Optical zoom lets users capture distant subjects with less digital cropping. This can help maintain better detail when shooting from a distance. It could also make the Pro Max more useful for portraits. The iPhone 18 Pro is expected to keep its current telephoto setup.

Bigger Battery May Make the Pro Max Thicker

The iPhone 18 Pro Max may offer a larger battery than the standard iPhone 18 Pro. However, the larger battery could make the phone slightly thicker. The regular iPhone 18 Pro model is expected to retain its sleek design. As a result, it may be a good choice for users who prefer one-handed operation of the phone.

Despite their differences, both phones will share some features. For instance, both phones would use Apple’s new A20 Pro chip. They may also feature a smaller Dynamic Island compared with current models. Apple could improve the Camera Control feature on both phones. Both models are also expected to support Apple Intelligence features. Premium displays and titanium frames could remain part of the Pro lineup.