Computer manufacturing giant Dell is looking into claims that its infrastructure was breached and sensitive data on thousands of employees stolen.

Late last week, a threat actor with the alias ‘grep’ posted a new thread on the infamous dark web forum BreachForums. In it, they offered a large Dell database for sale, allegedly containing sensitive employee information.

The algorithms that underpin artificial intelligence systems like ChatGPT can’t learn as they go along, forcing tech companies to spend billions of dollars to train new models from scratch. While this has been a concern in the industry for some time, a new study suggests there is an inherent problem with the way models are designed – but there may be a way to solve it.

Most AIs today are so-called neural networks inspired by how brains work, with processing units known as artificial neurons. They typically go through distinct phases in their development. First, the AI is trained, which sees its artificial neurons fine-tuned by an algorithm to better reflect a given dataset. Then, the AI can be used to respond to new data, such as text inputs like those put into ChatGPT. However, once the model’s neurons have been set in the training phase, they can’t update and learn from new data.

This means that most large AI models must be retrained if new data becomes available, which can be prohibitively expensive, especially when those new datasets consist of large portions of the entire internet.

Researchers have wondered whether these models can incorporate new knowledge after the initial training, which would reduce costs, but it has been unclear whether they are capable of it.

Now, Shibhansh Dohare at the University of Alberta in Canada and his colleagues have tested whether the most common AI models can be adapted to continually learn. The team found that they quickly lose the ability to learn anything new, with vast numbers of artificial neurons getting stuck on a value of zero after they are exposed to new data.

“If you think of it like your brain, then it’ll be like 90 per cent of the neurons are dead,” says Dohare. “There’s just not enough left for you to learn.”

Dohare and his team first trained AI systems from the ImageNet database, which consists of 14 million labelled images of simple objects like houses or cats. But rather than train the AI once and then test it by trying to distinguish between two images multiple times, as is standard, they retrained the model after each pair of images.

They tested a range of different learning algorithms in this way and found that after a couple of thousand retraining cycles, the networks appeared unable to learn and performed poorly, with many neurons appearing “dead”, or with a value of zero.

The team also trained AIs to simulate an ant learning to walk through reinforcement learning, a common method where an AI is taught what success looks like and figures out the rules using trial and error. When they tried to adapt this technique to enable continual learning by retraining the algorithm after walking on different surfaces, they found that it also leads to a significant inability to learn.

This problem seems inherent to the way these systems learn, says Dohare, but there is a possible way around it. The researchers developed an algorithm that randomly turns some neurons on after each training round, and it appeared to reduce the poor performance. “If a [neuron] has died, then we just revive it,” says Dohare. “Now it’s able to learn again.”

The algorithm looks promising, but it will need to be tested for much larger systems before we can be sure that it will help, says Mark van der Wilk at the University of Oxford.

“A solution to continual learning is literally a billion dollar question,” he says. “A real, comprehensive solution that would allow you to continuously update a model would reduce the cost of training these models significantly.”

Intel’s list of best processors is about to expand with the upcoming launch of Arrow Lake-S — but what comes next is less certain, and even more so now. According to a post on the Chiphell forums, Intel may have decided to cancel the Arrow Lake-S refresh that was reportedly slated for sometime next year.  On the upside, the code name for an upcoming desktop CPU generation was also leaked.

The rumored Arrow Lake-S refresh was never confirmed, but there have been many whispers about it from various leakers. Much like the Raptor Lake refresh, it was never meant to be a groundbreaking update; the neural processing unit (NPU) was the main thing that was going to be updated.

With no core configuration or clock speed adjustments, Intel’s choice becomes clearer. The NPU doesn’t play as big a role in a desktop PC as it does in a thin-and-light laptop, as a discrete graphics card can handle AI workloads much better in most scenarios. If the NPU was the only thing to be changed, we likely wouldn’t have seen any performance improvements in things like gaming and many other workloads.

By the sound of it, many of us would do just fine without an Arrow Lake-S refresh. However, this does mean that the LGA1851 socket might end up being used in just one generation of processors unless Intel sticks to it with Nova Lake. Nova Lake-S was initially rumored to use a new socket, and it’s also not expected to hit the market until 2026 and beyond. This would leave Intel with an up to two-year gap between CPU generations.

There’s some hope for that Arrow Lake-S refresh yet, though. For starters, the user who originally said that it’s been canceled and that ARL-S would last for two years edited their post. They’re still implying that a refresh isn’t happening, but it’s now less of a definite statement — and it wasn’t confirmed to begin with, so take it with some skepticism. Moreover, even if the desktop refresh is never released, Intel might still refresh Arrow Lake-H and HX, meaning the laptop variants.

Razer Lake

— HXL (@9550pro) September 23, 2024

Although Intel’s roadmap may now be looking pretty bare as far as desktops are concerned, we might have just learned the name of what comes after Nova Lake. According to frequent hardware leaker HXL on X (formerly Twitter), the successor to Nova Lake-S will be called Razer Lake. Another source, Bionic_Squash, agreed with this leak.

As things stand now, we’re still waiting for an official announcement regarding Arrow Lake-S. Rumor has it that Intel might unveil its new desktop CPUs on October 10, followed by an October 24 release date.

The iPhone 16 series, especially the vanilla and “Plus” models, are the most repairable iPhones in years. Probably the EU’s pressures to enhance repairability in tech devices have a lot to do with this. One of the biggest internal changes is a new adhesive that makes battery replacement much easier. A new teardown from iFixit reveals just how simple it is to remove the battery on an iPhone 16.

The new iPhone 16’s battery glue releases the cell when you “electrocute” it

If you’re not aware, the iPhone 16 and iPhone 16 Plus use a new, neat method of battery removal. While glue is still there, it is now responsive to low-voltage electrical currents thanks to new special properties. Basically, the adhesive “releases” the cell when you infuse it with a small current of around 9V. You can do this with a simple battery that is available in many places and some alligator clips (or another conductive tool).

The teardown shows that the change is pretty impactful, especially considering the risks of removing the battery in previous models. Until now, all iPhones used traditional glue and the classic removal tabs for the battery. However, if for some reason you were unable to remove the cell by pulling the tabs, you had to resort to pressure tools. This entailed a potential risk of damage to the device or even fires.

For unknown reasons, the new “voltage-sensitive” glue is not present in the iPhone 16 Pro and Pro Max. These models keep the battery removal system from previous generations. However, they also got some tweaks in favor of repairability. There is now a new metal enclosure for the battery, and the LiDAR scanner and TrueDepth camera are more easily accessible to technicians. Apple will likely implement the new battery glue in all future iPhone models.

More internal changes across the series

The iFixit teardown shows more than just the iPhone 16’s battery glue. It also provides a closer look at the new Camera Control button that’s present across the lineup. You can check out the full teardown video below.

If events in the last few years have felt like a higher power playing Mad Libs with our lives, then it looks as if it’s running out of options. “SpaceX,” you imagine it pulling out “gets sued by…” and then the sounds of paper rustling until it says, “Cards Against Humanity.” Turns out the silly game jokesters own an acre of land near to SpaceX’s facility in Texas, which the latter has been using for its own purposes.

. The lawsuit says the previously pristine land has been turned into an ersatz staging ground and parking lot for nearby construction work. It asks for $15 million compensation for the trespass and damage. If successful, it says it’ll share the win with the crowdfunding backers who coughed up to buy the land.

If the higher power in question wants some help with some suggestions, it could do worse than suggesting Engadget’s humble newsletter writers win a billion pounds each on the same day in completely different competitions.

— Dan Cooper

The biggest tech stories you missed

Don’t worry, it’s not using that power for anything good!

Microsoft is . Those with short memories may not know it was the site of a 1979 meltdown that helped halt the development of nuclear energy in the US.

Now, I’m not mad the plant is being reactivated, since nuclear offers safe, clean and abundant energy. TMI reactor one remained in operation from 1985 until 2019, so it’s got a healthy track record too. But I am mad at the reason Microsoft’s cutting the check: to power its AI data centers!

Anyone for pickleball?

Here’s a non-exhaustive list of reasons I rarely visit movie theaters:

1. The cost

2. Not being able to pause the movie to visit the restroom

3. People talking during the movie

4. People checking Facebook on their phone during the movie

5. People making phone calls during the movie

Here’s a non-exhaustive list of things movie chains will do to lure me back:

.

OpenAI is reportedly ditching its increasingly recognizable hexagonal flower logo in favor of something a bit more sinister. . At the same time, a new profile of Sir Jonathan Ive reveals the long-rumored partnership between him and OpenAI CEO Sam Altman is real. . Let’s hope it does better than the Humane Pin, the last AI hardware tool cooked up by a storied designer with Apple on their resume.

The Food and Drug Administration has approved a nasally-administered influenza immunization treatment that can be taken at home. FluMist — a nasal spray vaccine that AstraZeneca initially developed as an in-office treatment two decades ago — will still require a prescription to obtain and is expected to be made available via a new online pharmacy next year, according to The New York Times.

The treatment will require people to fill out a questionnaire on the upcoming FlueMist Home website. Once approved by a pharmacist, the nasal spray will be shipped directly to the customer’s door. The current out-of-pocket cost is around $35 to $45 per dose according to the NYT, but that may drop depending on insurance coverage.

The Centers for Disease Control and Prevention found that FluMist has a similar efficacy to the traditional shot. It has been available through healthcare providers ever since it was first approved by the FDA back in 2003. It’s suitable for people aged between 2 to 49 years old — though the FDA recommends that the nasal spray be administered by a parent/caregiver to anyone under 18. The spray could be an alternative for people who are averse to getting injections, or otherwise find it difficult or inconvenient to travel for flu immunization treatments.

“Today’s approval of the first influenza vaccine for self- or caregiver-administration provides a new option for receiving a safe and effective seasonal influenza vaccine potentially with greater convenience, flexibility and accessibility for individuals and families,” said the FDA’s vaccine center director, Dr. Peter Marks. The World Health Organization reports that there are around a billion annual cases of seasonal influenza, resulting in 290,000 to 650,000 respiratory deaths each year.