Wired

For a while now, companies like OpenAI and Google have been touting advanced “reasoning” capabilities as the next big step in their latest artificial intelligence models. Now, though, a new study from six Apple engineers shows that the mathematical “reasoning” displayed by advanced large language models can be extremely brittle and unreliable in the face of seemingly trivial changes to common benchmark problems.

The fragility highlighted in these new results helps support previous research suggesting that LLMs’ use of probabilistic pattern matching is missing the formal understanding of underlying concepts needed for truly reliable mathematical reasoning capabilities. “Current LLMs are not capable of genuine logical reasoning,” the researchers hypothesize based on these results. “Instead, they attempt to replicate the reasoning steps observed in their training data.”

Drew Crecente’s daughter died in 2006, killed by an ex-boyfriend in Austin, Texas, when she was just 18. Her murder was highly publicized—so much so that Drew would still occasionally see Google alerts for her name, Jennifer Ann Crecente.

The alert Drew received a few weeks ago wasn’t the same as the others. It was for an AI chatbot, created in Jennifer’s image and likeness, on the buzzy, Google-backed platform Character.AI.

Slowly but surely,Apple TV+ has found its feet. The streaming service, which at launch we called “odd, angsty, and horny as hell,” has evolved into a diverse library of dramas, documentaries, and comedies. Now, its library is so packed, we’ve declared it “the new HBO.”

Curious but don’t know where to get started? Below are our picks for the best shows on the service. (Also, here are our picks for the best movies on Apple TV+.) When you’re done, head over to our guides to the best shows on Netflix, best movies on Hulu, and best movies on Amazon Prime, because you can never have too much television.

Humans have been trying to talk to animals ever since we figured out how to form words. In modern times, we turn to technology for the solution—giving our dogs talking buttons to paw at, or trying to use artificial intelligence to help us understand whales.

The latest and perhaps most direct approach at human-animal communication is a voice-activated collar that gives your pet the power to talk back to you. Or at least, that’s the idea.

If you buy something using links in our stories, we may earn a commission. This helps support our journalism. Learn more. Please also consider subscribing to WIRED

The seventh generation of Apple’s tiniest iPad is finally here, three years since its last refresh. There was no fancy pretaped or live event—the company quietly announced the new tablet through its online newsroom. The headline feature encapsulating this device? Apple Intelligence. Apple’s artificial intelligence features still aren’t available for the iPhone 16, but they’re due to arrive this month via iOS 18.1.

If you buy something using links in our stories, we may earn a commission. This helps support our journalism. Learn more. Please also consider subscribing to WIRED

It’s been a few years since the Xbox Series X and Series S launch, and while the console hasn’t broken sales records, there’s a respectable number in the wild. It’s about time for a refresh, and Microsoft is delivering new all-digital versions of its consoles (say goodbye to the optical drive). You can order themright now.

The European New Car Assessment Program (Euro NCAP) has not yet measured, prodded, weighed, or smashed a Tesla Cybertruck—or checked whether pedestrians could survive a hit from this angular beast—but Elon Musk’s electric pickup truck might potentially score poorly on the tests anyway.

“Based only on the car’s visual appearance, there are several aspects of this vehicle that look like they may be a threat to pedestrians,” claims Euro NCAP’s director of strategic development, Matthew Avery. “You cannot fail Euro NCAP,” he adds, “but you can get a bad score.”

How are you feeling today? Hopefully you’re good. But hey, remember how you were feeling two Tuesdays ago? That answer might feel a little hazier. If only there were some kind of tech that kept track of that sort of thing.

A company called Emteq Labs has unveiled its vision for a new type of smart glasses—ones that peer inward at you rather than outward at the world. The glasses, called Sense, have a series of sensors in them that the company says can monitor facial movements in real time to discern the emotional state of the wearer. Sensors at the top of the frame monitor eye and eyebrow movements, while sensors on the bottom rim can detect cheek and jaw movements. Together, the sensors can pick up the subtle movements that occur when people make expressions like smiling or frowning, or when they are chewing food.

The world’s first assisted suicide pod wraps around the human body like a space capsule, tilting gently toward the sky. The device is designed to look as if the person inside were embarking on a journey, says its inventor, the Australian right-to-die activist Philip Nitschke. “It gives you the idea you’re saying goodbye to the world.”

Last month, the 3D-printed pod was used for the first time. In a forest on the Swiss-German border, an unnamed 64-year-old American woman pressed the pod’s button to release deadly nitrogen gas. She died seven minutes later, estimated the Swiss assisted suicide group The Last Resort, whose president Florian Willet was present at her death and was later detained for “aiding and abetting” the woman’s suicide. Nitschke tried to watch by video link, although he describes the signal as patchy.

In early 2020, deepfake expert Henry Ajder uncovered one of the first Telegram bots built to “undress” photos of women using artificial intelligence. At the time, Ajder recalls, the bot had been used to generate more than 100,000 explicit photos—including those of children—and its development marked a “watershed” moment for the horrors deepfakes could create. Since then, deepfakes have become more prevalent, more damaging, and easier to produce.

Now, a WIRED review of Telegram communities involved with the explicit nonconsensual content has identified at least 50 bots that claim to create explicit photos or videos of people with only a couple of clicks. The bots vary in capabilities, with many suggesting they can “remove clothes” from photos while others claim to create images depicting people in various sexual acts.