Tesla has invited a whole bunch of lawsuits and investigations over the safety of its self-driving tech, and it continues to flub it despite all the progress. But people, apparently, want self-driving in their cars so bad they can install a DIY mod, too. The federal government is now scrutinizing Comma AI after a string of serious crashes tied to its open-source Openpilot system raised urgent questions about who is responsible when DIY self-driving software goes wrong. AI — yeah, I know — just brought another long-dead screen legend to life. And then there’s Tinder, which now wants you to find love by hopping within groups.

Send us feedback at [email protected].

The National Highway Traffic Safety Administration has opened a formal investigation into Comma AI, the company behind the open-source assisted driving platform Openpilot, following five crashes involving vehicles running the software. Across those incidents, cars equipped with Comma hardware slammed into slow-moving or stopped vehicles, leaving eleven people injured and three dead. The investigation covers roughly 30,000 Comma hardware kits, including the latest Comma Four device, which retails for around $1,000 and is marketed as delivering hands-free driving for cars people already own, a claim that sits in tension with fine print requiring constant driver supervision.

The probe gained public attention after a video posted to X showed a Tesla Model Y navigating roads using Comma Four hardware running Sunnypilot, a community-built fork of Openpilot. The clip drew both admiration and alarm, with critics pointing out that aftermarket self-driving software is being installed on public roads with none of the institutional accountability that comes with commercially developed systems. The concern deepens when accounting for the modding culture around Openpilot, where users run customized versions of the software whose safety implications are difficult to trace, and where the rise of AI-assisted coding tools makes it increasingly easy for inexperienced developers to alter how a car behaves at speed.

Greta Garbo, who defined Hollywood glamour across the silent era and the early talkies, may be heading for a third act. The Swedish-American star's descendants, who manage her name, voice, image, and publicity rights, have said they are open to an AI version of Garbo taking on genuine film roles after being struck by the quality of a recent digital recreation. That resurrection came in the form of a 99-second advertisement for Swedish ball-bearings manufacturer SKF, a company with a genuine historical link to Garbo, whose second-ever screen appearance was in a 1921 promotional film connected to the Gothenburg firm. Craig Reisfield, a great-nephew who once drove for Garbo and saw her regularly during her years in New York, described himself as "flabbergasted" by the result. "The mannerisms that they could capture and manipulate, her dress, facial images and even her voice," he said. "It was startling what they could do."

The technical approach was unusually constrained. Because still and moving images of Garbo are protected by copyright, the creators did not feed any of them directly into the AI systems, which included China's Seedream and Google's Nano Banana Pro. Instead, they worked entirely from text prompts describing her features, among them her ice-blue eyes and wavy blond bob, and overlaid the AI-generated appearance onto a human performer. Her voice was reconstructed from a recording made during her film career. Reisfield said the outcome cleared the so-called uncanny valley that typically plagues AI likenesses, and it was enough to make him reconsider what might be possible. "I won't say no, especially having seen what I've seen," he told the Guardian when asked whether an AI Garbo could be cast in other roles. Whether a full film role follows will depend on that same careful judgment, but for now, at least, the door is open.

Tinder is rolling out a new feature called Group Hangouts that lets friend groups form a shared profile and match with other groups directly inside the app. The way it works is straightforward: a user creates a group, selects a few shared activities or interests, and invites friends to join. Once a group reaches at least three members, its profile becomes visible on the main Tinder feed, where other groups can discover it, drop a like, and, if the interest is mutual, open up a group chat together.

The feature grew out of the success of Tinder's Double Date option launched last year, and early testing in Australia suggests real momentum behind the concept. Women were 2.2 times more likely to like a group profile than a traditional one-on-one profile, and the acceptance rate for Group Hangouts invites came in close to 40 percent. Group recommendations, like individual ones, are shaped by a user's stated interests and preferences, and standard safety tools, blocking and reporting, carry over to the group setting as well. Group Hangouts is already live for users in the United States, with a broader global rollout planned for the coming months.

Xbox Disc to Digital converts physical games to digital

Apple patched an iPhone flaw already under attack

Honor's Life S3 snaps magnetically onto your phone's back

Apple Pay launches in India with limited card support

Zelda's 40th anniversary concert tickets go on sale Sept.

GTA 6's Game Informer reveal crashed the app

Hulk returns in Avengers: Doomsday, Russos confirm

Avatar: Seven Havens episode titles have been revealed