Category: Technology

Imagine going to the doctor, telling them exactly how you’re feeling and then a transcription later adds false information and alters your story. That could be the case in medical centers that use Whisper, OpenAI’s transcription tool. Over a dozen developers, software engineers and academic researchers have found evidence that Whisper creates hallucinations — invented text — that includes made up medications, racial commentary and violent remarks, ABC News reports. Yet, in the last month, open-source AI platform HuggingFace saw 4.2 million downloads of Whisper’s latest version. The tool is also built into Oracle and Microsoft’s cloud computing platforms, along with some versions of ChatGPT.

The harmful evidence is quite extensive, with experts finding significant faults with Whisper across the board. Take a University of Michigan researcher who found invented text in eight out of ten audio transcriptions of public meetings. In another study, computer scientists found 187 hallucinations while analyzing over 13,000 audio recordings. The trend continues: A machine learning engineer found them in about half of 100 hours-plus worth of transcriptions, while a developer spotted hallucinations in almost all of the 26,000 transcriptions he had Whisper create.  

The potential danger becomes even clearer when looking at specific examples of these hallucinations. Two professors, Allison Koenecke and Mona Sloane of Cornell University and the University of Virginia, respectively, looked at clips from a research repository called TalkBank. The pair found that nearly 40 percent of the hallucinations had the potential to be misinterpreted or misrepresented. In one case, Whisper invented that three people discussed were Black. In another, Whisper changed “He, the boy, was going to, I’m not sure exactly, take the umbrella.” to “He took a big piece of a cross, a teeny, small piece … I’m sure he didn’t have a terror knife so he killed a number of people.”

Whisper’s hallucinations also have risky medical implications. A company called Nabla utilizes Whisper for its medical transcription tool, used by over 30,000 clinicians and 40 health systems — so far transcribing an estimated seven million visits. Though the company is aware of the issue and claims to be addressing it, there is currently no way to check the validity of the transcripts. The tool erases all audio for “data safety reasons,” according to Nabla’s chief technology officer Martin Raison. The company also claims that providers must quickly edit and approve the transcriptions (with all the extra time doctors have?), but that this system may change. Meanwhile, no one else can confirm the transcriptions are accurate because of privacy laws. 

This article originally appeared on Engadget at https://www.engadget.com/ai/openais-whisper-invents-parts-of-transcriptions–a-lot-120039028.html?src=rss 

In an AMA this weekend, Instagram head Adam Mosseri shared some insight into why some videos on the platform appear reduced in quality well after they’re posted, and it all boils down to performance. Responding to a question about old stories looking “blurry” in highlights, Mosseri said, “In general, we want to show the highest-quality video we can. But if something isn’t watched for a long time — because the vast majority of views are in the beginning — we will move to a lower quality video.” If the video later spikes in popularity again, “then we will re-render the higher quality video,” he said in the response, which was reposted by a Threads user (spotted by The Verge).

Further elaborating in a follow-up reply, though, Mosseri added, “We bias to higher quality (more CPU intensive encoding and more expensive storage for bigger files) for creators who drive more views.” The comment has sparked concern from small creators in the replies who say it puts them at a disadvantage competing with others who have larger platforms. Meta has previously said it uses “different encoding configurations to process videos based on their popularity” as part of how it manages its computing resources. 

The performance system “works at an aggregate level,” Mosseri said, “not an individual viewer level… It’s not a binary theshhold [sic], but rather a sliding scale.” In response to one user who questioned its fairness for smaller creators, Mosseri said the quality shift “doesn’t seem to matter much” in practice as it “isn’t huge” and viewers appear to care more about video content over quality. “Quality seems to be much more important to the original creator, who is more likely to delete the video if it looks poor, than to their viewers,” he said. Understandably, not everyone seems convinced.

This article originally appeared on Engadget at https://www.engadget.com/social-media/mosseri-confirms-instagram-reduces-video-quality-for-posts-that-arent-raking-in-views-233033536.html?src=rss 

Apple’s rumored smart home display may look a lot like an old friend — the iMac G4 of the early 2000s. In the Power On newsletter this weekend, Mark Gurman reports that the countertop tablet will feature a square screen that’s “positioned at an angle on a small base, making it reminiscent of the circular bottom of the iMac G4.” That may be where the speakers are housed too. The device is expected to arrive sometime next year, followed later on by a higher-end version with a robotic limb that can change the position of the display.

The more affordable model, which will be released first, will have a relatively small screen, according to Gurman, being “about the size of two iPhones side by side.” It’s intended to be used for smart home controls, but will also run apps like FaceTime and Calendar, and display photos and videos. The more expensive model will boast a larger screen in addition to its robotic capabilities, according to Gurman. We likely won’t see that product until at least 2026, as he previously predicted, and it may run you almost $1,000. Both models are expected to come with Apple Intelligence.

This article originally appeared on Engadget at https://www.engadget.com/home/smart-home/apple-may-tap-into-a-beloved-retro-design-for-its-smart-home-display-210412601.html?src=rss 

The long-promised Playdate Stereo Dock is officially dead, at least for now. After many months with the desktop accessory stuck in development limbo, Panic has updated the product page so it no longer says it’s “coming soon,” but instead that it’s been put on pause indefinitely due to issues with the Bluetooth experience, the factory that handled it and the rising costs of resolving it all. The Stereo Dock, first announced in 2021, was meant to be a charging dock that doubles as a Bluetooth speaker (and pen holder), along with being a cute way to display your Playdate when it’s not in use. But, things didn’t quite work out as planned.

The Playdate team said it hit some unexpected hurdles in trying to bring the idea to life, and found that “making the Stereo Dock was almost more challenging than making the Playdate itself.” The statement explains:

What happened? Well, our first regret was assuming that our factory at that time — a factory that made lots of Bluetooth speakers — could easily design the electronics for us, saving us lots of time and money. That went pretty well at first. But they, in turn, outsourced the Bluetooth software and chipset to another company in another country.

As a result, the Stereo Dock would crash often — rapidly adjusting the volume would do it. The Bluetooth pairing/unpairing experience was rough. And when we played music, it just didn’t sound very good. The back-and-forth between three companies trying to fix bugs became a huge challenge. And worst of all: the cost of the Stereo Dock kept getting higher and higher.

It goes on to say that the team worked to improve the experience, even buying the Bluetooth chipset source code in hopes to fix the issues, to no avail. While there are “a number of beautiful finished prototypes,” Panic says “they weren’t very good to use.” The official update was foreshadowed in a social media exchange shared on Reddit earlier this month in which Panic’s Cabel Sasser, replying to a question about the Stereo Dock’s status, said it “turned out to be incredibly hard for lots of incredibly dumb reasons” and was put on pause.

The team says it’s now turning its focus wholly back to the continuing production of the Playdate. There may still be a chance that we see the Stereo Dock at some point down the line, but don’t hold your breath. “Will it arrive someday? We still hope so,” Panic said. “Will it arrive anytime soon? Unlikely.”

This article originally appeared on Engadget at https://www.engadget.com/gaming/panic-hits-pause-on-the-playdate-stereo-dock-says-its-unlikely-well-see-it-anytime-soon-191607744.html?src=rss 

Google may be close to unveiling an AI agent that can operate a web browser to help users automate everyday tasks. The Information reports that the company is working on a “computer-using agent” under the codename Project Jarvis, and it may be ready to be previewed as soon as December. According to sources that spoke to The Information, Jarvis “responds to a person’s commands by capturing frequent screenshots of what’s on their computer screen, and interpreting the shots before taking actions like clicking on a button or typing into a text field.”

Jarvis is reportedly made to work only with web browsers — particularly Chrome — to assist with common tasks like research, shopping and booking flights. It comes as Google continues to expand the capabilities of its Gemini AI, the next-gen model of which is expected to be revealed in December, as reported by The Verge. Gemini Live, Google’s AI chatbot, gained support for dozens of new languages this month, and Gemini integration has recently made it to Google Meet, Photos and other applications.

The news of Jarvis comes days after Anthropic introduced a similar but seemingly more expansive feature for its Claude AI, which it says has been equipped with computer skills so it can “use a wide range of standard tools and software programs designed for people.” That’s available now in a public beta.

This article originally appeared on Engadget at https://www.engadget.com/ai/google-is-reportedly-developing-jarvis-ai-that-could-take-over-your-web-browser-163404705.html?src=rss 

Google is reportedly working on a new Android API for what it’s calling Rich Ongoing Notifications, which would allow apps to display at-a-glance information in a status bar much like Apple’s Live Activities in the Dynamic Island on iPhone. This is according to journalist Mishaal Rahman, who spotted the code in the Android 15 QPR1 Beta 3 release. It could work a lot like the time tracker that currently appears when you’re on a phone call, with a bit of text in a bubble at the top of the display that you can tap to open the app for more details.

Writing for Android Authority, Rahman says the API “will let apps create chips with their own text and background color that live in the status bar.” It could be especially useful for things like transit updates, allowing users to keep track of pertinent information like departure times or an Uber’s ETA while using other apps. The feature isn’t yet complete, though, and it could still be some time before we see it. Rahman predicts it’ll arrive with Android 16.

This article originally appeared on Engadget at https://www.engadget.com/mobile/ongoing-notifications-similar-to-apples-live-activities-could-be-coming-to-android-210443338.html?src=rss 

New releases in fiction, nonfiction and comics that caught our attention.

This article originally appeared on Engadget at https://www.engadget.com/entertainment/what-to-read-this-weekend-jeff-vandermeer-returns-to-area-x-160004379.html?src=rss 

The legal battle between Apple and medical technology company Masimo rages on, with the bigger company — sorta, kinda — winning their latest face off. A federal jury has agreed with Apple that previous versions of Masimo’s W1 and Freedom (pictured above) watches infringed on its design patents, according to Reuters. It only awarded Apple $250 in damages, which is the smallest amount that could be awarded for patent infringement, but the company’s lawyers reportedly told the court that it wasn’t after money anyway. 

What Apple, which is worth $3.5 trillion, wanted was an injunction on the sales of Masimo’s current smartwatch models. However, the jury determined that those newer models don’t violate Apple’s intellectual property. That is why Masimo is also treating the jury’s decision as a win, telling the news organization that it’s thankful for the verdict that’s “in favor of Masimo and against Apple on nearly all issues.” Apparently, the ruling only affects a “discontinued module and charger.” As for Apple, it told Reuters that it was “glad the jury’s decision today will protect the innovations [it advances] on behalf of [its] customers.”

Masimo sued Apple in 2021, accusing it of infringing on several of its light-based blood-oxygen monitoring patents, while the tech giant countersued a year later. A court sided with Masimo in 2023, forcing Apple to pause sales on its latest smartwatch models, as the US International Trade Commission blocked all Watch Series 9 and Ultra 2 imports into the country. The company appealed and was ultimately able to sell its watches in the country earlier this year by removing the technology from the units offered in the US. 

This article originally appeared on Engadget at https://www.engadget.com/big-tech/apple-wins-250-in-masimo-smartwatch-patent-case-150020340.html?src=rss 

Lyft has agreed to to tell its drivers how much they can truly earn on the ride-hailing platform — and back it up with evidence — as part of its settlement for a lawsuit filed by the US Justice Department and the Federal Trade Commission. The lawsuit accused the company of making “numerous false and misleading claims” in the advertisements it released in 2021 and 2022, when the demand for rides recovered following COVID-19 lockdowns in the previous years. Lyft promised drivers up to $43 an hour in some locations, the FTC said, without revealing that those numbers were based on the earnings of its top drivers. 

The rates it published allegedly didn’t represent drivers’ average earnings and inflated actual earnings by up to 30 percent. Further, the FTC said that Lyft “failed to disclose” that information, as well as the fact that the amounts it published included passengers’ tips. The company also promised in its ads that drivers will get paid a set amount if they complete a certain number of rides within a specific timeframe. A driver is supposed to make $975, for instance, if they complete 45 rides over a weekend. 

Lyft allegedly didn’t clarify that it will only pay the difference between the what the drivers’ earn and its promised guaranteed earnings. Drivers thought they were getting those guaranteed payments on top of their ride payments as a bonus for completing a specific number of rides. The FTC accused Lyft of continuing to make “deceptive earnings claims” even after it sent the company a notice of its concerns in October 2021, as well. 

Earlier this month, the company launched an earnings dashboard that showed the estimated hourly rate for each ride, along with the driver’s daily, weekly and yearly earnings. But under the settlement, Lyft will have to explicitly tell drivers how much their potential take-home pay is based on typical, instead of inflated, earnings. It has to take tips out of the equation, and it has to to clarify that it will only pay the difference between what the drivers get from rides and its guaranteed earnings promise. Finally, it will have to pay a $2.1 million civil penalty. 

This article originally appeared on Engadget at https://www.engadget.com/transportation/lyft-will-have-to-tell-drivers-how-much-they-can-truly-earn-with-evidence-120011572.html?src=rss