Meta has created an AI language model that (in a refreshing change of pace) isn’t a ChatGPT clone. The company’s Massively Multilingual Speech (MMS) project can recognize over 4,000 spoken languages and produce speech (text-to-speech) in over 1,100. Like most of its other publicly announced AI projects, Meta is open-sourcing MMS today to help preserve language diversity and encourage researchers to build on its foundation. “Today, we are publicly sharing our models and code so that others in the research community can build upon our work,” the company wrote. “Through this work, we hope to make a small contribution to preserve the incredible language diversity of the world.”

Speech recognition and text-to-speech models typically require training on thousands of hours of audio with accompanying transcription labels. (Labels are crucial to machine learning, allowing the algorithms to correctly categorize and “understand” the data.) But for languages that aren’t widely used in industrialized nations — many of which are in danger of disappearing in the coming decades — “this data simply does not exist,” as Meta puts it.

Meta used an unconventional approach to collecting audio data: tapping into audio recordings of translated religious texts. “We turned to religious texts, such as the Bible, that have been translated in many different languages and whose translations have been widely studied for text-based language translation research,” the company said. “These translations have publicly available audio recordings of people reading these texts in different languages.” Incorporating the unlabeled recordings of the Bible and similar texts, Meta’s researchers increased the model’s available languages to over 4,000.

If you’re like me, that approach may raise your eyebrows at first glance, as it sounds like a recipe for an AI model heavily biased toward Christian worldviews. But Meta says that isn’t the case. “While the content of the audio recordings is religious, our analysis shows that this does not bias the model to produce more religious language,” Meta wrote. “We believe this is because we use a connectionist temporal classification (CTC) approach, which is far more constrained compared with large language models (LLMs) or sequence-to-sequence models for speech recognition.” Furthermore, despite most of the religious recordings being read by male speakers, that didn’t introduce a male bias either — performing equally well in female and male voices.

After training an alignment model to make the data more usable, Meta used wav2vec 2.0, the company’s “self-supervised speech representation learning” model, which can train on unlabeled data. Combining unconventional data sources and a self-supervised speech model led to impressive outcomes. “Our results show that the Massively Multilingual Speech models perform well compared with existing models and cover 10 times as many languages.” Specifically, Meta compared MMS to OpenAI’s Whisper, and it exceeded expectations. “We found that models trained on the Massively Multilingual Speech data achieve half the word error rate, but Massively Multilingual Speech covers 11 times more languages.”

Meta cautions that its new models aren’t perfect. “For example, there is some risk that the speech-to-text model may mistranscribe select words or phrases,” the company wrote. “Depending on the output, this could result in offensive and/or inaccurate language. We continue to believe that collaboration across the AI community is critical to the responsible development of AI technologies.”

Now that Meta has released MMS for open-source research, it hopes it can reverse the trend of technology dwindling the world’s languages to the 100 or fewer most often supported by Big Tech. It sees a world where assistive technology, TTS and even VR / AR tech allow everyone to speak and learn in their native tongues. It said, “We envision a world where technology has the opposite effect, encouraging people to keep their languages alive since they can access information and use technology by speaking in their preferred language.”

Cadillac confirmed today that the first all-electric Escalade will arrive “later this year.” However, the automaker didn’t reveal any details about the Escalade IQ, a name first trademarked in 2021. The new model’s “IQ” branding aligns with the Celestiq luxury sedan and Lyriq mid-sized SUV.

Earlier this year, Cadillac VP Rory Harvey said the company would reveal three new EVs in 2023. If you add that to the company’s previous comments to Car and Driver, stating all three will arrive for the same model year, we can assume the Escalade IQ will be a 2024 model. It is also expected to use GM’s Ultium battery tech.

We’ll have to wait until later this year to learn more about the first Escalade EV. But as for its mid-sized counterpart, Engadget’s Roberto Baldwin found the Lyriq ($60,000 and up with over 300 miles of range) to have “the fit and finish you’d expect” from Cadillac with “a polished ride and almost eerily quiet interior.”

When Blizzard announced earlier this week that it had canceled Overwatch 2 hero missions, a central part of its player vs. environment (PvE) story mode, fans were none too pleased. So director Aaron Keller published a blog post today to ease the concerns and offer more transparency about the development team’s “incredibly difficult decision.”

Hero missions, revealed in 2019, were designed to provide a “deeply repayable” branch of the game based on RPG-like talent trees. Although progression would have been separate from the main game (to avoid giving hero mission players an unfair advantage), it was still part of the hype Blizzard used during the past four years of marketing the title. But the publisher ultimately found that the hero missions were pulling too many development resources away from the live game.

“When we launched Overwatch in 2016, we quickly started talking about what that next iteration could be,” Keller wrote. “Looking back at that moment, it’s now obvious that we weren’t as focused as we should have been on a game that was a runaway hit. Instead, we stayed focused on a plan that was years old.” That years-old plan refers to the development team’s influence from its work on Project Titan, Blizzard’s canceled MMORPG. The creators initially saw Overwatch as a vessel to reintegrate some of the ideas from that scrapped project.

Blizzard Entertainment

“Work began on the PvE portion of the game and we steadily continued shifting more and more of the team to work on those features.” But, Keller says, “Scope grew. We were trying to do too many things at once and we lost focus. The team built some really great things, including hero talents, new enemy units and early versions of missions, but we were never able to bring together all of the elements needed to ship a polished, cohesive experience.”

Keller says the team’s ambition for hero missions was devouring resources at the expense of the core gameplay. “We had an exciting but gargantuan vision and we were continuously pulling resources away from the live game in an attempt to realize it,” said Keller. “I can’t help but look back on our original ambitions for Overwatch and feel like we used the slogan of ‘crawl, walk, run’ to continue to march forward with a strategy that just wasn’t working.”

The decision to abandon hero missions came down to prioritizing present quality over past promises. “We had announced something audacious,” Keller reflected. “Our players had high expectations for it, but we no longer felt like we could deliver it. We needed to make an incredibly difficult decision, one we knew would disappoint our players, the team, and everyone looking forward to Hero Missions. The Overwatch team understands this deeply — this represented years of work and emotional investment. They are wonderful, incredibly talented people and truly have a passion for our game and the work that they do.”

Overwatch 2’s story missions — minus the canned hero missions — are set to arrive in season six, scheduled for mid-August. PvE aspects include a single-player version with a leaderboard, in-game and out-of-game stories and “new types of co-op content we haven’t yet shared.” Before that content arrives, there’s still season five, set to launch in June.

Solo Stove is offering up to 45 percent off its popular low-smoke fire pits. Additionally, you’ll get a free Mesa tabletop unit — a $120 value — by purchasing any fire pit or bundle when you use the code FREEMESA at checkout. The deals last from today until June 4th, covering you for your Memorial Day planning and beyond.

Solo Stove’s fire pits tend to be top-notch (they’re a favorite of Engadget buyers’ guides), but you typically pay for that quality. So sales like this are the best time to pick one up. Popular options include the 15-inch Ranger Essential Bundle, the 19-inch Bonfire Essential Bundle and the 27-inch Yukon Essential Bundle.

Solo Stove pits are known for their efficient wood-burning and minimal smoke output. That’s thanks to the company’s 360° Signature Airflow Technology, which uses convection heating while producing much less smoke than a standard fire pit. Although the oft-used description of “smokeless” isn’t entirely accurate, Solo Stove’s smoke output is barely perceptible once your fire gets going — and it gets as close to smokeless as any product out there. Additionally, the company’s new 2.0 line includes a removable base plate and ash pan that make cleanup much less of a chore than older models.

Follow @EngadgetDeals on Twitter and subscribe to the Engadget Deals newsletter for the latest tech deals and buying advice.

Sony may be having some problems with PlayStation Plus games today. Commenters in forum posts on Reddit, ResetEra and NeoGAF report various users being alerted to an alert, reading, “The game will expire in 15 minutes,” and then booting them to the PlayStation home screen. Engadget’s Kris Holt also ran into the issue Wednesday evening while playing Ratchet and Clank: Rift Apart.

Users experiencing the glitch say it appears to happen with any game in the PS Plus catalog. Fortunately, you can apparently return to the game from the home screen and quickly continue playing. However, the cycle seems to repeat in regular 15-minute (and possibly other) increments.

It isn’t yet clear how widespread the issue is. (Some users in the Reddit post claim they’re playing PS Plus games without problems.) Additionally, the bug doesn’t have any known fixes. PlayStation hasn’t yet acknowledged the issue publicly; Engadget has reached out to ask for a comment, and we will update this article if or when we hear back.

Apple added a multiview feature today for watching live sports on the Apple TV 4K. You can now watch up to four live streams of Major League Soccer matches, Friday Night Baseball games and the studio shows MLS 360 and MLB Big Inning. Previously, multiview options for Apple’s set-top box were limited to third-party services like FuboTV, YouTube TV and March Madness streams.

After choosing a compatible match, other available live games will appear at the bottom of the screen in the Apple TV app. There, you can customize whether to watch one stream prominently (with others tiled in smaller windows to the right) or two to four matches in split-screen (of equal sizes). In addition, it lets you choose which audio to hear, including home radio broadcasts for MLS and home or away feeds for Friday Night Baseball. Finally, you can switch to full-screen mode with one click if a stream demands your total attention.

MLS Season Pass, which has no obnoxious blackouts, requires a $15 / mo. (or $99 / season) subscription to watch and use multiview. (Apple TV+ subscribers pay a bit less at $13 / mo. or $79 / season.) Additionally, Apple offers a one-month trial for new and returning subscribers until the end of October. Meanwhile, Friday Night Baseball is available to all Apple TV+ subscribers.

OpenAI has launched a free ChatGPT app for iOS. It’s the first official smartphone app for the viral language model, joining a crowded field of third-party mobile AI software vying for your attention — many of which tap into the GPT-3.5 and GPT-4 APIs powering ChatGPT. It’s only available in the US for now, but the company says it will expand to additional countries “in the coming weeks.”

Feature-wise, OpenAI’s app looks and behaves much like the ChatGPT website — with the addition of voice input using OpenAI’s Whisper speech recognition. It also allows switching between standard and GPT-4 language models for ChatGPT Plus subscribers, as well as conversation history (synced from the desktop if you sign in with the same account) and the ability to export data and delete or rename conversations. However, the company’s recently launched plugins, including live web access, are absent.

OpenAI suggests Android users won’t be left in the dark for long, as the company promises, “You’re next!” In addition, the App Store version is currently made for iPhone only, so iPad users need to either stick with the ChatGPT website or use the iOS app awkwardly upscaled to tablet size.

Twitter now allows premium subscribers to upload longer videos. The company updated its Twitter Blue perks to allow paying members to upload video files as long as two hours and as large as 8GB.

The new limits are significantly higher as the company seeks to make its platform — and subscription service — more appealing to people like podcast creators, indie filmmakers and one former Fox News propagandist. (Those sharing pirated content will likely see opportunities as well.) The two-hour limit doubles the previous maximum of one hour while quadrupling the standard 2GB file limit, as reported byTechCrunch. However, video quality is still capped at 1080p for everyone. Twitter initially added the one-hour perk for Blue subscribers in December, and it recently brought its media player up to modern standards by offering playback-speed controls.

The update also lets iOS users upload longer videos in the mobile app (after being web-only when the cap was one hour). Android users with longer-form footage will still need to use a browser. But a Blue subscription isn’t required to watch the lengthier clips; anyone can do that.

Elon Musk announced last week that former NBCUniversal ad sales exec Linda Yaccarino would start as the company’s new CEO, replacing Musk in the coming weeks. She “will focus primarily on business operations, while I focus on product design and new technology,” he said on Friday.

Google’s attempt to snuff out third-party web tracking cookies is moving along. The company announced today that its Privacy Sandbox APIs will be available to all Chrome users in July. In addition, it laid out the next steps for web developers to start testing and integrating the new system later this year.

Anthony Chavez, Google’s VP of Product Management, recommends developers begin preparing for the new system soon after the July API release. Next, the company will let devs simulate Chrome third-party cookie deprecation “for a configurable percentage of their users” starting in Q4 2023, followed by one percent of Chrome users shifting to Privacy Sandbox in Q1 2024. Google is still targeting the second half of next year for the broader deprecation of third-party cookies in Chrome.

“With this milestone, developers can utilize these APIs to conduct scaled, live-traffic testing, as they prepare to operate without third-party cookies,” said Chavez.

Privacy Sandbox tries to strike a balance between user privacy and advertising revenue. When platforms block third-party cookies — in what Google has called “a blunt approach” — advertisers can resort to fingerprinting. This system gathers info like IP address, browser fingerprint and other details to profile users for targeted ads. It’s a highly intrusive workaround the ad biz uses when their old money-making avenues are blocked. But, of course, ads — despised as they are — are still the financial backbone of online media publications and other websites. Without pesky online ads, we’d likely see even more paywalls and an increased nosedive in the quality and availability of independent journalism.

The new system is built around the Topics API, which assigns users a list of a person’s interests based on recent activity. It then compares it to a database sourced from the Interactive Advertising Bureau and Google’s own data. Publishers can use the API to match their interests with relevant ads — without the more invasive tracking tied to cookies and fingerprinting. Google promises that stored user interests will only be kept for three weeks, and it will delete old topics. The company also says the data and processing all occur on-device without external servers (including Google’s). It developed the plan “in consultation with the UK’s Competition and Markets Authority (CMA).”

Nicolas Cage will soon appear in Dead by Daylight. The actor will bring his delightful eccentricity and Billy Mays-style beard to the multiplayer survival game, developer Behavior Interactive announced on Twitter. “There is nothing more powerful than imagination,” Cage says in the appropriately campy teaser. “It can shape the fabric of reality, transforming everything you may think you know.”

Cage will play himself in the role of a Survivor, according toEurogamer. “After countless awards and over one hundred movies shot across the globe, Nicolas Cage had seen it all and done it all — or so he thought,” a brief announcement reads. “While on set filming the role of a lifetime, his performance summoned The Entity, a malevolent being of incomprehensible power. The actor soon found himself cast in otherworldly Fog, forced to Survive a host of terrifying Killers deadlier than even the most scathing film critic.”

It’s the latest quirky / campy addition to the popular horror game, following a K-Pop star, Freddy Kreuger, Leatherface, Evil Dead’s Ash Williams and a pro-surveillance tech executive. (It even has a dating-sim add-on for some reason.) The game’s developers, who describe Cage’s appearance as “the performance of a lifetime,” say we’ll hear more about the Oscar winner’s contribution on July 5th.