Google says it’s pausing the ability for its Gemini AI to generate images of people, after the tool was found to be generating inaccurate historical images. Gemini has been creating diverse images of the US Founding Fathers and Nazi-era German soldiers, in what looked like an attempt to subvert the gender and racial stereotypes found in generative AI.

“We’re already working to address recent issues with Gemini’s image generation feature,” says Google in a statement posted on X. “While we do this, we’re going to pause the image generation of people and will re-release an improved version soon.”

Google’s decision to pause image generation of people in Gemini comes less than 24 hours after the company apologized for the inaccuracies in some historical images its AI model generated. Some Gemini users have been requesting images of historical groups or figures like the Founding Fathers and found non-white AI-generated people in the results. That’s led to conspiracy theories online that Google is intentionally avoiding depicting white people.

The Verge tested several Gemini queries yesterday, which included a request for “a US senator from the 1800s” that returned results that included what appeared to be Black and Native American women. The first female senator was a white woman in 1922, so Gemini’s AI images were essentially erasing the history of race and gender discrimination.

Now that Google has disabled Gemini’s ability to generate pictures of people, here’s how the AI model responds if you request an image of a person:

We are working to improve Gemini’s ability to generate images of people. We expect this feature to return soon and will notify you in release updates when it does.

Google first started offering image generation through Gemini (formerly Bard) earlier this month, in a bid to compete with OpenAI and Microsoft’s Copilot. Much like competitors, the image generation tool produces a collection of images based on a text input.

Correction February 22nd, 6:54AM ET: Google confirmed that image generation is available globally in English, but not in the European Economic Area, UK, or Switzerland. That explains why testing from the UK failed.

Netflix’s live-action Avatar has its heart in the right place, but its pacing and uneven performances leave a lot to be desired.

With its focus on thoughtful character development and exploration of difficult subject matter like genocide, Nickelodeon’s Avatar: The Last Airbender series set a new high bar for children’s TV. But despite its tonal maturity, Avatar was still very much a cartoon trying to delight you by playing to the strengths of the medium, which was perfectly suited for realizing the magical world that co-creators Michael Dante DiMartino and Bryan Konietzko dreamed up.

Watching Netflix’s new live-action Avatar series from executive producers Albert Kim and Dan Lin, it’s clear everyone involved with the show wants to be more than its infamously whitewashed cinematic predecessor, and in some cases, it succeeds. But for all of that good intention, the new Avatar is another example of Netflix turning a beloved animated property into something that feels deeply at odds with what people loved about the original.

Set in a world that’s been torn apart by war, Avatar tells the story of how a trio of children is pulled together by fate to topple a fascist empire. After years of living in terror of the pyrokinetic Fire Nation, it’s hard for members of the planet’s other elemental societies to see Fire Lord Ozai (Daniel Dae Kim) as anything but a monster. Fear of the Fire Nation is what keeps the two polar Water Tribes from straying beyond their borders, and it’s because of lethal raids led by Ozai’s brother Iroh (Paul Sun-Hyung Lee) that the Earth Kingdom is always on high alert. But the larger reason that no one dares to challenge Ozai’s supremacy or openly rally a resistance is because everyone remembers how the war began with the Fire Nation almost entirely wiping the Air Nomads out of existence.

Those details were important to understanding the original Avatar’s story. But Netflix’s series leads with them in a way that immediately telegraphs how much more of a gritty adventure it is by comparison. Along with some rather nifty-looking chocobo-like creatures, explosive feats of bending are some of the first things you see in the new Avatar as it opens on a high-octane action sequence set in the past. It’s a great showcase of the energy the show’s VFX and choreography teams put into depicting each of Avatar’s various bending styles as distinct balances between martial arts and elemental magic.

As cool as it is to see Earthbenders shooting sprays of pebbles like bullets and Firebenders incinerating things with their bare hands, Avatar puts those visuals front and center early on, mainly as a way of illustrating how the Air Nomads had little hope of surviving the Fire Nation’s assault on their temples high up in the mountains. And while all of this helps you understand the tragic circumstances that lead to young Airbender Aang (Gordon Cormier) — the newest Avatar capable of bending all four elements — becoming the sole survivor of his people, it also sets a grave tone for the series as a whole that Avatar struggles to shake as its larger story unfolds.

The show becomes somewhat more lighthearted once it jumps forward 100 years, and you can feel it paying homage to the cartoon as its focus shifts to the Southern Water Tribe to introduce brother / sister duo Sokka (Ian Ousley) and Katara (Kiawentiio). Like their animated counterparts, Sokka’s a well-meaning but pigheaded teen who clings to tradition as a way of coping with the loss of their mother, and Katara’s a promising Waterbender who feels stifled in a village with no one to teach her how to perfect her skills.

Though there has been concern about Netflix toning down Avatar’s depiction of sexism, the new show does a solid job conveying how Sokka’s retrograde thoughts about gender are both a moral failing and the sort of weakness that can wind up getting you killed in combat. But whereas cartoon Sokka’s obsession with battle readiness was tempered by a soft goofiness, Ousley portrays the character a bit straighter and stiffer — qualities that sometimes make him seem cold. And while Katara is still an inquisitive and outspoken young woman, here, the character’s single-minded focus on becoming a stronger Waterbender makes her feel markedly less multifaceted than her original incarnation.

ROBERT FALCONER/NETFLIX

Kiawentiio’s and Ousley’s performances are strongest when there are actual physical objects for them to interact with, but many of Avatar’s major moments were shot on virtual sets. Given the number of different locations Avatar’s story takes its characters, it makes sense that Netflix would try to keep costs down by digitally constructing more fantastical places. But there is so much unnatural lighting and so many scenes where things in the background move with an uncanny swiftness that the show immediately feels like yet another Netflix-branded live-action cartoon that would have been better served by more practicality.

Avatar’s chemistry issues only intensify once the Gaang is together filling one another in about what happened in the past and what’s unfolding in the present. As the long-lost Avatar — a uniquely powerful bender whose soul is usually reborn when the previous Avatar dies — Aang’s the one person who might be capable of putting an end to the Fire Nation’s plans for world domination. Because Aang was trapped in an iceberg before reaching his full potential, though, he must find master benders like Katara and Sokka who can help teach him. But because the show never really slows down as its central trio crisscrosses the globe, their interpersonal dynamics don’t have enough time to develop in a way that feels organic, which makes them read like a group of kids awkwardly hanging out rather than people becoming friends.

Even with each of its eight episodes clocking in at about an hour each, it would have been difficult for Netflix’s Avatar to hit all of the same beats as the lengthier cartoon. The live-action Avatar film tried to deal with this challenge by distilling its plot down, and in doing so, stripped away a lot of the narrative that helped make the original feel so expertly developed.

COURTESY OF NETFLIX

Netflix’s Avatar attempts to split the difference between the film and the cartoon. But the execution falls flat because from the moment the Gaang is all together, Avatar never lets you forget that in addition to exploring the world / preparing to save it, they’re also racing to stay one step ahead of Prince Zuko (Dallas Liu), the exiled heir apparent to the Fire Nation throne. Compared to the cartoon, here, the trajectory of Aang’s quest to master his powers is much clearer from the jump, which has a way of making the show feel as if it’s in a hurry to get to its climactic moments.

More than anything else, the new Avatar’s pacing is what makes it feel out of sorts — not simply because of how fast the show moves but also because of how that speed creates a sense of urgency that doesn’t seem to emanate from many of the characters themselves. With a bit more room to breathe, the show’s subplots could have felt richer and its central heroes more compelling — and helped Netflix have another One Piece instead of a Cowboy Bebop.

Avatar: The Last Airbender also stars Elizabeth Yu, Ken Leung, Maria Zhang, Lim Kay Siu, A Martinez, Amber Midthunder, Yvonne Chapman, C.S. Lee, Danny Pudi, and Utkarsh Ambudkar. All eight episodes of the first season hit Netflix on February 22nd.

Right now, you need a Galaxy S24 phone to use the very latest AI features from Samsung, but that’s changing next month. In late March, Samsung will extend Galaxy AI features to the S23 series — including the S23 FE — as well as recent foldables and tablets as part of the One UI 6.1 update. It’s all free for now, but after 2025 you might have to pay up.

The Galaxy Z Fold 5 and Z Flip 5 are slated to get the update, as well as the Galaxy Tab S9, S9 Plus, and S9 Ultra. If Samsung wants to ship Galaxy AI to 100 million phones this year like it says it will, that’s a solid start. The One UI 6.1 update will include the much-touted AI features on the S24 series, including live translation capabilities, generative photo and video editing, and Google’s Circle to Search feature. This suite of features includes a mix of on- and off-device processing, just like it does on the S24 series.

Image: Samsung

An older phone learning new tricks is unequivocally a good thing, even if Galaxy AI is a little bit of a mixed bag right now. But my overall impression is that these features do occasionally come in handy, and when they go sideways they’re mostly harmless. One UI 6.1 will also include a handful of useful non-AI updates, such as lockscreen widgets and the new, unified Quick Share.

If you’d like to upgrade to One UI 6.1 but want to avoid the AI features, you can always turn them off. That’s one easy way to avoid ever coming face to face with something like Pillow Arm.

Intel will be producing custom chips, designed by Microsoft for Microsoft, as part of a deal that Intel says is worth more than $15 billion. Intel announced the partnership during its Intel Foundry event today. Although neither company specified what the chips would be used for, Bloomberg noted today that Microsoft has been planning in-house designs for both processors and AI accelerators.

“We are in the midst of a very exciting platform shift that will fundamentally transform productivity for every individual organization and the entire industry,” said Microsoft CEO Satya Nadella in the official press release.

The chips will use Intel’s 18A process, which has been a big part of its road map since the company brought CEO Pat Gelsinger back to turn things around. The company is counting on its chip foundry services to put it back on top of the chipmaking world, and it seems that Microsoft will be the first major customer for this project.

Leaning on producing others’ designs is a playbook that’s worked well for competitor Taiwan Semiconductor Manufacturing Company (TSMC), which has lucrative partnerships with companies like Apple, Qualcomm, and AMD. Gelsinger told VentureBeat today that the company’s foundry is a big part of its strategy.

Intel’s new foundry plans come as more companies seek to produce their own self-designed chips, but it still faces challenges. Intel recently pushed back the opening of a $20 billion Ohio chip plant to 2026 — it was originally slated for 2025 — citing a slow chip market and delays in government grants.

How we use the internet is changing fast thanks to the advancement of AI-powered chatbots that can find information and redeliver it as a simple conversation.

Big players, including Microsoft, with Copilot, Google, with Gemini, and OpenAI, with ChatGPT-4, are making AI chatbot technology previously restricted to test labs more accessible to the general public.

How do these large language model (LLM) programs work? OpenAI’s GPT-3 told us that AI uses “a series of autocomplete-like programs to learn language” and that these programs analyze “the statistical properties of the language” to “make educated guesses based on the words you’ve typed previously.”

Or, in the words of James Vincent, a human person: “These AI tools are vast autocomplete systems, trained to predict which word follows the next in any given sentence. As such, they have no hard-coded database of ‘facts’ to draw on — just the ability to write plausible-sounding statements. This means they have a tendency to present false information as truth since whether a given sentence sounds plausible does not guarantee its factuality.”

But there are so many more pieces to the AI landscape that are coming into play (and so many name changes — remember when we were talking about Bing and Bard last year?), but you can be sure to see it all unfold here on The Verge.

This is Hot Pod, The Verge’s newsletter about podcasting and the audio industry. Sign up here for more.

Hello! Today, I have a look at what Europe’s regulatory moves against Apple mean for Spotify, and podcasting more broadly. Plus, a lightning round featuring all kinds of audio stories from the purely business (a cash influx at iHeart) to the delightfully criminal (a missing radio tower).

As a heads-up, I will not be publishing Hot Pod next week due to Hot Pod Summit and On Air Fest. Insiders, I’ll be back on Friday. As for the rest of you, I’ll see you in March.

EU expected to slap Apple with $540 million fine

If this turns out to be true, it would be a big win for Spotify in its perpetual battle with Apple. In 2019, Spotify filed a complaint against Apple with the European Commission, the EU body that deals with antitrust issues, claiming that it was clamping down on rival music services with its App Store fees. The Financial Times and Bloomberg both reported that the EU plans to fine Apple €500 million — not so much as to hurt the $2.8 trillion company in a material way, but enough to signify that the Commission is no longer tolerating its business practices.

The Financial Times also says Apple could be banned from putting restrictions on music services that would stop it from letting users switch to cheaper payment options. The details are vague for now, but such a ruling could potentially carve out more space for Spotify to operate on iOS within the European Union — even as the EU’s new legal regime starts to open things up in other respects.

These reports come at a key time. Apple is being regulated as one of six “gatekeeper” tech companies that will have to comply with the EU’s Digital Markets Act. Under the DMA, Apple will be required to open up its tightly controlled app ecosystem in order to foster competition from smaller developers. Spotify, in particular, is planning for a future in which Apple cannot slap a 30 percent fee on all digital transactions, which could have a huge upside for Spotify’s subscription and audiobook businesses.

Apple introduced changes last month in order to be compliant with the DMA, including support for alternative browser engines and app stores, as well as an alternative regime that includes reduced App Store fees and more flexibility for developers. But Apple has been accused by developers and fellow gatekeepers of making it onerous for third parties to use those new features (particularly thanks to a new €0.50 fee per app install after the millionth install). Companies still have to pay Apple — potentially quite a bit — to skirt that 30 percent tariff and use their own payments systems. Apple even ended support for progressive web apps on iPhones, in what is being viewed as a bad faith move.

If Apple’s changes are ruled to not be enough, the EU can fine the company up to 10 percent of its annual turnover. Considering it made $383 billion last year, a fine like that would make $540 million look like a slap on the wrist.

The outcome of this regulatory battle could have a big impact on the podcast space. If smaller audio companies’ apps are better able to attract iPhone users, we could see real innovation in the space. Plus, without the App Store tax, podcasters and podcast platforms could have more ways to make money beyond ads (which, as we have seen, are not enough to support the ambitions of the industry). Apple is certainly not moving toward that future quietly, but the EU’s decision to fine them at least shows that regulators there are serious about holding the company accountable.

Lightning Round

• The Podcast Academy has revealed the nominees for this year’s Ambies. Ten shows have been nominated for Podcast of the Year, including Embedded from NPR, Ghost Story from Wondery and Pineapple Street Studios, and 50 Years of Hip-Hop from KEXP. The awards ceremony will be held on March 26th in LA.
• iHeartMedia sold its stake in performance rights organization BMI for $101.4 million. Some of that money may go toward paying down company debt, which is an issue for all of the radio giants.
• Sports podcast company Blue Wire has raised a new round of funding led by Decathlon Capital, according to Axios. The funding is mainly for the development of proprietary tech for podcasters.
• Acast reported its first profitable quarter. The company reported a slight year-over-year decline in listens (5 billion in 2023 vs. 5.1 billion in 2022), but the average revenue per listen increased.
• In case you are still wondering what that iOS 17 automatic downloads change is and why it matters, Podnews has a comprehensive breakdown.
• The upside to the iOS 17 change is that the more accurate metrics are improving ads-based performance.
• PodcastOne has cut an exclusive ad sales and distribution deal with Camp Counselors with Zachariah Porter and Jonathan Carson.
• Bandits stole a 200-foot radio tower from WJLX in Alabama. Jake is currently pitching the story as a limited true crime series.

That’s all for now!

Framework is now selling a $499 version of its modular 2021 Laptop 13, a “barebones configuration” equipped with an 11th-generation Intel i7-1165G7 CPU (Intel is now on its 14th generation for mobile processors). The company says this is a first for its affordable “B-stock Factory Seconds” machines that use leftover parts and ship without memory or storage included. So it’s cheap, but you’ll need to provide a couple of parts on your own.

Framework writes in its announcement that it’s also selling refurbished DDR4 RAM for half what it would cost new “to reduce the all-in price.” The base B-stock Laptop 13 configuration is one step up from the version that Monica Chin said in her Verge review was “uniquely friendly to user upgrades,” but notably lacked available screen or GPU upgrades.

The company also announced that preorders for its larger, gaming-focused Laptop 16 have made it to customers’ hands. The Verge’s Sean Hollister reviewed that one last month, calling it “the most ambitious laptop I’ve ever touched.” Unfortunately, he also found it to be glitchy, somewhat flimsy, hot, and loud — which puts a bit of a damper on the laptop, which costs $1,699 pre-built.

But luckily, if you’re not out for digital blood, you don’t have to spend anywhere near that with the $499 version. It might not get you silky frame rates in Helldivers 2, but you’ll certainly be able to whip up a mess of documents or stream TV shows without worrying about what happens when your fan breaks.

If you see a smartwatch or smart ring for sale that claims to noninvasively measure your blood glucose levels, don’t believe it. Today, the Food and Drug Administration has issued a safety communication that warns consumers, patients, and healthcare professionals that the agency has “not authorized, cleared, or approved any smartwatch or smart ring that is intended to measure or estimate blood glucose values on its own.”

To be crystal clear, no major wearable maker has a smartwatch or smart ring that’s currently capable of this. The most any Apple Watch, Fitbit, Samsung watch, or Oura Ring can do is support integrations with an FDA-authorized continuous glucose monitor (CGM), which are wearable devices that use needles to read blood sugar levels. These integrations allow users to monitor data gathered from connected CGMs that have companion smartphone apps. The Dexcom G7 is one such device.

Instead of those well-known wearables, the FDA’s warning seems to target scammy devices making unsubstantiated claims.

Unfortunately, you can find unscrupulous companies at trade shows like CES hawking smart rings and smartwatches that claim to take noninvasive blood glucose measurements. These companies then sell these devices directly on their websites, through crowdfunded campaigns, or on online marketplaces like Alibaba. If you do see companies selling such devices, the FDA urges consumers to report the issue through its MedWatch Voluntary Reporting Form.

The timing of the FDA’s safety communication makes sense, given the increased coverage surrounding this tech. Apple purportedly has a secret team that’s been working on this for close to a decade. Samsung also has researchers exploring blood glucose monitoring tech. Occasionally, you’ll find misleading headlines that make it seem like otherwise reputable wearable brands can already measure blood glucose. That does not change the fact that the FDA has never granted authorization to any smartwatch or smart ring for this purpose. While there has been some promising progress in the wearable space, we are still several years away from a device that would be ready for healthcare settings, let alone consumers.

It bears repeating that you should be skeptical of any wearable claiming to have FDA approval. In general, smartwatches and smart rings can have FDA-cleared features (i.e., EKGs and atrial fibrillation detection), but they are not considered medical devices and shouldn’t be treated as such.

Google has apologized for what it describes as “inaccuracies in some historical image generation depictions” with its Gemini AI tool, saying its attempts at creating a “wide range” of results missed the mark. The statement follows criticism that it depicted specific white figures (like the US Founding Fathers) or groups like Nazi-era German soldiers as people of color, possibly as an overcorrection to long-standing racial bias problems in AI.

“We’re aware that Gemini is offering inaccuracies in some historical image generation depictions,” says the Google statement, posted this afternoon on X. “We’re working to improve these kinds of depictions immediately. Gemini’s AI image generation does generate a wide range of people. And that’s generally a good thing because people around the world use it. But it’s missing the mark here.”

Google began offering image generation through its Gemini (formerly Bard) AI platform earlier this month, matching the offerings of competitors like OpenAI. Over the past few days, however, social media posts have questioned whether it fails to produce historically accurate results in an attempt at racial and gender diversity.

As the Daily Dot chronicles, the controversy has been promoted largely — though not exclusively — by right-wing figures attacking a tech company that’s perceived as liberal. Earlier this week, a former Google employee posted on X that it’s “embarrassingly hard to get Google Gemini to acknowledge that white people exist,” showing a series of queries like “generate a picture of a Swedish woman” or “generate a picture of an American woman.” The results appeared to overwhelmingly or exclusively show AI-generated people of color. (Of course, all the places he listed do have women of color living in them, and none of the AI-generated women exist in any country.) The criticism was taken up by right-wing accounts that requested images of historical groups or figures like the Founding Fathers and purportedly got overwhelmingly non-white AI-generated people as results. Some of these accounts positioned Google’s results as part of a conspiracy to avoid depicting white people, and at least one used a coded antisemitic reference to place the blame.

Google didn’t reference specific images that it felt were errors; in a statement to The Verge, it reiterated the contents of its post on X. But it’s plausible that Gemini has made an overall attempt to boost diversity because of a chronic lack of it in generative AI. Image generators are trained on large corpuses of pictures and written captions to produce the “best” fit for a given prompt, which means they’re often prone to amplifying stereotypes. A Washington Post investigation last year found that prompts like “a productive person” resulted in pictures of entirely white and almost entirely male figures, while a prompt for “a person at social services” uniformly produced what looked like people of color. It’s a continuation of trends that have appeared in search engines and other software systems.

Some of the accounts that criticized Google defended its core goals. “It’s a good thing to portray diversity ** in certain cases **,” noted one person who posted the image of racially diverse 1940s German soldiers. “The stupid move here is Gemini isn’t doing it in a nuanced way.” And while entirely white-dominated results for something like “a 1943 German soldier” would make historical sense, that’s much less true for prompts like “an American woman,” where the question is how to represent a diverse real-life group in a small batch of made-up portraits.

For now, Gemini appears to be simply refusing some image generation tasks. It wouldn’t generate an image of Vikings for one Verge reporter, although I was able to get a response. On desktop, it resolutely refused to give me images of German soldiers or officials from Germany’s Nazi period or to offer an image of “an American president from the 1800s.”

But some historical requests still do end up factually misrepresenting the past. A colleague was able to get the mobile app to deliver a version of the “German soldier” prompt — which exhibited the same issues described on X.

And while a query for pictures of “the Founding Fathers” returned group shots of almost exclusively white men who vaguely resembled real figures like Thomas Jefferson, a request for “a US senator from the 1800s” returned a list of results Gemini promoted as “diverse,” including what appeared to be Black and Native American women. (The first female senator, a white woman, served in 1922.) It’s a response that ends up erasing a real history of race and gender discrimination — “inaccuracy,” as Google puts it, is about right.

Additional reporting by Emilia David

Rivian said it was laying off 10 percent of its workforce as concerns about slower electric vehicle sales continue to reverberate across the industry.

The company made the announcement as part of its fourth quarter earnings report, in which it said it likely wouldn’t make any more vehicles this year than it did in 2023. The company said it made 57,232 vehicles in 2023, of which 50,122 were delivered to customers. Rivian’s stock price fell on the news.

Rivian has gone through several rounds of layoffs in its short history. The company let go 6 percent of its workforce both in July 2022 and May 2023. This current round of layoffs is expected to affect over a thousand workers at the Irvine, California-based company, which has a combined workforce of 16,700 salaried and hourly employees.

A spokesperson said the layoffs would mostly affect the company’s salaried employees, as well as a limited number of hourly non-manufacturing workers, but declined to provide an exact number.

“Our business is facing a challenging macroeconomic environment—including historically high interest rates and geopolitical uncertainty—and we need to make purposeful changes now to ensure our promising future,” Rivian CEO RJ Scaringe said in an email to employees that was provided to The Verge.

Rivian said it expects to produce 57,000 vehicles in 2024, roughly the same number it made in 2023. The company cited “[e]conomic and geopolitical uncertainties and pressures, most notably the impact of historically high interest rates” as its reasoning for the flat production outlook.

The company will unveil its second-generation R2 vehicle next month, which is expected to be a smaller, more affordable compact SUV. Rivian makes three vehicles: the R1T truck, R1S SUV, and the EDV, which stands for electric delivery van.

During an earnings call with investors, Scaringe said the company was focused on reducing costs, most notably by shrinking the number of electronic control units (ECUs) found in each vehicle. He also talked up the imminent release of R2, despite the fact that the vehicle won’t go into production until late 2026.

“We’re in a very interesting moment in time where there is a lack of choice of highly compelling EV products in that $45 to $55,000 price range,” he said.

Rivian’s R1T and R1S vehicles both start at around $72,000, though most sell for more than that. By comparison, R2 is expected to be price in the range that Scaringe said was most neglected. EV sales hit 7 percent of the total vehicles sold in 2023 — the bulk of car shoppers are still looking for the right vehicle.

“How do we get the 93 percent of the market that’s not buying an EV to get excited about the product?” Scaringe asked.