Emails from the Department of Justice’s antitrust case against Google revealed how Microsoft executives were alarmed by and even envious of Google’s AI lead.

In an email thread, CTO Kevin Scott wrote he was “very, very worried” about Google’s rapidly growing AI capabilities. He said he initially dismissed the company’s “game-playing stunts,” likely referring to Google’s AlphaGo models. The emails reference Gmail’s autocomplete features, which execs called “scary good.” Microsoft struggled to copy Google’s BERT-large, an AI model that deciphers the meaning and context of words in a sentence. It took the company six hours to replicate the model, while Google inched further ahead on more elaborate, bigger models.

Scott said Microsoft had “very smart” people on its machine-learning teams but their ambitions had been curbed and that their company was “multiple years behind the competition in terms of ML scale.” This all led to a billion-dollar push into OpenAI in 2019. It’s since invested $13 billion.

— Mat Smith

The biggest stories you might have missed

The Cheyenne Supercomputer is going for a fraction of its list price at auction right now

Batman: Arkham Shadow is the first big exclusive VR game for the Quest 3

May's PlayStation Plus games include Ghostrunner 2 and the modern classic Tunic

​​You can get these reports delivered daily direct to your inbox. Subscribe right here!

LinkedIn now has daily Wordle-style games

What connects you with a B2B marketer in West Virginia? Four letters.

LinkedIn, the career-centric social network, is getting into gaming. But the kind of earnest, word-based games your mom would let you play when you were a kid. LinkedIn describes them as “thinking-oriented games,” though the format will likely look familiar to fans of The New York Times Games app. You can only play each game once a day, and you can share your score with friends. And just maybe... strike up a conversation on how you can help each other with targeted SaaS projects. Yes, I have feelings about who hits me up on LinkedIn.

Continue reading.

TikTok might be trying to circumvent Apple’s in-app purchase rules

It appears to be directing users to “avoid in-app service fees.”

TikTok is allegedly violating Apple’s App Store rules, with the app allowing (even recommending) particular users to purchase its coins directly from its website. TikTok has apparently given some iOS users the option to “Try recharging on tiktok.com to avoid in-app service fees” — namely Apple’s 30 percent commission on purchases, which are more likely than not passed onto those users. It’s definitely not available to all users and seems to be there for TikTok users who have previously bought a large number of coins — the TikTok whales, if you will.

Continue reading.

Rabbit denies claims its R1 virtual assistant is a glorified Android app

Someone pulled the APK out and put it on an Android phone.

Rabbit

The Rabbit R1, a pocket-sized AI virtual assistant device, runs Android under the hood. Now early users have been able to tease out the R1 APK, install it on an Android phone and make it work — if not with all the features. If that’s the case, what’s the point in the $200 gadget?

In a statement sent to Android Authority, Rabbit CEO Jesse Lyu, said the Rabbit R1 is “not an Android app.” He added the R1 ran on very bespoke AOSP (Android Open Source Project) build and lower-level firmware modifications, so a local bootleg APK won’t be able to access most R1 services. We’re wrapping up our own detailed review — stay tuned.

Continue reading.

Crunchyroll, like many other streaming services recently, is raising its subscription prices. The anime streaming service has announced its first price hike since it was acquired by Funimation in 2020. Subscribers in Argentina, Colombia, France, Portugal, the United States and select additional countries will now have to pay $12 for the Mega Fan tier, up $2 from $10. Meanwhile, the Ultimate Fan tier will now be $1 more expensive at $16 a month. 

Both options give subscribers access to offline viewing and the Crunchyroll Game Vault, which contains a library of mobile games. The Mega Fan tier allows streaming on up to four devices at a time, while the Ultimate Fan tier allows streaming on up to six. People subscribed to the most expensive option also get a swag bag if they keep paying for the service for 12 consecutive months. The basic Fan Tier doesn't come with the perks these two have, but its price remains unchanged at $8 a month. Unfortunately, those who haven't decided whether to pay for a subscription yet can now only test the service out for seven days instead of 14 like before. But if they don't mind watching their anime with ads, they can still view more than 1,000 hours of content for free.

Sony's Funimation purchased Crunchyroll from WarnerMedia for $1.175 billion back in 2020, but it took a while before they were able to complete their transformation into a unified anime subscription service under the latter's name. Funimation didn't shut down its old app and website until April 2 this year after it moved its available titles to Crunchyroll's service. 

TikTok and Universal Music Group (UMG) have signed a deal that will allow Taylor Swift, Olivia Rodrigo, The Weeknd and other artists to return to the platform, the companies announced in a press release. Universal pulled songs from its roster of performers back in February and with some exceptions, its music hasn't been there since. Both sides are now "working expeditiously" to get content back on the platform that's home to a billion-plus users. 

A key part of the deal is artist protection from generative AI. "TikTok and UMG will work together to ensure AI development across the music industry will protect human artistry and the economics that flow to those artists and songwriters," the companies wrote. "TikTok is also committed to working with UMG to remove unauthorized AI-generated music from the platform, as well as tools to improve artist and songwriter attribution."

Also part of the deal are "new monetization opportunities" from TikTok's recent expansion into e-commerce. TikTok will reportedly also assist artists by providing tools around analytics, integrated ticketing, an "Add to Music App" and more. 

Universal took the drastic move of pulling music earlier this year, forcing the platform to mute videos or replace tracks with options from other labels. "As our negotiations continued, TikTok attempted to bully us into accepting a deal worth less than the previous deal, far less than fair market value and not reflective of their exponential growth," UMG said at the time. 

Last month, Taylor Swift's songs returned to TikTok, likely because she has full control of her own catalogue and was able to strike a separate deal. Some songs by other UMG artists, including Ariana Grande, also started appearing on the platform.

The dispute appears to be water under the bridge, but it's the least of TikTok's problems at the moment. US Congress recently voted in favor of a bill that would see TikTok banned in a year unless owner ByteDance sells the app. 

Over a year after announcing it would acquire Mint Mobile for up to $1.35 billion, T-Mobile has closed the deal. With the Un-Carrier's purchase of parent Ka'ena Corporation, it will not only get Mint, but internationally focused prepaid operator Ultra Mobile and wholesale wireless provider Plum. T-Mobile also promised to keep Mint Mobile's $15 per month/5GB offering that's among the least costly in the US.

Mint Mobile is backed by Ryan Reynolds, who is believed to own 20 to 25 percent of the company. The purchase was announced back in March 2023, but the FCC only approved the deal last week. Mint will continue to be operated as a separate brand with Reynolds as pitchman and founders David Glickman and Rizwan Kassim joining T-Mobile to guide the brands. 

Mint Mobile's $15 plan has a few loopholes, namely you have to pay for three months at the start to get that rate, then pay for an entire year to keep it ($180 in total). As a perk, T-Mobile is offering "unlimited" (40GB with throttling, really) data for the first months. Customers will also get unlimited talk and text in Canada, along with 3GB of roaming data. 

Mint's rivals include T-Mobile itself, which has a very similar plan but a hard cap at 5GB, along with AT&T, which offers 15GB for $25 including 10GB of hotspot data. A one-year contract and up-front payment is required for the latter. 

Anthropic is making its Claude AI easier to access on mobile. The company has released a Claude mobile app for iOS that any user can download for free. Similar to the mobile web version of the chatbot, the app syncs users' conversations with Claude across devices, allowing them to jump from a computer to the app (or vice versa) without losing their chat history. Users will also be able to upload files and images straight from their iPhone's gallery — or take a photo on the spot — if they need Claude to process or analyze them in real time. They'll be able to download and access the Claude app whatever plan they're using, even if they're not paying for the service. 

If they do decide to pay for Claude, they now have a new option other than Pro. The new Team plan provides greater usage than the Pro tier so that members can have more conversations with the chatbot. It also enables users to process longer documents, such as research papers and contracts, thanks to its 200,000 context window. The Team plan gives users access to the Claude 3 model family, as well, which includes Opus, Sonnet and Haiku. It will cost subscribers $30 per user per month, with a minimum head count of five users per team. 

Back in March, Anthropic claimed in a blog post that its Claude 3 language model had outperformed ChatGPT and Google's Gemini in several key industry benchmarks. It was better at graduate-level reasoning, multilingual math and coding (among many other metrics), the company said, showing Claude 3's benchmark results against its staunchest rivals. The most powerful Claude 3 model, the Opus, even apparently showed "near-human" abilities with rapid response rates that make it ideal for more complex and time-sensitive tasks. 

This one's a bummer. Mega-publisher Take-Two Interactive is shuttering Rollerdrome studio Roll7 and Kerbal Space Program 2 team Intercept Games, according to paperwork seen by Bloomberg.

Roll7 is based in London, and was founded in 2008 by lifelong friends Tom Hegarty and Simon Bennett. Roll7 is the studio behind OlliOlli, OlliOlli World and Rollerdrome, all fantastic games with wheel-based mechanics. OlliOlli was a Vita hit in 2014 and World landed in early 2022 — they're both great, and the latter in particular is a flow-state-inducing skateboarding platformer with an adorable art style. Rollerdrome was one of our favorite games of 2022; it's a luscious third-person rollerskating-and-gunplay game that looks like a slice of 1970s dystopian sci-fi. 

Roll7 has picked up multiple prestigious awards over the years, including recent wins at BAFTA and DICE. As the studio name implies, Roll7 developers know how to make incredibly smooth action games.

Take-Two purchased Roll7 in November 2021 and made it a subsidiary of Private Division, the company's label for small- and mid-size publishing deals. According to Bloomberg, Take-Two plans to close Roll7 and will offer severance packages to staff.

Intercept Games is based in Seattle and is responsible for Kerbal Space Program 2, a popular flight-simulation title that's still technically in early access on Steam. Take-Two founded Intercept in 2020 specifically to manage Kerbal Space Program 2, and the game has been receiving updates since going live in February 2023.

Take-Two has yet to confirm that it's closing Intercept Games — but it hasn't said it isn't, either. The company filed a notice in Washington on Monday outlining plans to lay off 70 people in the state and permanently close their place of business, and some Kerbal developers have confirmed their recent departures. Private Division will continue to update Kerbal Space Program 2, Take-Two said in a statement.

Take-Two is one of the largest video game companies around, reporting $5.3 billion in revenue last year. It's the owner of Grand Theft Auto and the parent company of Rockstar Games, 2K, Private Division, Zynga and — very recently — Gearbox Software. Take-Two purchased Borderlands studio Gearbox in March for $460 million. Grand Theft Auto VI, arguably the most anticipated game of the decade, is due to add billions to Take-Two's bottom line in 2025.

In April, Take-Two announced plans to lay off 5 percent of its employees, or roughly 600 people, by the end of 2024. It also canceled some in-development projects. When news of the planned firings broke last month, Take-Two didn't identify which studios would take the hit, but now we know it includes Roll7 and Intercept. The company laid off some Private Division workers in 2023 as well.

An estimated 9,400 people have been laid off in the video game industry so far in 2024, and a total of 10,500 workers were let go in 2023. Sony, Microsoft and Riot Games have fired a combined 3,300 employees this year alone, and the fallout from Embracer Group's funding implosion keeps spreading, with numerous shuttered studios and more than 1,400 displaced workers.

Snapchat will finally join most of its messaging app peers and allow users to edit their chats. The feature, which will be rolling out “soon,” will initially be limited to Snapchat+ subscribers, the company said.

With the change, Snapchat users will have a five-minute window to rephrase their message, fix typos or otherwise edit their chats. Messages that have been edited will have a label indicating the text has been changed. The company didn’t say when the feature might be available to more of its users, but the company often brings sought after features to its subscription service first. Snap announced last week that Snapchat+, which costs $3.99 a month, had reached 9 million subscribers.

The app is also adding several non-exclusive features, including updated emoji reactions for chats, the ability to use the My AI assistant to set reminders and AI-generated outfits for Bitmoji. Snap also showed off a new AI lens that transforms users’ selfies into 1990’s-themed snapshots (just don’t look too closely at the wireless headphones appearing in many of the images.)

Facebook’s News Feed algorithm has long been at the center of debates about some of Meta’s biggest problems. It’s also been a near constant source of complaints from users. But, if a newly filed lawsuit is successful, Facebook users may be able to use the social network with a vastly different feed. The Knight First Amendment Institute at Columbia University is suing Meta on behalf of a researcher who wants to release a browser extension that would allow people to “effectively turn off” their algorithmic feeds.

The extension was created by Ethan Zuckerman, a researcher and professor at the University of Massachusetts Amherst. He argues that Facebook users would be better off with more control over their feeds. “The tool, called Unfollow Everything 2.0, would allow users to unfollow their friends, groups, and pages, and, in doing so, to effectively turn off their newsfeed—the endless scroll of posts that users see when they log into Facebook,” the lawsuit explains. “Users who download the tool would be free to use the platform without the feed, or to curate the feed by refollowing only those friends and groups whose posts they really want to see.” (Meta officially renamed the News Feed to “Feed” in 2022.)

Zuckerman isn’t the first to come up with such a tool. He was inspired by a similar project, also called “Unfollow Everything,” from 2021. Facebook sued the U.K man who created that extension and permanently disabled his account. Zuckerman is trying to avoid a similar fate with his lawsuit. The suit, filed in San Francisco federal court Wednesday, asks the court “to recognize that Section 230 protects the development of tools designed to empower people to better control their social media experiences.”

The case could be a novel test of Section 230 of the 1996 Communications Decency Act, which is mostly known as the law that shields online platforms from legal liability for the actions of their users. But unlike recent Supreme Court cases involving the statute, Zuckerman’s case “relies on a separate provision protecting the developers of third-party tools that allow people to curate what they see online, including by blocking content they consider objectionable.”

A spokesperson for Meta declined to comment on the lawsuit. The company has a history of heavy-handed tactics when it comes to independent researchers. In addition to shutting down the earlier version “Unfollow Everything,” the company disabled the Facebook accounts of a group of NYU researchers attempting to study political ad targeting in 2021. Those types of tactics have led to some researchers pursuing “data donation” programs, which recruit volunteers to “donate” their own browsing data for academic studies.

If released, Zuckerman’s browser extension would also have a data donation component, allowing users to opt-in to sharing “anonymized data about their Facebook usage.” The data would then be used for research into the effects of Facebook’s feed algorithm.

It turns out the lay of today’s AI landscape can be traced back to — what do you know — fear, jealousy and intense capitalist ambition. Emails revealed in the Department of Justice’s antitrust case against Google, first reported by Business Insider, show Microsoft executives expressing alarm and envy over Google’s AI lead. That spurred an urgency that led to the Windows maker’s initial billion-dollar investment in its now-indispensable partner, OpenAI.

In a heavily redacted 2019 email thread titled “Thoughts on OpenAI,” Microsoft CEO Satya Nadella forwards a lengthy message from CTO Kevin Scott to CFO Amy Hood. “Very good email that explains, why I want us to do this ... and also why we will then ensure our infra folks execute,” Nadella wrote.

Scott wrote that he was “very, very worried” about Google’s rapidly growing AI capabilities. He says he initially dismissed the company’s “game-playing stunts,” likely referring to Google’s AlphaGo models. One of them beat Go world champion Ke Jie in 2017, a remarkable feat at the time. (Google’s later models surpassed that one, dropping the need for human training altogether.)

But Scott says brushing off Google’s game-playing progress “was a mistake.” “When they took all of the infrastructure that they had built to build [natural language] models that we couldn’t easily replicate, I started to take things more seriously,” Scott wrote. “And as I dug in to try to understand where all of the capability gaps were between Google and us for model training, I got very, very worried.”

Microsoft

Scott recounts how Microsoft struggled to copy Google’s BERT-large, an AI model that deciphers the meaning and context of words in a sentence. Scott pinned the blame on infrastructure leaps its rival had made — and that Microsoft hadn’t.

“Turns out, just replicating BERT-large wasn’t easy to do for us. Even though we had the template for the model, it took us ~6 months to get the model trained because our infrastructure wasn’t up to the task,” the Microsoft CTO wrote. “Google had BERT for at least six months prior to that, so in the time that it took us to hack together the capability to train a 340M parameter model, they had a year to figure out how to get it into production and to move on to larger scale, more interesting models.”

He also admired and envied Google’s Gmail auto-complete capabilities, saying it was “getting scarily good.” He commented that Microsoft was “multiple years behind the competition in terms of [machine learning] scale.” He commented on the “interesting” growth of OpenAI, DeepMind and Google Brain.

Scott touted Microsoft’s “very smart” people on its machine-learning teams but said their ambitions were curbed. “But the core deep learning teams within each of these bigger teams are very small, and their ambitions have also been constrained, which means that even as we start to feed them resources, they still have to go through a learning process to scale up,” Scott wrote. “And we are multiple years behind the competition in terms of ML scale.”

After prompting Hood that Scott’s concerns were “why I want us to do this,” meaning invest in OpenAI, the company made good on its CEO’s wishes. Microsoft invested a billion dollars in the Sam Altman-led startup in 2019, and the rest is a rapidly changing history. (It’s now invested $13 billion.) It’s a technology that does some incredible things but threatens to gut the labor market and give propagandists their most powerful tools to date in what was already an age of rampant disinformation.

Starfield, the excellent Digipick puzzle game surrounded by 80 hours of sci-fi mediocrity, is getting a performance update on Xbox Series X that unlocks frame rates above 30 fps. Starfield's May update adds the option to target 30 fps, 40 fps, 60 fps or an uncapped frame rate — for displays that support VRR running at 120hz. Displays without VRR will have the choice of 30 fps or 60 fps. 

The May update also includes the ability to prioritize visuals or performance at each frame rate target. Visuals mode means the game will do its best to maintain a high resolution and full detail in lighting, special effects and NPCs, while performance lowers the resolution and clarity of those same details. Of course, both modes adjust the game's base resolution alongside heavy on-screen action. 

Bethesda recommends performance mode when playing at 60 fps and above. For Xbox Series X players with 120hz VRR displays, Starfield's settings now default to 40 fps, prioritizing visuals.

The May 1 display updates bring the Xbox Series X version of Starfield closer to its PC counterpart in terms of customization options. The Xbox Series S edition remains capped at 30 fps. This is the version I played when I reviewed Starfield last year, and while a frame rate upgrade won't make the game less bland, its combat scenes would definitely benefit from a boost to 40 fps, at least. It's a shame that the most popular Xbox Series console isn't seeing any frame rate love in today's update.

Additionally in the May update, Starfield's surface maps have been overhauled in order to increase legibility on all platforms. The new design shows top-down 3D images of terrain, buildings, and objects like trees and rocks, which makes a lot of sense for, you know, a map. The original surface map tries to make landscapes out of white dots on a bright blue background, so this is a welcome improvement. The update also allows players to customize their difficulty options on the ground and in ship battles, and it adds navigation markers to the environment when walking around a planet.

This is Bethesda's fourth and largest Starfield update since the game came out in September 2023. It's all scheduled to go live by May 15.