"The iMessage connection software that powers Beeper Mini and Beeper Cloud is now 100% open source," Beeper announced late this week. " Anyone who wants can use it or continue development."

But while Beeper says it's done trying to bring iMessage to Android, CNET reports that the whole battle was "deeply tied" to Apple's ongoing strategy to control the mobile market: The tide seems to be changing, however: Apple said last month it would be opening up its Messages app (likely due to European regulation) to work with the newer, more feature-rich texting protocol called RCS. This hopefully will lead to a more modern and secure messaging experience when texting between an iPhone and an Android phone, and lead away from the aging SMS and MMS standards. Unfortunately, green bubbles will continue to persist even if there might be little to no functional difference. While third-party apps like Nothing Chats attempted and ultimately failed to bring iMessage to Android, Apple will likely never release the app on Google's mobile operating system.

Until RCS is fully adopted, companies are creating services to allow access to iMessage via Android phones. Apple, for its part, has been quick to block apps like Beeper Mini, citing security concerns. This, however, is raising eyebrows from lawmakers regarding competition in the messaging space and Apple's tight control over the market...

Beeper in a December 21 blog post told users to grab a jailbroken iPhone and install a free Beeper tool that'll generate iMessage registration codes to keep the service operational. It's such a roundabout and potentially expensive way of trying to get iMessage on Android that it likely won't be worth it for most people. For those not willing to go out and jailbreak an iPhone, Beeper said in a now-deleted blog post that it would allow people to rent a jailbroken unit for a small monthly fee starting next year.

Longtime Slashdot reader theodp writes: Way back in September 2012, Microsoft President Brad Smith discussed the idea of "producing a crisis" to advance Microsoft's "two-pronged" National Talent Strategy to increase K-12 CS education and the number of H-1B visas. Not long thereafter, the tech-backed nonprofit Code.org (which promotes and provides K-12 CS education and is led by Smith's next-door neighbor) and Mark Zuckerberg's FWD.us PAC (which lobbied for H-1B reform) were born, with Smith on board both. Over the past 10+ years, Smith has played a key role in establishing Code.org's influence in the new K-12 CS education "grassroots" movement, including getting buy-in from three Presidential administrations -- Obama, Trump, and Biden -- as well as the U.S. Dept. of Education and the nation's Governors.

But after recent updates, Code.org's Leadership page now indicates that Smith has quietly left Code.org's Board of Directors and thanks him for his past help and advice. Since November (when archive.org indicates Smith's photo was yanked from Code.org's Leadership page), Smith has been in the news in conjunction with Microsoft's relationship with another Microsoft-bankrolled nonprofit, OpenAI, which has come under scrutiny by the Feds and in the UK. Smith, who noted he and Microsoft helped OpenAI and CEO Sam Altman craft messaging ahead of a White House meeting, announced in a Dec. 8th tweet that Microsoft will be getting a non-voting OpenAI Board seat in connection with Altman's return to power (who that non-voting Microsoft OpenAI board member will be has not been announced).

OpenAI, Microsoft, and Code.org teamed up in December to provide K-12 CS+AI tutorials for this December's AI-themed Hour of Code (the trio has also partnered with Amazon and Google on the Code.org-led TeachAI initiative). And while Smith has left Code.org's Board, Microsoft's influence there will live on as Microsoft CTO Kevin Scott -- credited for forging Microsoft's OpenAI partnership -- remains a Code.org Board member together with execs from other Code.org Platinum Supporters ($3+ million in past 2 years) Google and Amazon.

For years, Amazon Web Services' annual Las Vegas trade show functioned as an infomercial for its cloud computing platform, rarely mentioning the competition. The pitch was so successful that AWS pulls in $90 billion per year. Then generative AI emerged, with Microsoft and Google baking it into products their cloud units sell. Suddenly, AWS faced startups building businesses on rivals' AI-powered platforms. So at AWS's 2023 event, AI was ubiquitous -- in presentations, launches, partnerships. AWS announced more models powering AI services and its largest-ever tech investment, $4 billion in generative AI startup Anthropic. AWS aims to show that, despite stiffening competition, it remains the leader in cloud computing. From a report: If Amazon had been caught off guard by the dawn of the generative AI age, here was evidence of a massive, companywide effort to catch up. "This is what last place looks like," analysts with Sanford C. Bernstein quipped in a research note. In the short term, AWS is going to be fine. Slowing sales growth aside, Amazon's servers remain the default starting point for companies looking to modernize old infrastructure or do much of anything online. And though generative AI makes for an impressive demo, the technology is error-prone and expensive. For most companies, it's an experiment, not a necessity.

Still, "to remain relevant," AWS needs to have a handle on generative AI, according to JB McGinnis, a principal at Deloitte who helps companies use AWS. "If they're not competing, they might lose the cloud game, too." Late in the week of the conference, Amazon invited thousands of attendees with ties to startups to the Las Vegas Raiders' stadium, which it had rented out for the occasion, plying them with drinks and AWS swag and giant versions of bar games. Before a panel discussion on artificial intelligence, Swami Sivasubramanian, the Amazon executive in charge of the company's AI services, declared 2023 the year of generative AI. Nearby, an AWS product leader walked up to the founder of a tiny startup, introduced himself, and asked what Amazon could do better. This was a humbled AWS, one that has to fight for business.

Google's Safety Check feature for Chrome, which, among other things, checks the internet to see if any of your saved passwords have been compromised, will now "run automatically in the background" on desktop, the company said in a blog post on Thursday. From a report: The constant checks could mean that you're alerted about a password that you should change sooner than you would have before. Safety Check also watches for bad extensions or site permissions you need to look at, and you can act on Safety Check alerts from Chrome's three-dot menu. In addition, Google says that Safety Check can revoke a site's permissions if you haven't visited it in a while. Google also announced an upcoming feature for Chrome's tab groups, also on desktop: Chrome will let you save tab groups so that you can use those groups across devices, which might be handy when moving between a PC at home and a laptop when traveling. Google says this feature will roll out "over the next few weeks."

India is testing AI to build climate models to improve weather forecasting as torrential rains, floods and droughts proliferate across the vast country, a top weather official said. From a report: Global warming has triggered more intense clashes of weather systems in India in recent years, increasing extreme weather events, which the independent Centre for Science and Environment estimates have killed nearly 3,000 people this year. Weather agencies around the world are focussing on AI, which can bring down cost and improve speed, and which Britain's Met Office says could "revolutionise" weather forecasting, with a recent Google-funded model found to have outperformed conventional methods.

Accurate weather forecasting is particularly crucial in India, a country of 1.4 billion people, many impoverished, and the world's second-largest producer of rice, wheat and sugar. The India Meteorological Department (IMD) provides forecasts based on mathematical models using supercomputers. Using AI with an expanded observation network could help generate higher-quality forecast data at lower cost. The department expects the AI-based climate models and advisories it is developing to help improve forecasts, K.S. Hosalikar, head of climate research and services at IMD, told Reuters.

An anonymous reader quotes a report from Ars Technica: A report at The Information says that AI might already be taking people's jobs at Google. The report cites people briefed on the plans and says Google intends to "consolidate staff, including through possible layoffs, by reassigning employees at its large customer sales unit who oversee relationships with major advertisers." According to the report, the jobs are being vacated because Google's new AI tools have automated them. The report says a future restructuring was apparently already announced at a department-wide Google Ads meeting last week.

Google announced a "new era of AI-powered ads" in May, featuring a "natural-language conversational experience within Google Ads, designed to jump-start campaign creation and simplify Search ads." Google said its new AI could scan your website and "generate relevant and effective keywords, headlines, descriptions, images, and other assets," making the Google Ads chatbot one part designer and one part sales expert. [...] The report also notes another benefit of making AI do this work: "Because these tools don't require much employee attention, they carry relatively few expenses, so the ad revenue carries a high-profit margin."

The Information report says, "A growing number of advertisers have adopted PMax since [launch], eliminating the need for some employees who specialized in selling ads for a particular Google service, like search, working together to design ad campaigns for big customers." [Google's Performance Max, or "PMax," is a Google ad tool that can help advertisers actually make ad content and determine the best places for it -- YouTube, Search, Gmail, etc.] According to the report, as of a year ago, Google had about 13,500 people devoted to this kind of sales work, a huge chunk of the 30,000-strong ad division. These 13,500 people aren't necessarily all going to be affected, and those who are won't necessarily be laid off -- they could be reassigned to other areas in Google. We should know the scale of Google Ad's big re-org soon. The report says, "Some employees expect the changes to be announced next month."

The next version of Android could give you an estimate of your battery's remaining capacity, which naturally degrades over time. "Android 14 laid the initial groundwork for the OS to track battery health information, but Android 15 could actually bring that information in front of users," reports Android Authority. It could also tell you whether your device's battery has been replaced. From the report: The manufacture date and cycle count aren't the only battery-related statistics that Android 14 exposes to apps through new APIs, though. Other battery health details like the date of first use, charging policy, charging status, and state of health are also available. The state of health is particularly interesting because it's an estimate of the battery's current full charge capacity, expressed as a percentage relative to the battery's rated capacity. For example, if your Pixel 8 battery's state of health is measured at 90%, that means its remaining full charge capacity is estimated to be about 4118mAh (compared to the rated 4575mAh).

The Settings app currently doesn't show the battery state of health, but that's set to change in the future, as the latest version of the Settings Services app (an extension to the Settings app on Pixel and other devices) found within Android 14 QPR2 Beta 2 has a new "battery health" page that is set to show the state of health. [...] Strings within the APK suggest this page will show you the "estimated percentage of charge the battery can currently hold compared to when it was new" (i.e. the state of health) before and after "recalibration" of the battery. We don't have the exact details on what "recalibration" entails, but given that one string suggests the "process may take a few weeks," we're guessing that it's simply the system collecting data over a longer period to provide a more accurate estimate of the battery capacity. Meanwhile, the "initial battery health values" are "based on lab results" and hence "may vary from your actual battery state."

[...] We also learned that the Settings app itself will surface "tips" to the user when either the battery capacity is degraded or can't be detected, so the user doesn't have to manually check the "battery health" page. Lastly, we learned that Google is working on exposing more battery-related information to the OS, such as the part status and the serial number. [...] At the very least, we do know that Android will support reading the battery's part status and serial number, provided the battery exposes that information to the OS, and the vendor implements the new version of the Android health HAL. The health HAL is the software responsible for bridging the gap between the OS APIs that read battery/charging information (i.e. everything we talked about before) with the software that controls the battery/charging chips. Version 2.0 of the health HAL needs to be implemented to support all the new Android 14 battery health APIs like state of health, which is why so few devices support that right now.

An anonymous reader quotes a report from ZDNet: Long before Facebook existed, or even before the Internet, there was Usenet. Usenet was the first social network. Now, with Google Groups abandoning Usenet, this oldest of all social networks is doomed to disappear. Some might say it's well past time. As Google declared, "Over the last several years, legitimate activity in text-based Usenet groups has declined significantly because users have moved to more modern technologies and formats such as social media and web-based forums. Much of the content being disseminated via Usenet today is binary (non-text) file sharing, which Google Groups does not support, as well as spam." True, these days, Usenet's content is almost entirely spam, but in its day, Usenet was everything that Twitter and Reddit would become and more.

In 1979, Duke University computer science graduate students Tom Truscott and Jim Ellis conceived of a network of shared messages under various topics. These messages, also known as articles or posts, were submitted to topic categories, which became known as newsgroups. Within those groups, messages were bound together in threads and sub-threads. [...] In 1980, Truscott and Ellis, using the Unix to Unix Copy Protocol (UUCP), hooked up with the University of North Carolina to form the first Usenet nodes. From there, it would rapidly spread over the pre-Internet ARPANet and other early networks. These messages would be stored and retrieved from news servers. These would "peer" to each other so that messages to a newsgroup would be shared from server to server and to user to user so that within hours, your messages would reach the entire networked world. Usenet would evolve its own network protocol, Network News Transfer Protocol (NNTP), to speed the transfer of these messages. Today, the social network Mastodon uses a similar approach with the ActivityPub protocol, while other social networks, such as Threads, are exploring using ActivityPub to connect with Mastodon and the other social networks that support ActivityPub. As the saying goes, everything old is new again.

[...] Usenet was never an organized social network. Each server owner could -- and did -- set its own rules. Mind you, there was some organization to begin with. The first 'mainstream' Usenet groups, comp, misc, news, rec, soc, and sci hierarchies, were widely accepted and disseminated until 1987. Then, faced with a flood of new groups, a new naming plan emerged in what was called the Great Renaming. This led to a lot of disputes and the creation of the talk hierarchy. This and the first six became known as the Big Seven. Then the alt groups emerged as a free speech protest. Afterward, fewer Usenet sites made it possible to access all the newsgroups. Instead, maintainers and users would have to decide which one they'd support. Over the years, Usenet began to decline as discussions were replaced both by spam and flame wars. Group discussions were also overwhelmed by flame wars. "If, going forward, you want to keep an eye on Usenet -- things could change, miracles can happen -- you'll need to get an account from a Usenet provider," writes ZDNet's Steven Vaughan-Nichols. "I favor Eternal September, which offers free access to the discussion Usenet groups; NewsHosting, $9.99 a month with access to all the Usenet groups; EasyNews, $9.98 a month with fast downloads, and a good search engine; and Eweka, 9.50 Euros a month and EU only servers."

"You'll also need a Usenet client. One popular free one is Mozilla's Thunderbird E-Mail client, which doubles as a Usenet client. EasyNews also offers a client as part of its service. If you're all about downloading files, check out SABnzbd."

samleecole writes: The LAION-5B machine learning dataset used by Google, Stable Diffusion, and other major AI products has been removed by the organization that created it after a Stanford study found that it contained 3,226 suspected instances of child sexual abuse material, 1,008 of which were externally validated.

LAION told 404 Media on Tuesday that out of "an abundance of caution," it was taking down its datasets temporarily "to ensure they are safe before republishing them." According to a new study by the Stanford Internet Observatory shared with 404 Media ahead of publication, the researchers found the suspected instances of CSAM through a combination of perceptual and cryptographic hash-based detection and analysis of the images themselves.

An anonymous reader writes: Digital mapping specialist TomTom said on Tuesday it has partnered with tech giant Microsoft to create an artificial intelligence (AI)-powered conversational assistant for vehicles. The assistant will allow users to "converse naturally with their vehicles" and enable voice interaction with infotainment, location search, and vehicle command systems, the company said.

TomTom, which competes with Google Maps and the world's biggest mapping platform HERE, used various Microsoft services like its Azure OpenAI Service to create the voice assistant. The Microsoft Azure OpenAI Service allows enterprises to leverage ChatGPT maker OpenAI's large language models (LLM). The voice assistant can be integrated into other automotive infotainment systems and is also built into TomTom's Digital Cockpit, an open, modular in-vehicle infotainment platform, the Dutch map maker said. The company began working with Microsoft in 2016, when it first started powering Azure Maps location services.

An anonymous reader quotes a report from the BBC: A bill that mandates tech giants pay news outlets for their content has come into effect in Canada amid an ongoing dispute with Facebook and Instagram owner Meta over the law. Some have hailed it as a game-changer that sets out a permanent framework that will see a steady drip of funds from wealthy tech companies to Canada's struggling journalism industry. But it has also been met with resistance by Google and Meta -- the only two companies big enough to be encompassed by the law. In response, over the summer, Meta blocked access to news on Facebook and Instagram for Canadians. Google looked set to follow, but after months of talks, the federal government was able to negotiate a deal with the search giant as the company has agreed to pay Canadian news outlets $75 million annually.

No such agreement appears to be on the horizon with Meta, which has called the law "fundamentally flawed." If Meta is refusing to budge, so is the government. "We will continue to push Meta, that makes billions of dollars in profits, even though it is refusing to invest in the journalistic rigor and stability of the media," Prime Minister Justin Trudeau told reporters on Friday. According to a study by the Media Ecosystem Observatory, the views of Canadian news on Facebook dropped 90% after the company blocked access to news on the platform. Local news outlets have been hit particularly hard.

"The loss of journalism on Meta platforms represents a significant decline in the resiliency of the Canadian media ecosystem," said Taylor Owen, a researcher at McGill and the co-author of the study. He believes it also hurts Meta's brand in the long run, pointing to the fact that the Canada's federal government, as well as that of British Columbia, other municipalities and a handful of large Canadian corporations, have all pulled their advertising off Facebook and Instagram in retaliation.

Alphabet will pay $700 million and alter its Google Play policies to settle claims that the app store unlawfully dominates the Android mobile applications market, resolving antitrust complaints brought by attorneys general of about three dozen states and consumers. From a report: The deal disclosed in a court filing late Monday calls for tweaks to Google Play policies designed to reduce barriers to competition in the markets for app distribution and payment processing. The lawsuits that were grouped together in federal court in California had threatened billions of dollars in revenue generated by the sale and distribution of apps through Google Play. Google will also make a series of changes to its business practices as part of the settlement. In a blog post, the Android-maker said: Streamlining sideloading while prioritizing security: Unlike on iOS, Android users have the option to sideload apps, meaning they can download directly from a developer's website without going through an app store like Google Play. While we maintain it is critical to our safety efforts to inform users that sideloading on mobile could come with unique risks, as part of our settlement we will be further simplifying the sideloading process and updating the language that informs users about these potential risks of downloading apps directly from the web for the first time.
Expanding user choice billing to more people: App and game developers will be able to implement an alternative billing option alongside Google Play's billing system for their U.S. users who can then choose which option to use when making in-app purchases. We have been piloting user choice billing in the U.S. for over a year and will now expand this option further.
Expanding open communication on pricing: We have always given developers more ways to interact with their customers than iOS and other operating systems. For example, Google Play allows developers to communicate freely with their customers outside the app about subscription offers or lower-cost options available on a rival app store or the developer's website. This openness has spurred competition and benefited consumers and developers. As part of user choice billing, which we're expanding with today's settlement announcement, developers are also able to show different pricing options within the app when a user makes a digital purchase.

Ron Amadeo reports via Ars Technica: Stadia might be dead, but the controllers for Google's cloud-based gaming platform are still out there. With the service permanently offline, the proprietary Stadia Controller threatened to fill up landfills until Google devised a plan to convert them to generic Bluetooth devices that can work on almost anything. The app to open up the controller to other devices is a web service, which previously had a shutdown date of December 2023. That apparently isn't enough time to convert all these controllers, so the Stadia Controller Salvage operation will run for a whole additional year. X (formerly Twitter) user Wario64 was the first to spot the announcement, which says the online tool will continue running until December 31, 2024.

Travel website Expedia wants to get people to start their travel search on its site with AI instead of using an external search engine. From a report: Expedia already uses AI for some customer service features and to help property owners describe their homes and hotels. The company hopes in the future that AI will help it recommend travel destinations to customers based on previous trips and bring more direct traffic to its site. It's a long-term plan to shift the balance of power on the web -- albeit one that's still in its earliest stages for the company.

Rajesh Naidu, chief architect and head of data management at Expedia, says the goal is to get users started on their trips in one place. Expedia hopes to produce recommendations trained with its library of flight and hotel information and informed by users' travel preferences. "By being able to train large language models on our data, this rich 70 petabytes' worth of data we've gathered over the years, we can eventually recommend places to go and stay and do and continue to refine and personalize that," Naidu tells The Verge in an interview. According to Naidu, when people plan trips, they often start by going to a search engine to look for a destination. Only then do they visit services like Expedia to start booking travel and accommodation. There's nothing inherently wrong with going to Google and typing "best vacation that isn't cold and not that far from New York," but Naidu believes there's value in streamlining the travel planning process even more.

Forbes recently published this article by author/speaker Nina Xiang, who reports that Huawei is pushing forward with "an amibitious plan to dethrone Android." Hundreds of technical experts from many of China's biggest state-owned and private companies, including the Industrial and Commercial Bank of China (ICBC), China Telecom, Meituan, and Baidu, all gathered in Beijing last month. The purpose behind the meeting was for their staff to receive training so they could be certified as developers on Huawei's Harmony Operation System (OS).

While most observers were looking the other way, Huawei has been quietly building an independent Chinese operating system that isn't subject to U.S. sanctions. In the four years after the telecom giant was banned from using Google apps, the Shenzhen-based company has been making significant strides toward achieving its long-term goal: To dethrone Android and make its HarmonyOS the default operating system in China.

Looking at the data for smartphone sales in China shows that HarmonyOS had the third-largest share with 10% in the second quarter of 2023, thanks to a strong resurgence in sales of Huawei smartphones. Although it's still well below Android's dominant 72%, it's not far from iOS's 17%... Huawei already says more than 700 million devices (including phones, smart devices, computers, and others) were equipped with HarmonyOS as of August this year, with over 2.2 million developers actively building within the ecosystem...

A key moment will come next year, when Huawei says HarmonyOS will no longer be compatible with Android apps.

Long-time Slashdot reader theodp writes: Earlier this month, Amazon came under fire as the Los Angeles Times reported on a leaked confidential document that "reveals an extensive public relations strategy by Amazon to donate to community groups, school districts, institutions and charities" to advance the company's business objectives. "We will not fund organizations that have positioned themselves antagonistically toward our interests," explained Amazon officials of the decision to cut off donations to the Cheech Marin Center for Chicano Art and Culture after it ran an exhibit ("Burn Them All Down") that the artist called a commentary on how public officials were not listening to community concerns about the growing number of Amazon warehouses in Southern California's Inland Empire neighborhoods...

Interestingly on the same day the Los Angeles Times was sounding the alarm on Amazon philanthropy, the White House and National Science Foundation (NSF) held a White House-hosted event on K-12 AI education. There it was announced that the Amazon-backed nonprofit Computer Science Teachers Association (CSTA) will develop new K-12 computer science standards that incorporate AI into foundational computer science education with support from the NSF, Amazon, Google, and Microsoft. CSTA separately announced it had received a $1.5 million donation from Amazon to "support efforts to update the CSTA K-12 Computer Science Standards to reflect the rapid advancements in technologies like artificial intelligence (AI)," adding that the CSTA standards — which CSTA credited Microsoft Philanthropies for helping to advance — "serve as a model for CS teaching and learning across grades K-12" in 42 states.

The announcements, the White House noted, came during Computer Science Education Week, the signature event of which is Amazon, Google, and Microsoft-backed Code.org's Hour of Code (which was AI-themed this year), for which Amazon, Google, and Microsoft — not teachers — provided the event's signature tutorials used by the nation's K-12 students. Amazon, Google, and Microsoft are also advisors to Code.org's TeachAI initiative, which was launched in May "to provide thought leadership to guide governments and educational leaders in aligning education with the needs of an increasingly AI-driven world and connecting the discussion of teaching with AI to teaching about AI and computer science."

Earlier this week Google Maps stopped storing user location histories in the cloud. But why did Google make this move? Bloomberg reports that it was "so that the company no longer has access to users' individual location histories, cutting off its ability to respond to law enforcement warrants that ask for data on everyone who was in the vicinity of a crime." The company said Thursday that for users who have it enabled, location data will soon be saved directly on users' devices, blocking Google from being able to see it, and, by extension, blocking law enforcement from being able to demand that information from Google. "Your location information is personal," said Marlo McGriff, director of product for Google Maps, in the blog post. "We're committed to keeping it safe, private and in your control."

The change comes three months after a Bloomberg Businessweek investigation that found police across the US were increasingly using warrants to obtain location and search data from Google, even for nonviolent cases, and even for people who had nothing to do with the crime. "It's well past time," said Jennifer Lynch, the general counsel at the Electronic Frontier Foundation, a San Francisco-based nonprofit that defends digital civil liberties. "We've been calling on Google to make these changes for years, and I think it's fantastic for Google users, because it means that they can take advantage of features like location history without having to fear that the police will get access to all of that data."

Google said it would roll out the changes gradually through the next year on its own Android and Apple Inc.'s iOS mobile operating systems, and that users will receive a notification when the update comes to their account. The company won't be able to respond to new geofence warrants once the update is complete, including for people who choose to save encrypted backups of their location data to the cloud.
The EFF general counsel also pointed out to Bloomberg that "nobody else has been storing and collecting data in the same way as Google." (Apple, for example, is technically unable to provide the same data to police.)

An anonymous reader quotes a report from : Microsoft Research, the blue sky division of the software giant, [...] announced the release of its Phi-2 small language model (SML), a text-to-text AI program that is "small enough to run on a laptop or mobile device," according to a post on X. At the same time, Phi-2 with its 2.7 billion parameters (connections between artificial neurons) boasts performance that is comparable to other, much larger models including Meta's Llama 2-7B with its 7 billion parameters and even Mistral-7B, another 7 billion parameter model.

Microsoft researchers also noted in their blog post on the Phi-2 release that it outperforms Google's brand new Gemini Nano 2 model despite it having half a billion more parameters, and delivers less "toxicity" and bias in its responses than Llama 2. Microsoft also couldn't resist taking a little dig at Google's now much-criticized, staged demo video for Gemini in which it showed off how its forthcoming largest and most powerful new AI model, Gemini Ultra, was able to solve fairly complex physics problems and even correct students' mistakes on them. As it turned out, even though it is likely a fraction of the size of Gemini Ultra, Phi-2 also was able to correctly answer the question and correct the student using the same prompts.

However, despite these encouraging findings, there is a big limitation with Phi-2, at least for the time being: it is licensed only for "research purposes only," not commercial usage, under a custom Microsoft Research License, which further states Phi-2 may only be used for "non-commercial, non-revenue generating, research purposes." So, businesses looking to build products atop it are out of luck.

An anonymous reader quotes a report from MIT Technology Review: In a paper published in Nature today, the researchers say it is the first time a large language model has been used to discover a solution to a long-standing scientific puzzle -- producing verifiable and valuable new information that did not previously exist. "It's not in the training data -- it wasn't even known," says coauthor Pushmeet Kohli, vice president of research at Google DeepMind. Large language models have a reputation for making things up, not for providing new facts. Google DeepMind's new tool, called FunSearch, could change that. It shows that they can indeed make discoveries -- if they are coaxed just so, and if you throw out the majority of what they come up with.

FunSearch (so called because it searches for mathematical functions, not because it's fun) continues a streak of discoveries in fundamental math and computer science that DeepMind has made using AI. First Alpha Tensor found a way to speed up a calculation at the heart of many different kinds of code, beating a 50-year record. Then AlphaDev found ways to make key algorithms used trillions of times a day run faster. Yet those tools did not use large language models. Built on top of DeepMind's game-playing AI AlphaZero, both solved math problems by treating them as if they were puzzles in Go or chess. The trouble is that they are stuck in their lanes, says Bernardino Romera-Paredes, a researcher at the company who worked on both AlphaTensor and FunSearch: "AlphaTensor is great at matrix multiplication, but basically nothing else." FunSearch takes a different tack. It combines a large language model called Codey, a version of Google's PaLM 2 that isfine-tuned on computer code, with other systems that reject incorrect or nonsensical answers and plug good ones back in.

The researchers started by sketching out the problem they wanted to solve in Python, a popular programming language. But they left out the lines in the program that would specify how to solve it. That is where FunSearch comes in. It gets Codey to fill in the blanks -- in effect, to suggest code that will solve the problem. A second algorithm then checks and scores what Codey comes up with. The best suggestions -- even if not yet correct -- are saved and given back to Codey, which tries to complete the program again. After a couple of million suggestions and a few dozen repetitions of the overall process -- which took a few days -- FunSearch was able to come up with code that produced a correct and previously unknown solution to the cap set problem, which involves finding the largest size of a certain type of set. Imagine plotting dots on graph paper. [...] To test its versatility, the researchers used FunSearch to approach another hard problem in math: the bin packing problem, which involves trying to pack items into as few bins as possible. This is important for a range of applications in computer science, from data center management to e-commerce. FunSearch came up with a way to solve it that's faster than human-devised ones.

Emma Roth reports via The Verge: Google is releasing a tool to help users diagnose problems with their Pixel phones. Users can launch the app by entering #*#7287#*# on the dial pad, allowing them to check whether their phone is working correctly before or after a repair. There are a bunch of diagnostic tools available within the app. While users can run a full diagnostic test to detect issues across the entire device, there are also options to run individual tests for physical defects and problems affecting the phone's display, sensors, and connectivity. The new tool is available on all Pixel phones in English.

Aside from the diagnostics app, Google is introducing redesigned repair manuals the company says are "easier for technicians and DIYers to use." Users can download repair manuals from Google's website in English and French, but they're only available for the Pixel Fold, Pixel 8, and Pixel 8 Pro for now. Google says it will add repair manuals for previous and future devices "in the coming months." There's also a new Repair Mode Google introduced earlier this month, which Pixel Phone owners can toggle on when their device is getting repaired. The feature is meant to protect users' private information while their phone is in the hands of a technician [...].