AI – Page 26

Meta takes us a step closer to Star Trek’s universal translator

AI, Computer science, human speech, Science, text processing, translation / 9u50fv / January 15, 2025

The computer science behind translating speech from 100 source languages.

In 2023, AI researchers at Meta interviewed 34 native Spanish and Mandarin speakers who lived in the US but didn’t speak English. The goal was to find out what people who constantly rely on translation in their day-to-day activities expect from an AI translation tool. What those participants wanted was basically a Star Trek universal translator or the Babel Fish from the Hitchhiker’s Guide to the Galaxy: an AI that could not only translate speech to speech in real time across multiple languages, but also preserve their voice, tone, mannerisms, and emotions. So, Meta assembled a team of over 50 people and got busy building it.

What this team came up with was a next-gen translation system called Seamless. The first building block of this system is described in Wednesday’s issue of Nature; it can translate speech among 36 different languages.

Language data problems

AI translation systems today are mostly focused on text, because huge amounts of text are available in a wide range of languages thanks to digitization and the Internet. Institutions like the United Nations or European Parliament routinely translate all their proceedings into the languages of all their member states, which means there are enormous databases comprising aligned documents prepared by professional human translators. You just needed to feed those huge, aligned text corpora into neural nets (or hidden Markov models before neural nets became all the rage) and you ended up with a reasonably good machine translation system. But there were two problems with that.

The first issue was those databases comprised formal documents, which made the AI translators default to the same boring legalese in the target language even if you tried to translate comedy. The second problem was speech—none of this included audio data.

The problem of language formality was mostly solved by including less formal sources like books, Wikipedia, and similar material in AI training databases. The scarcity of aligned audio data, however, remained. Both issues were at least theoretically manageable in high-resource languages like English or Spanish, but they got dramatically worse in low-resource languages like Icelandic or Zulu.

As a result, the AI translators we have today support an impressive number of languages in text, but things are complicated when it comes to translating speech. There are cascading systems that simply do this trick in stages. An utterance is first converted to text just as it would be in any dictation service. Then comes text-to-text translation, and finally the resulting text in the target language is synthesized into speech. Because errors accumulate at each of those stages, the performance you get this way is usually poor, and it doesn’t work in real time.

A few systems that can translate speech-to-speech directly do exist, but in most cases they only translate into English and not in the opposite way. Your foreign language interlocutor can say something to you in one of the languages supported by tools like Google’s AudioPaLM, and they will translate that to English speech, but you can’t have a conversation going both ways.

So, to pull off the Star Trek universal translator thing Meta’s interviewees dreamt about, the Seamless team started with sorting out the data scarcity problem. And they did it in a quite creative way.

Building a universal language

Warren Weaver, a mathematician and pioneer of machine translation, argued in 1949 that there might be a yet undiscovered universal language working as a common base of human communication. This common base of all our communication was exactly what the Seamless team went for in its search for data more than 70 years later. Weaver’s universal language turned out to be math—more precisely, multidimensional vectors.

Machines do not understand words as humans do. To make sense of them, they need to first turn them into sequences of numbers that represent their meaning. Those sequences of numbers are numerical vectors that are termed word embeddings. When you vectorize tens of millions of documents this way, you’ll end up with a huge multidimensional space where words with similar meaning that often go together, like “tea” and “coffee,” are placed close to each other. When you vectorize aligned text in two languages like those European Parliament proceedings, you end up with two separate vector spaces, and then you can run a neural net to learn how those two spaces map onto each other.

But the Meta team didn’t have those nicely aligned texts for all the languages they wanted to cover. So, they vectorized all texts in all languages as if they were just a single language and dumped them into one embedding space called SONAR (Sentence-level Multimodal and Language-Agnostic Representations). Once the text part was done, they went to speech data, which was vectorized using a popular W2v (word to vector) tool and added it to the same massive multilingual, multimodal space. Of course, each embedding carried metadata identifying its source language and whether it was text or speech before vectorization.

The team just used huge amounts of raw data—no fancy human labeling, no human-aligned translations. And then, the data mining magic happened.

SONAR embeddings represented entire sentences instead of single words. Part of the reason behind that was to control for differences between morphologically rich languages, where a single word may correspond to multiple words in morphologically simple languages. But the most important thing was that it ensured that sentences with similar meaning in multiple languages ended up close to each other in the vector space.

It was the same story with speech, too—a spoken sentence in one language was close to spoken sentences in other languages with similar meaning. It even worked between text and speech. So, the team simply assumed that embeddings in two different languages or two different modalities (speech or text) that are at a sufficiently close distance to each other are equivalent to the manually aligned texts of translated documents.

This produced huge amounts of automatically aligned data. The Seamless team suddenly got access to millions of aligned texts, even in low-resource languages, along with thousands of hours of transcribed audio. And they used all this data to train their next-gen translator.

Seamless translation

The automatically generated data set was augmented with human-curated texts and speech samples where possible and used to train multiple AI translation models. The largest one was called SEAMLESSM4T v2. It could translate speech to speech from 101 source languages into any of 36 output languages, and translate text to text. It would also work as an automatic speech recognition system in 96 languages, translate speech to text from 101 into 96 languages, and translate text to speech from 96 into 36 languages—all from a single unified model. It also outperformed state-of-the-art cascading systems by 8 percent in a speech-to-text and by 23 percent in a speech-to-speech translations based on the scores in Bilingual Evaluation Understudy (an algorithm commonly used to evaluate the quality of machine translation).

But it can now do even more than that. The Nature paper published by Meta’s Seamless ends at the SEAMLESSM4T models, but Nature has a long editorial process to ensure scientific accuracy. The paper published on January 15, 2025, was submitted in late November 2023. But in a quick search of the arXiv.org, a repository of not-yet-peer-reviewed papers, you can find the details of two other models that the Seamless team has already integrated on top of the SEAMLESSM4T: SeamlessStreaming and SeamlessExpressive, which take this AI even closer to making a Star Trek universal translator a reality.

SeamlessStreaming is meant to solve the translation latency problem. The baseline SEAMLESSM4T, despite all the bells and whistles, worked as a standard AI translation tool. You had to say what you wanted to say, push “translate,” and it spat out the translation. SeamlessStreaming was designed to take this experience a bit closer to what human simultaneous translator do—it translates what you’re saying as you speak in a streaming fashion. SeamlessExpressive, on the other hand, is aimed at preserving the way you express yourself in translations. When you whisper or say something in a cheerful manner or shout out with anger, SeamlessExpressive will encode the features of your voice, like tone, prosody, volume, tempo, and so on, and transfer those into the output speech in the target language.

Sadly, it still can’t do both at the same time; you can only choose to go for either streaming or expressivity, at least at the moment. Also, the expressivity variant is very limited in supported languages—it only works in English, Spanish, French, and German. But at least it’s online so you can go ahead and give it a spin.

Nature, 2025. DOI: 10.1038/s41586-024-08359-z

Jacek Krywko is a freelance science and technology writer who covers space exploration, artificial intelligence research, computer science, and all sorts of engineering wizardry.

Meta takes us a step closer to Star Trek’s universal translator Read More »

ChatGPT becomes more Siri-like with new scheduled tasks feature

agentic AI, AI, AI Agent, chatgpt, ChatGPT Tasks, openai / Mike M. / January 14, 2025

OpenAI is making ChatGPT work a little more like older digital assistants with a new feature called Tasks, as reported by TechCrunch and others.

Currently in beta, Tasks allows users to direct the chatbot to send reminders or to generate responses to specific prompts at certain times; recurring tasks are also supported.

The feature is available to Plus, Team, and Pro subscribers starting today, while free users don’t have access.

To create a task, users need to select “4o with scheduled tasks” from the model picker and then direct ChatGPT using the same kind of plain language text prompts that drive everything else it does. ChatGPT will sometimes suggest tasks, too, but they won’t go into effect unless the user approves them.

The user can then make changes to assigned tasks through the same chat conversation, or they can use a new Tasks section of the ChatGPT apps to manage all currently assigned items. There’s currently a 10-task limit.

When the time comes to perform an assigned task, the ChatGPT mobile or desktop app will send a notification on schedule.

This update can be seen as OpenAI’s first step into the agentic AI space, where applications built using deep learning can operate relatively independently within certain boundaries, either replacing or easing the day-to-day responsibilities of information workers.

ChatGPT becomes more Siri-like with new scheduled tasks feature Read More »

Amid a flurry of hype, Microsoft reorganizes entire dev team around AI

agentic AI, AI, copilot, deep learning, GitHub, llm, machine learning, microsoft, Satya Nadella, software development, Tech, visual studio / Shannon Garcia / January 14, 2025

Microsoft CEO Satya Nadella has announced a dramatic restructuring of the company’s engineering organization, which is pivoting the company’s focus to developing the tools that will underpin agentic AI.

Dubbed “CoreAI – Platform and Tools,” the new division rolls the existing AI platform team and the previous developer division (responsible for everything from .NET to Visual Studio) along with some other teams into one big group.

As for what this group will be doing specifically, it’s basically everything that’s mission-critical to Microsoft in 2025, as Nadella tells it:

This new division will bring together Dev Div, AI Platform, and some key teams from the Office of the CTO (AI Supercomputer, AI Agentic Runtimes, and Engineering Thrive), with the mission to build the end-to-end Copilot & AI stack for both our first-party and third-party customers to build and run AI apps and agents. This group will also build out GitHub Copilot, thus having a tight feedback loop between the leading AI-first product and the AI platform to motivate the stack and its roadmap.

To accomplish all that, “Jay Parikh will lead this group as EVP.” Parikh was hired by Microsoft in October; he previously worked as the VP and global head of engineering at Meta.

The fact that the blog post doesn’t say anything about .NET or Visual Studio, instead emphasizing GitHub Copilot and anything and everything related to agentic AI, says a lot about how Nadella sees Microsoft’s future priorities.

So-called AI agents are applications that are given specified boundaries (action spaces) and a large memory capacity to independently do subsets of the kinds of work that human office workers do today. Some company leaders and AI commentators believe these agents will outright replace jobs, while others are more conservative, suggesting they’ll simply be powerful tools to streamline the jobs people already have.

Amid a flurry of hype, Microsoft reorganizes entire dev team around AI Read More »

Meta to cut 5% of employees deemed unfit for Zuckerberg’s AI-fueled future

AI, augmented reality, facebook, Instagram, layoffs, mark zuckerberg, Meta, Policy, smart glasses / Mike M. / January 14, 2025

Anticipating that 2025 will be an “intense year” requiring rapid innovation, Mark Zuckerberg reportedly announced that Meta would be cutting 5 percent of its workforce—targeting “lowest performers.”

Bloomberg reviewed the internal memo explaining the cuts, which was posted to Meta’s internal Workplace forum Tuesday. In it, Zuckerberg confirmed that Meta was shifting its strategy to “move out low performers faster” so that Meta can hire new talent to fill those vacancies this year.

“I’ve decided to raise the bar on performance management,” Zuckerberg said. “We typically manage out people who aren’t meeting expectations over the course of a year, but now we’re going to do more extensive performance-based cuts during this cycle.”

Cuts will likely impact more than 3,600 employees, as Meta’s most recent headcount in September totaled about 72,000 employees. It may not be as straightforward as letting go anyone with an unsatisfactory performance review, as Zuckerberg said that any employee not currently meeting expectations could be spared if Meta is “optimistic about their future performance,” The Wall Street Journal reported.

Any employees affected will be notified by February 10 and receive “generous severance,” Zuckerberg’s memo promised.

This is the biggest round of cuts at Meta since 2023, when Meta laid off 10,000 employees during what Zuckerberg dubbed the “year of efficiency.” Those layoffs followed a prior round where 11,000 lost their jobs and Zuckerberg realized that “leaner is better.” He told employees in 2023 that a “surprising result” from reducing the workforce was “that many things have gone faster.”

“A leaner org will execute its highest priorities faster,” Zuckerberg wrote in 2023. “People will be more productive, and their work will be more fun and fulfilling. We will become an even greater magnet for the most talented people. That’s why in our Year of Efficiency, we are focused on canceling projects that are duplicative or lower priority and making every organization as lean as possible.”

Meta to cut 5% of employees deemed unfit for Zuckerberg’s AI-fueled future Read More »

$getting-an-all-optical-ai-to-handle-non-linear-math$

Getting an all-optical AI to handle non-linear math

AI, Computer science, optical processing, optics, Science / DJ Henderson / January 12, 2025

The problem is that this cascading requires massive parallel computations that, when done on standard computers, take tons of energy and time. Bandyopadhyay’s team feels this problem can be solved by performing the equivalent operations using photons rather than electrons. In photonic chips, information can be encoded in optical properties like polarization, phase, magnitude, frequency, and wavevector. While this would be extremely fast and energy-efficient, building such chips isn’t easy.

Siphoning light

“Conveniently, photonics turned out to be particularly good at linear matrix operations,” Bandyopadhyay claims. A group at MIT led by Dirk Englund, a professor who is a co-author of Bandyopadhyay’s study, demonstrated a photonic chip doing matrix multiplication entirely with light in 2017. What the field struggled with, though, was implementing non-linear functions in photonics.

The usual solution, so far, relied on bypassing the problem by doing linear algebra on photonic chips and offloading non-linear operations to external electronics. This, however, increased latency, since the information had to be converted from light to electrical signals, processed on an external processor, and converted back to light. “And bringing the latency down is the primary reason why we want to build neural networks in photonics,” Bandyopadhyay says.

To solve this problem, Bandyopadhyay and his colleagues designed and built what is likely to be the world’s first chip that can compute the entire deep neural net, including both linear and non-linear operations, using photons. “The process starts with an external laser with a modulator that feeds light into the chip through an optical fiber. This way we convert electrical inputs to light,” Bandyopadhyay explains.

The light is then fanned out to six channels and fed into a layer of six neurons that perform linear matrix multiplication using an array of devices called Mach-Zehnder interferometers. “They are essentially programmable beam splitters, taking two optical fields and mixing them coherently to produce two output optical fields. By applying the voltage, you can control how much those the two inputs mix,” Bandyopadhyay says.

Getting an all-optical AI to handle non-linear math Read More »

Three bizarre home devices and a couple good things at CES 2025

AI, ai slop, CES, CES 2025, Google Home, Google Nest Hub, home automation, lg, matter, philips hue, Smart Home, switchbot, Tech / DJ Henderson / January 11, 2025

You can’t replace cats with AI, not yet

Some quietly good things made an appearance at CES 2025, amidst the AI slush.

Credit: Verity Burns/WIRED UK

Every year, thousands of product vendors, journalists, and gadget enthusiasts gather in an unreasonable city to gawk at mostly unrealistic products.

To be of service to our readers, Ars has done the work of looking through hundreds of such items presented at the 2025 Consumer Electronic Show, pulling out the most bizarre, unnecessary, and head-scratching items. Andrew Cunningham swept across PC and gaming accessories. This writer stuck to goods related to the home.

It’s a lie to say it’s all a prank, so I snuck in a couple of actually good things for human domiciles announced during CES. But the stuff you’ll want to tell your family and friends about in mock disbelief? Plenty of that, still.

AI-powered spice dispenser: Spicerr

A hand holding a white tubular device, with spice tubes loaded into a bottom area, spices dropping out of the bottom. — Credit: Spicerr

Part of my job is to try and stretch my viewpoint outward—to encompass people who might not have the same experiences and who might want different things from technology. Not everybody is a professional writer, pecking away in Markdown about the latest turn-based strategy game. You must try to hear many timbres inside the common voice in your head when addressing new products and technologies.

I cannot get there with Spicerr, the “world’s first AI-powered spice dispenser,” even leaving aside the AI bit. Is the measurement and dumping of spices into a dish even five percent of the overall challenge? Will a mechanical dispenser be any more precise than standard teaspoons? Are there many kinds of food on which you would want to sprinkle a “customized blend” of spices? Are there home cooks so dedicated to fresh, bright flavors that they want their spices delivered in small vials, at presumably premium prices, rather than simply having small quantities of regularly restocked essentials?

Maybe the Spicerr would be a boon to inexperienced cooks, whose relatives all know them to under-season their food. Rather than buying them a battery-powered device, they must charge to “take the guesswork out of seasoning,” though, you could … buy them good cookbooks, or a Times Cooking subscription, or just a few new bottles of paprika, oregano, cumin, cayenne, and turmeric.

Philips Hue’s (sigh) AI-powered lighting assistants

Image of AI assistant responding to prompts from user, — Credit: Signify

I’m not dismayed that Philips Hue is jumping on the “This has AI now” bandwagon. Well, I am, but not specifically dismayed, because every vendor at CES this year is hawking AI. No, the bad thing here is that Hue lights are devices that work great. Maybe Philips’ pursuit of an “AI assistant” to help you figure out that Halloween lights should be orange-ish won’t distract them from their core product’s reliability. But I have my doubts.

Hue has recently moved from a relatively open lighting system to an app-and-account-required, cloud-controlled scheme, supposedly in the name of security and user control. Having an AI assistant is perhaps another way to sell services beyond hardware, like the $130 or $3/month LG TV app it now offers. The AI service is free for now, but charging for it in the future is far from impossible.

Again, none of this should necessarily affect people who, like me, use Hue bulbs to have a porch light come on at sunset or turn a dim, warm hue when it’s time to wind down. But it felt like Hue, which charges a very decent amount for their hardware, might have held off on chasing this trend.

Robot vacuums doing way too much

Switchbot K20+ Pro holding up a tablet while a woman does a yoga pose in front of an insanely wealthy-person view of a California cliffside. — Credit: Switchbot

Robot vacuums are sometimes worth the hassle and price… if you don’t mind doing a pre-vacuum sweep of things that might get stuck in its brushes, you’ve got room for an emptying base or will empty it yourself, and you don’t mind that they usually miss floor edges and corners. They’re fine, I’m saying.

Robot vacuum makers have steadfastly refused to accept “fine” and are out way over their skis this year. In one trade show, you can find:

Eureka’s J15 Max Ultra, incorporating “IntelliView AI 2.0,” infrared, and FHD vision, detects liquid spills and switches brushes and vacuums to better clean and avoid spreading.
Roborock’s Saros Z70 has a “mechanical task arm” that can pick up objects like socks and small debris (up to 10.5 ounces) and put them in a pre-determined pile spot.
SwitchBot’s modular K20+ Pro, which is a vacuum onto which you can attach air purifiers, tablet mounts, security cameras, or other things you want rolling around your home.
Dreame’s X50, which can pivot to clean some small ledges but cannot actually climb.
The Narwal Flow, which has a wide, flat, off-center mop to reach wall edges.

Pricing and availability are not available for these vacuums yet, but each is likely to set you back the equivalent of at least one new MacBook. They are also rather big devices to stash in your home (it’s hard to hide an arm or an air purifier). Each is an early adopter device, and getting replacement consumable parts for them long-term is an uncertain bet. I’m not sure who they are for, but that has not stopped this apparently fertile field from growing many new products.

Now for good things, starting with Google Home

Nest Hub second generation, on a nightstand with a bamboo top and dim lamp in the near background. — Credit: Corey Gaskin

I’ve been watching and occasionally writing about the progress of the nascent Matter smart home protocol, somewhat in the vein of a high school coach who knows their team is held back by a lack of coordination, communication, and consistent direction. What Matter wants to do is vital for the future of the smart home, but it’s very much a loose scrimmage right now.

And yet, this week, in a CES-adjacent announcement, Google reminded me that Matter can really, uh, matter. All of Google Home’s hub devices—Nest screens and speakers, Chromecasts, Google TV devices running at least Android 14, and a few other gadgets—can interoperate with Matter devices locally, with no cloud required.

That means people with a Google Home setup can switch devices, adjust volumes, and otherwise control devices, faster, with Internet outages or latency no longer an issue. Local, no-cloud-required control of devices across brands is one of Matter’s key promises, and seeing it happen inside one major home brand is encouraging.

More we’ll-see-what-happens news is the unveiling of the public Home APIs, which promise to make it easier for third-party devices to be set up, integrated, and automated in a Google Home setup. Even if you’re skeptical of Google’s long-term support for APIs, the company is also working with the Matter group to improve the Matter certification process for all devices. Device makers should then have Matter to fall back onto, failing enthusiasm for Google Home APIs.

This cat tower is also an air purifier; it is also good

Two fake cats, sitting on seats atop an air purifier at CES 2025 — Credit: Verity Burns/WIRED UK

There are a lot of phones out there that need charging and a bunch of gamers who, for some reason, need even more controllers and screens to play on. But there is another, eternally underserved market getting some attention at CES: cats wanting to sit.

LG, which primarily concerned itself with stuffing generative AI interfaces into every other device at CES 2025, crafted something that feels like a real old-time trade show gimmick. There is no guarantee that your cat will use the AeroCat Tower; some cats may just sit inside the cardboard box it came in out of spite. But should they deign to luxuriate on it, the AeroCat will provide gentle heat beneath them, weigh them, and give you a record of their sleep habits. Also, it purifies the air in that room.

There is no pricing or availability information yet. But if you like your cats, you want to combine the function of a cat tower and air purifier, or you just want to consider something even just a little bit fun about the march of technology, look out for this one.

Kevin is a senior technology reporter at Ars Technica, covering open-source software, PC gaming, home automation, repairability, e-bikes, and tech history. He has previously worked at Lifehacker, Wirecutter, iFixit, and Carbon Switch.

Three bizarre home devices and a couple good things at CES 2025 Read More »

161 years ago, a New Zealand sheep farmer predicted AI doom

AI, Biz & IT, charles darwin, machine learning / Shannon Garcia / January 11, 2025

The text anticipated several modern AI safety concerns, including the possibility of machine consciousness, self-replication, and humans losing control of their technological creations. These themes later appeared in works like Isaac Asimov’s The Evitable Conflict, Frank Herbert’s Dune novels (Butler possibly served as the inspiration for the term “Butlerian Jihad“), and the Matrix films.

A model of Charles Babbage's Analytical Engine, a calculating machine invented in 1837 but never built during Babbage's lifetime. — A model of Charles Babbage’s Analytical Engine, a calculating machine invented in 1837 but never built during Babbage’s lifetime. Credit: DE AGOSTINI PICTURE LIBRARY via Getty Images

Butler’s letter dug deep into the taxonomy of machine evolution, discussing mechanical “genera and sub-genera” and pointing to examples like how watches had evolved from “cumbrous clocks of the thirteenth century”—suggesting that, like some early vertebrates, mechanical species might get smaller as they became more sophisticated. He expanded these ideas in his 1872 novel Erewhon, which depicted a society that had banned most mechanical inventions. In his fictional society, citizens destroyed all machines invented within the previous 300 years.

Butler’s concerns about machine evolution received mixed reactions, according to Butler in the preface to the second edition of Erewhon. Some reviewers, he said, interpreted his work as an attempt to satirize Darwin’s evolutionary theory, though Butler denied this. In a letter to Darwin in 1865, Butler expressed his deep appreciation for The Origin of Species, writing that it “thoroughly fascinated” him and explained that he had defended Darwin’s theory against critics in New Zealand’s press.

What makes Butler’s vision particularly remarkable is that he was writing in a vastly different technological context when computing devices barely existed. While Charles Babbage had proposed his theoretical Analytical Engine in 1837—a mechanical computer using gears and levers that was never built in his lifetime—the most advanced calculating devices of 1863 were little more than mechanical calculators and slide rules.

Butler extrapolated from the simple machines of the Industrial Revolution, where mechanical automation was transforming manufacturing, but nothing resembling modern computers existed. The first working program-controlled computer wouldn’t appear for another 70 years, making his predictions of machine intelligence strikingly prescient.

Some things never change

The debate Butler started continues today. Two years ago, the world grappled with what one might call the “great AI takeover scare of 2023.” OpenAI’s GPT-4 had just been released, and researchers evaluated its “power-seeking behavior,” echoing concerns about potential self-replication and autonomous decision-making.

161 years ago, a New Zealand sheep farmer predicted AI doom Read More »

Viral ChatGPT-powered sentry gun gets shut down by OpenAI

AI / Mike M. / January 11, 2025

OpenAI says it has cut off API access to an engineer whose video of a motorized sentry gun controlled by ChatGPT-powered commands has set off a viral firestorm of concerns about AI-powered weapons.

An engineer going by the handle sts_3d started posting videos of a motorized, auto-rotating swivel chair project in August. By November, that same assembly appeared to seamlessly morph into the basis for a sentry gun that could quickly rotate to arbitrary angles and activate a servo to fire precisely aimed projectiles (though only blanks and simulated lasers are shown being fired in his videos).

Earlier this week, though, sts_3d started getting wider attention for a new video showing the sentry gun’s integration with OpenAI’s real-time API. In the video, the gun uses that ChatGPT integration to aim and fire based on spoken commands from sts_3d and even responds in a chirpy voice afterward.

@sts_3d OpenAI Realtime API project integration #robotics #ai #openai ♬ original sound – sts_3d

“If you need any other assistance, please let me know,” the ChatGPT-powered gun says after firing a volley at one point. “Good job, you saved us,” sts_3d responds, deadpan.

“I’m glad I could help!” ChatGPT intones happily.

In response to a comment request from Futurism, OpenAI said it had “proactively identified this violation of our policies and notified the developer to cease this activity ahead of receiving your inquiry. OpenAI’s Usage Policies prohibit the use of our services to develop or use weapons or to automate certain systems that can affect personal safety.”

Halt, intruder alert!

The “voice-powered killer AI robot angle” has garnered plenty of viral attention for sts_3d’s project in recent days. But the ChatGPT integration shown in his video doesn’t exactly reach Terminator levels of a terrifying killing machine. Here, ChatGPT instead ends up looking more like a fancy, overwrought voice-activated remote control for a legitimately impressive gun mount.

Viral ChatGPT-powered sentry gun gets shut down by OpenAI Read More »

Elon Musk wants courts to force OpenAI to auction off a large ownership stake

AI, elon musk, grok, openai, Policy, syndication / 9u50fv / January 11, 2025

Musk, who founded his own AI startup xAI in 2023, has recently stepped up efforts to derail OpenAI’s conversion.

In November, he sought to block the process with a request for a preliminary injunction filed in California. Meta has also thrown its weight behind the suit.

In legal filings from November, Musk’s team wrote: “OpenAI and Microsoft together exploiting Musk’s donations so they can build a for-profit monopoly, one now specifically targeting xAI, is just too much.”

Kathleen Jennings, attorney-general in Delaware—where OpenAI is incorporated—has since said her office was responsible for ensuring that OpenAI’s conversion was in the public interest and determining whether the transaction was at a fair price.

Members of Musk’s camp—wary of Delaware authorities after a state judge rejected a proposed $56 billion pay package for the Tesla boss last month—read that as a rebuke of his efforts to block the conversion, and worry it will be rushed through. They have also argued OpenAI’s PBC conversion should happen in California, where the company has its headquarters.

In a legal filing last week Musk’s attorneys said Delaware’s handling of the matter “does not inspire confidence.”

OpenAI committed to become a public benefit corporation within two years as part of a $6.6 billion funding round in October, which gave it a valuation of $157 billion. If it fails to do so, investors would be able to claw back their money.

There are a number of issues OpenAI is yet to resolve, including negotiating the value of Microsoft’s investment in the PBC. A conversion was not imminent and would be likely to take months, according to the person with knowledge of the company’s thinking.

A spokesperson for OpenAI said: “Elon is engaging in lawfare. We remain focused on our mission and work.” The California and Delaware attorneys-general did not immediately respond to a request for comment.

Elon Musk wants courts to force OpenAI to auction off a large ownership stake Read More »

Microsoft sues service for creating illicit content with its AI platform

AI, Artificial Intelligence, Biz & IT, generative ai, microsoft, Policy, Security / 9u50fv / January 10, 2025

Microsoft and others forbid using their generative AI systems to create various content. Content that is off limits includes materials that feature or promote sexual exploitation or abuse, is erotic or pornographic, or attacks, denigrates, or excludes people based on race, ethnicity, national origin, gender, gender identity, sexual orientation, religion, age, disability status, or similar traits. It also doesn’t allow the creation of content containing threats, intimidation, promotion of physical harm, or other abusive behavior.

Besides expressly banning such usage of its platform, Microsoft has also developed guardrails that inspect both prompts inputted by users and the resulting output for signs the content requested violates any of these terms. These code-based restrictions have been repeatedly bypassed in recent years through hacks, some benign and performed by researchers and others by malicious threat actors.

Microsoft didn’t outline precisely how the defendants’ software was allegedly designed to bypass the guardrails the company had created.

Masada wrote:

Microsoft’s AI services deploy strong safety measures, including built-in safety mitigations at the AI model, platform, and application levels. As alleged in our court filings unsealed today, Microsoft has observed a foreign-based threat–actor group develop sophisticated software that exploited exposed customer credentials scraped from public websites. In doing so, they sought to identify and unlawfully access accounts with certain generative AI services and purposely alter the capabilities of those services. Cybercriminals then used these services and resold access to other malicious actors with detailed instructions on how to use these custom tools to generate harmful and illicit content. Upon discovery, Microsoft revoked cybercriminal access, put in place countermeasures, and enhanced its safeguards to further block such malicious activity in the future.

The lawsuit alleges the defendants’ service violated the Computer Fraud and Abuse Act, the Digital Millennium Copyright Act, the Lanham Act, and the Racketeer Influenced and Corrupt Organizations Act and constitutes wire fraud, access device fraud, common law trespass, and tortious interference. The complaint seeks an injunction enjoining the defendants from engaging in “any activity herein.”

Microsoft sues service for creating illicit content with its AI platform Read More »

AI could create 78 million more jobs than it eliminates by 2030—report

AI, AI and jobs, AI layoffs, Biz & IT, machine learning, World Economic Forum / DJ Henderson / January 10, 2025

On Wednesday, the World Economic Forum (WEF) released its Future of Jobs Report 2025, with CNN immediately highlighting the finding that 40 percent of companies plan workforce reductions due to AI automation. But the report’s broader analysis paints a far more nuanced picture than CNN’s headline suggests: It finds that AI could create 170 million new jobs globally while eliminating 92 million positions, resulting in a net increase of 78 million jobs by 2030.

“Half of employers plan to re-orient their business in response to AI,” writes the WEF in the report. “Two-thirds plan to hire talent with specific AI skills, while 40% anticipate reducing their workforce where AI can automate tasks.”

The survey collected data from 1,000 companies that employ 14 million workers globally. The WEF conducts its employment analysis every two years to help policymakers, business leaders, and workers make decisions about hiring trends.

The new report points to specific skills that will dominate hiring by 2030. Companies ranked AI and big data expertise, networks and cybersecurity, and technological literacy as the three most in-demand skill sets.

The WEF identified AI as the biggest potential job creator among new technologies, with 86 percent of companies expecting AI to transform their operations by 2030.

Declining job categories

The WEF report also identifies specific job categories facing decline. Postal service clerks, executive secretaries, and payroll staff top the list of shrinking roles, with changes driven by factors including (but not limited to) AI adoption. And for the first time, graphic designers and legal secretaries appear among the fastest-declining positions, which the WEF tentatively links to generative AI’s expanding capabilities in creative and administrative work.

AI could create 78 million more jobs than it eliminates by 2030—report Read More »

Why I’m disappointed with the TVs at CES 2025

AI, CES, CES 2025, generative ai, Google, lg, Microsoft Copilot, oled, samsung, Software, Tech, TVs / Kelly Newman / January 9, 2025

Won’t someone please think of the viewer?

Op-ed: TVs miss opportunity for real improvement by prioritizing corporate needs.

The TV industry is hitting users over the head with AI and other questionable gimmicks Credit: Getty

If you asked someone what they wanted from TVs released in 2025, I doubt they’d say “more software and AI.” Yet, if you look at what TV companies have planned for this year, which is being primarily promoted at the CES technology trade show in Las Vegas this week, software and AI are where much of the focus is.

The trend reveals the implications of TV brands increasingly viewing themselves as software rather than hardware companies, with their products being customer data rather than TV sets. This points to an alarming future for smart TVs, where even premium models sought after for top-end image quality and hardware capabilities are stuffed with unwanted gimmicks.

LG’s remote regression

LG has long made some of the best—and most expensive—TVs available. Its OLED lineup, in particular, has appealed to people who use their TVs to watch Blu-rays, enjoy HDR, and the like. However, some features that LG is introducing to high-end TVs this year seem to better serve LG’s business interests than those users’ needs.

Take the new remote. Formerly known as the Magic Remote, LG is calling the 2025 edition the AI Remote. That is already likely to dissuade people who are skeptical about AI marketing in products (research suggests there are many such people). But the more immediately frustrating part is that the new remote doesn’t have a dedicated button for switching input modes, as previous remotes from LG and countless other remotes do.

LG AI remote — LG’s AI Remote. Credit: Tom’s Guide/YouTube

To use the AI Remote to change the TV’s input—a common task for people using their sets to play video games, watch Blu-rays or DVDs, connect their PC, et cetera—you have to long-press the Home Hub button. Single-pressing that button brings up a dashboard of webOS (the operating system for LG TVs) apps. That functionality isn’t immediately apparent to someone picking up the remote for the first time and detracts from the remote’s convenience.

By overlooking other obviously helpful controls (play/pause, fast forward/rewind, and numbers) while including buttons dedicated to things like LG’s free ad-supported streaming TV (FAST) channels and Amazon Alexa, LG missed an opportunity to update its remote in a way centered on how people frequently use TVs. That said, it feels like user convenience didn’t drive this change. Instead, LG seems more focused on getting people to use webOS apps. LG can monetize app usage through, i.e., getting a cut of streaming subscription sign-ups, selling ads on webOS, and selling and leveraging user data.

Moving from hardware provider to software platform

LG, like many other TV OEMs, has been growing its ads and data business. Deals with data analytics firms like Nielsen give it more incentive to acquire customer data. Declining TV margins and rock-bottom prices from budget brands (like Vizio and Roku, which sometimes lose money on TV hardware sales and make up for the losses through ad sales and data collection) are also pushing LG’s software focus. In the case of the AI Remote, software prioritization comes at the cost of an oft-used hardware capability.

Further demonstrating its motives, in September 2023, LG announced intentions to “become a media and entertainment platform company” by offering “services” and a “collection of curated content in products, including LG OLED and LG QNED TVs.” At the time, the South Korean firm said it would invest 1 trillion KRW (about $737.7 million) into its webOS business through 2028.

Low TV margins, improved TV durability, market saturation, and broader economic challenges are all serious challenges for an electronics company like LG and have pushed LG to explore alternative ways to make money off of TVs. However, after paying four figures for TV sets, LG customers shouldn’t be further burdened to help LG accrue revenue.

Google TVs gear up for subscription-based features

There are numerous TV manufacturers, including Sony, TCL, and Philips, relying on Google software to power their TV sets. Numerous TVs announced at CES 2025 will come with what Google calls Gemini Enhanced Google Assistant. The idea that this is something that people using Google TVs have requested is somewhat contradicted by Google Assistant interactions with TVs thus far being “somewhat limited,” per a Lowpass report.

Nevertheless, these TVs are adding far-field microphones so that they can hear commands directed at the voice assistant. For the first time, the voice assistant will include Google’s generative AI chatbot, Gemini, this year—another feature that TV users don’t typically ask for. Despite the lack of demand and the privacy concerns associated with microphones that can pick up audio from far away even when the TV is off, companies are still loading 2025 TVs with far-field mics to support Gemini. Notably, these TVs will likely allow the mics to be disabled, like you can with other TVs using far-field mics. But I still ponder about features/hardware that could have been implemented instead.

Google is also working toward having people pay a subscription fee to use Gemini on their TVs, PCWorld reported.

“For us, our biggest goal is to create enough value that yes, you would be willing to pay for [Gemini],” Google TV VP and GM Shalini Govil-Pai told the publication.

The executive pointed to future capabilities for the Gemini-driven Google Assistant on TVs, including asking it to “suggest a movie like Jurassic Park but suitable for young children” or to show “Bollywood movies that are similar to Mission: Impossible.”

She also pointed to future features like showing weather, top news stories, and upcoming calendar events when someone is near the TV, showing AI-generated news briefings, and the ability to respond to questions like “explain the solar system to a third-grader” with text, audio, and YouTube videos.

But when people have desktops, laptops, tablets, and phones in their homes already, how helpful are these features truly? Govil-Pai admitted to PCWorld that “people are not used to” using their TVs this way “so it will take some time for them to adapt to it.” With this in mind, it seems odd for TV companies to implement new, more powerful microphones to support features that Google acknowledges aren’t in demand. I’m not saying that tech companies shouldn’t get ahead of the curve and offer groundbreaking features that users hadn’t considered might benefit them. But already planning to monetize those capabilities—with a subscription, no less—suggests a prioritization of corporate needs.

Samsung is hungry for AI

People who want to use their TV for cooking inspiration often turn to cooking shows or online cooking videos. However, Samsung wants people to use its TV software to identify dishes they want to try making.

During CES, Samsung announced Samsung Food for TVs. The feature leverages Samsung TVs’ AI processors to identify food displayed on the screen and recommend relevant recipes. Samsung introduced the capability in 2023 as an iOS and Android app after buying the app Whisk in 2019. As noted by TechCrunch, though, other AI tools for providing recipes based on food images are flawed.

So why bother with such a feature? You can get a taste of Samsung’s motivation from its CES-announced deal with Instacart that lets people order off Instacart from Samsung smart fridges that support the capability. Samsung Food on TVs can show users the progress of food orders placed via the Samsung Food mobile app on their TVs. Samsung Food can also create a shopping list for recipe ingredients based on what it knows (using cameras and AI) is in your (supporting) Samsung fridge. The feature also requires a Samsung account, which allows the company to gather more information on users.

Other software-centric features loaded into Samsung TVs this year include a dedicated AI button on the new TVs’ remotes, the ability to use gestures to control the TV but only if you’re wearing a Samsung Galaxy Watch, and AI Karaoke, which lets people sing karaoke using their TVs by stripping vocals from music playing and using their phone as a mic.

Like LG, Samsung has shown growing interest in ads and data collection. In May, for example, it expanded its automatic content recognition tech to track ad exposure on streaming services viewed on its TVs. It also has an ads analytics partnership with Experian.

Large language models on TVs

TVs are mainstream technology in most US homes. Generative AI chatbots, on the other hand, are emerging technology that many people have yet to try. Despite these disparities, LG and Samsung are incorporating Microsoft’s Copilot chatbot into 2025 TVs.

LG claims that Copilot will help its TVs “understand conversational context and uncover subtle user intentions,” adding: “Access to Microsoft Copilot further streamlines the process, allowing users to efficiently find and organize complex information using contextual cues. For an even smoother and more engaging experience, the AI chatbot proactively identifies potential user challenges and offers timely, effective solutions.”

Similarly, Samsung, which is also adding Copilot to some of its smart monitors, said in its announcement that Copilot will help with “personalized content recommendations.” Samsung has also said that Copilot will help its TVs understand strings of commands, like increasing the volume and changing the channel, CNET noted. Samsung said it intends to work with additional AI partners, namely Google, but it’s unclear why it needs multiple AI partners, especially when it hasn’t yet seen how people use large language models on their TVs.

TV-as-a-platform

To be clear, this isn’t a condemnation against new, unexpected TV features. This also isn’t a censure against new TV apps or the usage of AI in TVs.

AI marketing hype is real and misleading regarding the demand, benefits, and possibilities of AI in consumer gadgets. However, there are some cases when innovative software, including AI, can improve things that TV users not only care about but actually want or need. For example, some TVs use AI for things like trying to optimize sound, color, and/or brightness, including based on current environmental conditions or upscaling. This week, Samsung announced AI Live Translate for TVs. The feature is supposed to be able to translate foreign language closed captions in real time, providing a way for people to watch more international content. It’s a feature I didn’t ask for but can see being useful and changing how I use my TV.

But a lot of this week’s TV announcements underscore an alarming TV-as-a-platform trend where TV sets are sold as a way to infiltrate people’s homes so that apps, AI, and ads can be pushed onto viewers. Even high-end TVs are moving in this direction and amplifying features with questionable usefulness, effectiveness, and privacy considerations. Again, I can’t help but wonder what better innovations could have come out this year if more R&D was directed toward hardware and other improvements that are more immediately rewarding for users than karaoke with AI.

The TV industry is facing economic challenges, and, understandably, TV brands are seeking creative solutions for making money. But for consumers, that means paying for features that you’re likely to ignore. Ultimately, many people just want a TV with amazing image and sound quality. Finding that without having to sift through a bunch of fluff is getting harder.

Scharon is a Senior Technology Reporter at Ars Technica writing news, reviews, and analysis on consumer gadgets and services. She’s been reporting on technology for over 10 years, with bylines at Tom’s Hardware, Channelnomics, and CRN UK.

Why I’m disappointed with the TVs at CES 2025 Read More »