“Since the rise of large language models like ChatGPT there have been lots of anecdotal reports about students submitting AI-generated work as their exam assignments and getting good grades. So, we stress-tested our university’s examination system against AI cheating in a controlled experiment,” says Peter Scarfe, a researcher at the School of Psychology and Clinical Language Sciences at the University of Reading.
His team created over 30 fake psychology student accounts and used them to submit ChatGPT-4-produced answers to examination questions. The anecdotal reports were true—the AI use went largely undetected, and, on average, ChatGPT scored better than human students.
Rules of engagement
Scarfe’s team submitted AI-generated work in five undergraduate modules, covering classes needed during all three years of study for a bachelor’s degree in psychology. The assignments were either 200-word answers to short questions or more elaborate essays, roughly 1,500 words long. “The markers of the exams didn’t know about the experiment. In a way, participants in the study didn’t know they were participating in the study, but we’ve got necessary permissions to go ahead with that”, Scarfe claims.
An anonymous reader shares a report: Bill Gates has defended the rapid rise in energy use caused by AI systems, arguing the technology would ultimately offset its heavy consumption of electricity. Speaking in London, Gates urged environmentalists and governments to "not go overboard" on concerns about the huge amounts of power required to run new generative AI systems, as Big Tech companies such as Microsoft race to invest tens of billions of dollars in vast new data centres.
Data centres will drive a rise in global electricity usage of between 2-6 per cent, the billionaire said. "The question is, will AI accelerate a more than 6 per cent reduction? And the answer is: certainly," said Gates, the Microsoft co-founder who has been a prolific investor in companies developing sustainable energy and carbon- reduction technologies. In May, Microsoft admitted that its greenhouse gas emissions had risen by almost a third since 2020, in large part due to the construction of data centres.
Gates, who left Microsoft's board in 2020 but remains an adviser to chief executive Satya Nadella, said tech companies would pay a "green premium" -- or higher price -- for clean energy as they seek new sources of power, which was helping to drive its development and deployment. "The tech companies are the people willing to pay a premium and to help bootstrap green energy capacity," he said at the Breakthrough Energy Summit in London on Thursday.
Google Translate can come in handy when you're traveling or communicating with someone who speaks another language, and thanks to a new update, you can now connect with some 614 million more people. Google is adding 110 new languages to its Translate tool using its AI PaLM 2 large language model (LLM), which brings the total of supported languages to nearly 250. This follows the 24 languages added in 2022, including Indigenous languages of the Americas as well as those spoken across Africa and central Asia.
Many of the recently added languages are ones you've probably never heard of, as they're spoken only by small communities or have no native speakers.
Cantonese, which is spoken predominantly in southeastern China, Hong Kong, and Macau as well as communities across the world, may be one of the most recognizable additions with this update. According to Google, the overlap between Cantonese and Mandarin—which was already available—makes it difficult to train LLMs. Punjabi (Shahmukhi), the most spoken language in Pakistan, is also now available.
A quarter of the newly supported languages come from Africa, and include Afar—spoken in Djibouti, Eritrea, and Ethiopia—and Tamazight (Amazigh), a Berber language used across North Africa, as well as NKo, Fon, Kikongo, Luo, Ga, Swati, Venda, and Wolof.
You can also now use Google Translate to communicate in Manx, a Celtic language from the Isle of Man in the Irish Sea. The last native speaker of Manx died in 1974, leading to its near-extinction, but thanks to revitalization efforts, there are now a few dozen first-language speakers, and a couple thousand who speak Manx as a second language.
The update also includes Portuguese (Portugal), Tongan, Tibetan, Tahitian, Venetian, Sicilian, Fijian, and Jamaican Patois.
The Google Translate app is available in the Apple App Store and the Google Play store. It can translate text you paste in, as well as text appearing in photos. It can also translate voice input as well as handwritten characters.
Enlarge/ EU competition chief Margrethe Vestager said the bloc was looking into practices that could in effect lead to a company controlling a greater share of the AI market. (credit: Thierry Monasse/Getty Images)
Brussels is preparing for an antitrust investigation into Microsoft’s $13 billion investment into OpenAI, after the European Union decided not to proceed with a merger review into the most powerful alliance in the artificial intelligence industry.
The European Commission, the EU’s executive arm, began to explore a review under merger control rules in January, but on Friday announced that it would not proceed due to a lack of evidence that Microsoft controls OpenAI.
However, the commission said it was now exploring the possibility of a traditional antitrust investigation into whether the tie-up between the world’s most valuable listed company and the best-funded AI start-up was harming competition in the fast-growing market.
Enlarge/ A movable robotic face covered with living human skin cells. (credit: Takeuchi et al.)
In a new study, researchers from the University of Tokyo, Harvard University, and the International Research Center for Neurointelligence have unveiled a technique for creating lifelike robotic skin using living human cells. As a proof of concept, the team engineered a small robotic face capable of smiling, covered entirely with a layer of pink living tissue.
The researchers note that using living skin tissue as a robot covering has benefits, as it's flexible enough to convey emotions and can potentially repair itself. "As the role of robots continues to evolve, the materials used to cover social robots need to exhibit lifelike functions, such as self-healing," wrote the researchers in the study.
Enlarge/ An illustration created by OpenAI. (credit: OpenAI)
On Thursday, OpenAI researchers unveiled CriticGPT, a new AI model designed to identify mistakes in code generated by ChatGPT. It aims to enhance the process of making AI systems behave in ways humans want (called "alignment") through Reinforcement Learning from Human Feedback (RLHF), which helps human reviewers make large language model (LLM) outputs more accurate.
As outlined in a new research paper called "LLM Critics Help Catch LLM Bugs," OpenAI created CriticGPT to act as an AI assistant to human trainers who review programming code generated by the ChatGPT AI assistant. CriticGPT—based on the GPT-4 family of LLMS—analyzes the code and points out potential errors, making it easier for humans to spot mistakes that might otherwise go unnoticed. The researchers trained CriticGPT on a dataset of code samples with intentionally inserted bugs, teaching it to recognize and flag various coding errors.
The researchers found that CriticGPT's critiques were preferred by annotators over human critiques in 63 percent of cases involving naturally occurring LLM errors and that human-machine teams using CriticGPT wrote more comprehensive critiques than humans alone while reducing confabulation (hallucination) rates compared to AI-only critiques.
Developing an automated critic
The development of CriticGPT involved training the model on a large number of inputs containing deliberately inserted mistakes. Human trainers were asked to modify code written by ChatGPT, introducing errors and then providing example feedback as if they had discovered these bugs. This process allowed the model to learn how to identify and critique various types of coding errors.
Al Michaels, the 79-year-old American broadcaster, who first covered the Olympics decades ago, is returning to broadcasting via an AI clone. NBCUniversal and Peacock will use AI-generated narration by Al Michaels for daily customized highlight reels of the Summer Olympics. Officials say they anticipate seven million different variations of the customized highlights throughout the games. The New York Times reports: Al Michaels, the 79-year-old American broadcaster, who first covered the Olympics decades ago, is coming back to primetime. It does raise a key question, one that recalls Mr. Michaels's most famous Olympic call: Do NBCUniversal executives believe in miracles? NBC has been exclusively broadcasting the Olympics in the United States since 1996, and the network frequently finds itself subject to intense public scrutiny for its coverage of the Games. [...]
Subscribers who want the daily Peacock highlight reel will be able choose the Olympic events that interest them most, and the types of highlights they want to see, such as viral clips, gold medalists or elimination events. From there, Peacock's A.I. machines will get to work each evening cranking out the most notable moments and putting them together in a tidy customized package. Mr. Michaels's recreated voice will be piped over the reels. (Humans will make quality control checks on the A.I. highlight reels.)
Alibaba's Qwen models dominated Hugging Face's latest LLM leaderboard, securing three top-ten spots. The new benchmark, launched Thursday, tests open-source models on tougher criteria including long-context reasoning and complex math. Meta's Llama3-70B also ranked highly, but several Chinese models outperformed Western counterparts. (Closed-source AIs like ChatGPT were excluded.) The leaderboard replaces an earlier version deemed too easy to game.
Enlarge/ Al Michaels looks on prior to the game between the Minnesota Vikings and Philadelphia Eagles at Lincoln Financial Field on September 14, 2023, in Philadelphia, Pennsylvania. (credit: Getty Images)
On Wednesday, NBC announced plans to use an AI-generated clone of famous sports commentator Al Michaels' voice to narrate daily streaming video recaps of the 2024 Summer Olympics in Paris, which start on July 26. The AI-powered narration will feature in "Your Daily Olympic Recap on Peacock," NBC's streaming service. But this new, high-profile use of voice cloning worries critics, who say the technology may muscle out upcoming sports commentators by keeping old personas around forever.
NBC says it has created a "high-quality AI re-creation" of Michaels' voice, trained on Michaels' past NBC appearances to capture his distinctive delivery style.
The veteran broadcaster, revered in the sports commentator world for his iconic "Do you believe in miracles? Yes!" call during the 1980 Winter Olympics, has been covering sports on TV since 1971, including a high-profile run of play-by-play coverage of NFL football games for both ABC and NBC since the 1980s. NBC dropped him from NFL coverage in 2023, however, possibly due to his age.
An anonymous reader quotes a report from Reuters: Seven content-licensing sellers of music, image, video and other datasets for use in training artificial intelligence systems have formed the sector's first trade group, they said on Wednesday. The Dataset Providers Alliance (DPA) will advocate for 'ethical data sourcing' in the training of AI systems, including rights for people depicted in datasets and the protection of content owners' intellectual property rights, the companies said in a statement. Founding members include U.S. music dataset company Rightsify, image licensing service vAIsual, Japanese stock photo provider Pixta and Germany-based data marketplace Datarade.
An anonymous reader quotes a report from Wired: In recent weeks, as so-called cheap fake video clips suggesting President Joe Biden is unfit for office have gone viral on social media, a Kremlin-affiliated disinformation network has been promoting a parody music video featuring Biden wearing a diaper and being pushed around in a wheelchair. The video is called "Bye, Bye Biden" and has been viewed more than 5 million times on X since it was first promoted in the middle of May. It depicts Biden as senile, wearing a hearing aid, and taking a lot of medication. It also shows him giving money to a character who seems to represent illegal migrants while denying money to US citizens until they change their costume to mimic the Ukrainian flag. Another scene shows Biden opening the front door of a family home that features a Confederate flag on the wall and allowing migrants to come in and take over. Finally, the video contains references to stolen election conspiracies pushed by former president Donald Trump.
The video was created by Little Bug, a group that mimics the style of Little Big, a real Russian band that fled the country in 2022 following Russia's invasion of Ukraine. The video features several Moscow-based actors -- who spoke with Russian media outlet Agency.Media -- but also appears to use artificial intelligence technology to make the actors resemble Biden and Trump, as well as Ilya Prusikin, the lead singer of Little Big. "Biden and Trump appear to be the same actor, with deepfake video-editing changing his facial features until he resembles Biden in one case and Trump in the other case," says Alex Fink, an AI and machine-vision expert who analyzed the video for WIRED. "The editing is inconsistent, so you can see that in some cases he resembles Biden more and in others less. The facial features keep changing." An analysis by True Media, a nonprofit that was founded to tackle the spread of election-related deepfakes, found with 100 percent confidence that there was AI-generated audio used in the video. It also assessed with 78 percent confidence that some AI technology was used to manipulate the faces of the actors.
Fink says the obvious nature of the deepfake technology on display here suggests that the video was created in a rush, using a small number of iterations of a generative adversarial network in order to create the characters of Biden and Trump. It is unclear who is behind the video, but "Bye, Bye Biden" has been promoted by the Kremlin-aligned network known as Doppelganger. The campaign posted tens of thousands of times on X and was uncovered by Antibot4Navalny, an anonymous collective of Russian researchers who have been tracking Doppelganger's activity for the past six months. The campaign first began on May 21, and there have been almost 4,000 posts on X promoting the video in 13 languages that were promoted by a network of almost 25,000 accounts. The Antibot4Navalny researchers concluded that the posts were written with the help of generative AI technology. The video has been shared 6.5 million times on X and has been viewed almost 5 million times.
An anonymous reader quotes a report from Ars Technica: On Monday, Toys "R" Us announced that it had partnered with an ad agency called Native Foreign to create what it calls "the first-ever brand film using OpenAI's new text-to-video tool, Sora." OpenAI debuted Sora in February, but the video synthesis tool has not yet become available to the public. The brand film tells the story of Toys "R" Us founder Charles Lazarus using AI-generated video clips. "We are thrilled to partner with Native Foreign to push the boundaries of Sora, a groundbreaking new technology from OpenAI that's gaining global attention," wrote Toys "R" Us on its website. "Sora can create up to one-minute-long videos featuring realistic scenes and multiple characters, all generated from text instruction. Imagine the excitement of creating a young Charles Lazarus, the founder of Toys "R" Us, and envisioning his dreams for our iconic brand and beloved mascot Geoffrey the Giraffe in the early 1930s."
The company says that The Origin of Toys "R" Us commercial was co-produced by Toys "R" Us Studios President Kim Miller Olko as executive producer and Native Foreign's Nik Kleverov as director. "Charles Lazarus was a visionary ahead of his time, and we wanted to honor his legacy with a spot using the most cutting-edge technology available," Miller Olko said in a statement. In the video, we see a child version of Lazarus, presumably generated using Sora, falling asleep and having a dream that he is flying through a land of toys. Along the way, he meets Geoffery, the store's mascot, who hands the child a small red car. Many of the scenes retain obvious hallmarks of AI-generated imagery, such as unnatural movement, strange visual artifacts, and the irregular shape of eyeglasses. [...] Although the Toys "R" Us video uses key visual elements from Sora, it still required quite a bit of human post-production work to put it together. Sora eliminated the need for actors and cameras, but creating successful generations and piecing together the rest still took human scriptwriters and VFX artists to fill in the AI model's shortcomings. "The brand film was almost entirely created with Sora, with some corrective VFX and an original music score composed by Aaron Marsh of famed indie rock band Copeland," wrote Toys "R" Us in a press release. Comedy writer Mike Drucker wrapped up several of these criticisms into one post, writing: "Love this commercial is like, 'Toys R Us started with the dream of a little boy who wanted to share his imagination with the world. And to show how, we fired our artists and dried Lake Superior using a server farm to generate what that would look like in Stephen King's nightmares.'"
Other critical comments were more frank. Filmmaker Joe Russo posted: "TOYS 'R US released an AI commercial and it fucking sucks."
Exam submissions generated by AI can not only evade detection but also earn higher grades than those submitted by university students, a real-world test has shown. From a report: The findings come as concerns mount about students submitting AI-generated work as their own, with questions being raised about the academic integrity of universities and other higher education institutions. It also shows even experienced markers could struggle to spot answers generated by AI, the University of Reading academics said.
Peter Scarfe, an associate professor at Reading's School of Psychology and Clinical Language Sciences said the findings should serve as a "wake-up call" for educational institutions as AI tools such as ChatGPT become more advanced and widespread. He said: "The data in our study shows it is very difficult to detect AI-generated answers. There has been quite a lot of talk about the use of so-called AI detectors, which are also another form of AI but (the scope here) is limited." For the study, published in the journal Plos One, Prof Scarfe and his team generated answers to exam questions using GPT-4 and submitted these on behalf of 33 fake students. Exam markers at Reading's School of Psychology and Clinical Language Sciences were unaware of the study. Answers submitted for many undergraduate psychology modules went undetected in 94% of cases and, on average, got higher grades than real student submissions, Prof Scarfe said.
Enlarge/ A screen capture from the partially AI-generated Toys "R" Us brand film created using Sora. (credit: Toys R Us)
On Monday, Toys "R" Us announced that it had partnered with an ad agency called Native Foreign to create what it calls "the first-ever brand film using OpenAI's new text-to-video tool, Sora." OpenAI debuted Sora in February, but the video synthesis tool has not yet become available to the public. The brand film tells the story of Toys "R" Us founder Charles Lazarus using AI-generated video clips.
"We are thrilled to partner with Native Foreign to push the boundaries of Sora, a groundbreaking new technology from OpenAI that's gaining global attention," wrote Toys "R" Us on its website. "Sora can create up to one-minute-long videos featuring realistic scenes and multiple characters, all generated from text instruction. Imagine the excitement of creating a young Charles Lazarus, the founder of Toys "R" Us, and envisioning his dreams for our iconic brand and beloved mascot Geoffrey the Giraffe in the early 1930s."
The company says that The Origin of Toys "R" Us commercial was co-produced by Toys "R" Us Studios President Kim Miller Olko as executive producer and Native Foreign's Nik Kleverov as director. "Charles Lazarus was a visionary ahead of his time, and we wanted to honor his legacy with a spot using the most cutting-edge technology available," Miller Olko said in a statement.
At first glance, OpenAI's Mac app offers the usual ChatGPT experience you're used to. When you log in, you'll find all your previous conversations saved to the sidebar, just as they are in the web and mobile apps. You can type your prompts in the text field, use the mic button to ask questions with your voice, and click the headphones icon to enter Voice mode. (Not the "Her" Voice mode, mind you: That feature has been delayed.) You can also use features like Temporary Chats (conversations that don't pull from your chat history), change your GPT model, generate images with DALL-E, and access GPTs.
A better experience than the web app
But there are some Mac-specific features that make this particular app worth using over the web option. First, in addition to uploading files and photos to ChatGPT, you can take a screenshot of any open window on your Mac directly from the app. If you click on the paperclip icon, and select Take Screenshot, you can select an active window from the pop-up list to share with ChatGPT. (The first time you do this, you'll need to grant the ChatGPT app access to screen recording.)
Alternatively, you can take a screenshot of the window manually, then share it to ChatGPT as an image, but this skips a step and makes the bot feel a bit more integrated with macOS.
Credit: Jake Peterson
But what's even more convenient, in my opinion, is the ChatGPT "launcher." This launcher is essentially Spotlight search, but for ChatGPT. Using a keyboard shortcut, you can bring up a ChatGPT text field directly over any window you're currently using on macOS to start a conversation with the bot. You'll then be taken to the app to continue chatting. This basically saves you the step of switching out of the current app you're in and starting a new thread in ChatGPT; if you see something on your Mac you want to know more about, you can hit Option + Spacebar, type your query, and get started.
Credit: Jake Peterson
This launcher also has the same paperclip icon as the app itself, which means you can upload files and take screenshots directly from the shortcut. If you're a ChatGPT power user, this launcher should be a welcome feature. (I don't even use ChatGPT that much, and I really like it.)
Unfortunately, OpenAI is only making the ChatGPT app available on M-series Macs—the machines running Apple silicon. If you have an older Intel-based Mac, you'll still have to head to the web app in order to use ChatGPT on your computer.
If you have a Mac with an M1 chip or newer, you can download the app from OpenAI's download site.
Enlarge (credit: Chris Ratcliffe/Bloomberg via Getty)
YouTube is in talks with record labels to license their songs for artificial intelligence tools that clone popular artists’ music, hoping to win over a skeptical industry with upfront payments.
The Google-owned video site needs labels’ content to legally train AI song generators, as it prepares to launch new tools this year, according to three people familiar with the matter.
The company has recently offered lump sums of cash to the major labels—Sony, Warner, and Universal—to try to convince more artists to allow their music to be used in training AI software, according to several people briefed on the talks.
Researchers claim to have developed a new way to run AI language models more efficiently by eliminating matrix multiplication from the process. This fundamentally redesigns neural network operations that are currently accelerated by GPU chips. The findings, detailed in a recent preprint paper from researchers at the University of California Santa Cruz, UC Davis, LuxiTech, and Soochow University, could have deep implications for the environmental impact and operational costs of AI systems.
Matrix multiplication (often abbreviated to "MatMul") is at the center of most neural network computational tasks today, and GPUs are particularly good at executing the math quickly because they can perform large numbers of multiplication operations in parallel. That ability momentarily made Nvidia the most valuable company in the world last week; the company currently holds an estimated 98 percent market share for data center GPUs, which are commonly used to power AI systems like ChatGPT and Google Gemini.
In the new paper, titled "Scalable MatMul-free Language Modeling," the researchers describe creating a custom 2.7 billion parameter model without using MatMul that features similar performance to conventional large language models (LLMs). They also demonstrate running a 1.3 billion parameter model at 23.8 tokens per second on a GPU that was accelerated by a custom-programmed FPGA chip that uses about 13 watts of power (not counting the GPU's power draw). The implication is that a more efficient FPGA "paves the way for the development of more efficient and hardware-friendly architectures," they write.
Enlarge/ The app lets you invoke ChatGPT from anywhere in the system with a keyboard shortcut, Spotlight-style. (credit: Samuel Axon)
OpenAI's official ChatGPT app for macOS is now available to all users for the first time, provided they're running macOS Sonoma or later.
It was previously being rolled out gradually to paid subscribers to ChatGPT's Plus premium plan.
The ChatGPT Mac app mostly acts as a desktop window version of the web app, allowing you to carry on back-and-forth prompt-and-response conversations. You can select between the GPT-3.5, GPT-4, and GPT-4o models. It also supports the more specialized GPTs available in the web version, including the DALL-E image generator and custom GPTs.
If you or your company pay for Workspace, you may have noticed Google's AI integration with apps like Docs, Sheets, and Drive. The company has been pushing Gemini in its products since their big rebrand from "Bard" back in February, and it appears that train isn't stopping anytime soon: Starting this week, you'll now have access to Gemini via a sidebar panel in some of Google's most-used Workspace apps.
Google says the side panel uses Gemini 1.5 Pro, the LLM the company rolled out back in February, equipped with a "longer context window and more advanced reasoning." That longer context window should be helpful when asking Gemini to analyze long documents or run through large sets of data in Drive, as it allows an LLM to handle more information at once in any given request.
Now, if you've ever used a generative AI experience—especially one from Google—this experience probably won't shock you: You'll see a pretty typical welcome screen when Gemini comes up, in addition to a series of prompt suggestions for you to ask the bot. When you pull up the side panel in a Google Doc, for example, Gemini may immediately offer you a summary of the doc, then present potential prompts, such as "Refine," "Suggest improvements," or "Rephrase." However, the prompt field at the bottom of the panel is always available for you to ask Gemini whatever you want.
Here are some of the uses Google envisions for Gemini in the side panel:
Docs: Help you write, summarize text, generate writing ideas, come up with content from other Google files
Slides: Create new slides, create images for slides, summarize existing presentations
Sheets: Follow and organize your data, create tables, run formulas, ask for help with tasks in the app
Drive: Summarize "one or two documents," ask for the highlights about a project, request a detailed report based on multiple files
Gmail: Summarize a thread, suggest replies to an email, advice on writing an email, ask about emails in your inbox or Drive
Credit: Google
None of these features are necessarily groundbreaking (Gemini has been generally available in Workspace since February) but Google's view is they're now available in a convenient location as you use these apps. In fact, Google announced that Gmail for Android and iOS are also getting Gemini—just not as a side panel. But while the company is convinced that adding its generative AI to its apps will have a positive impact on the end user, I'm not quite sold. After all, this is the first big AI development from Google since the company's catastrophic "AI Overviews" rollout. I, for one, am curious if Gemini will suggest that I respond to an email by sharing instructions on adding glue to pizza.
As companies like Google continue to add new AI features to their products, we're seeing the weak points in real time: Do you want to trust Gemini's summary of a presentation in Slides, or an important conversation in Gmail, when AI still makes things up and treats them like fact?
Who can try Gemini side panel in Google apps
That said, not everyone will actually see Gemini in their Workspace apps, even as Google rolls it out. As of now, Gemini's new side panel feature is only available to companies who purchase the Business and Enterprise Gemini add-on, schools that purchase the Education and Education Premium Gemini add-on, and Google One AI Premium subscribers. If you don't pay for Google's top tier subscription, and your business or school doesn't pay for Gemini, you're not seeing Google's AI in Gmail. Depending on who you are, that may be a good or bad thing.
Enlarge/ Someone just asked what it would look like if their girlfriend was a Smurf. Better add another rack of servers! (credit: Getty Images)
Late last week, both Bloomberg and The Washington Post published stories focused on the ostensibly disastrous impact artificial intelligence is having on the power grid and on efforts to collectively reduce our use of fossil fuels. The high-profile pieces lean heavily on recent projections from Goldman Sachs and the International Energy Agency (IEA) to cast AI's "insatiable" demand for energy as an almost apocalyptic threat to our power infrastructure. The Post piece even cites anonymous "some [people]" in reporting that "some worry whether there will be enough electricity to meet [the power demands] from any source."
Digging into the best available numbers and projections available, though, it's hard to see AI's current and near-future environmental impact in such a dire light. While generative AI models and tools can and will use a significant amount of energy, we shouldn't conflate AI energy usage with the larger and largely pre-existing energy usage of "data centers" as a whole. And just like any technology, whether that AI energy use is worthwhile depends largely on your wider opinion of the value of generative AI in the first place.
Not all data centers
While the headline focus of both Bloomberg and The Washington Post's recent pieces is on artificial intelligence, the actual numbers and projections cited in both pieces overwhelmingly focus on the energy used by Internet "data centers" as a whole. Long before generative AI became the current Silicon Valley buzzword, those data centers were already growing immensely in size and energy usage, powering everything from Amazon Web Services servers to online gaming services, Zoom video calls, and cloud storage and retrieval for billions of documents and photos, to name just a few of the more common uses.
Artificial intelligence-generated “deepfakes” that impersonate politicians and celebrities are far more prevalent than efforts to use AI to assist cyber attacks, according to the first research by Google’s DeepMind division into the most common malicious uses of the cutting-edge technology.
The study said the creation of realistic but fake images, video, and audio of people was almost twice as common as the next highest misuse of generative AI tools: the falsifying of information using text-based tools, such as chatbots, to generate misinformation to post online.
The most common goal of actors misusing generative AI was to shape or influence public opinion, the analysis, conducted with the search group’s research and development unit Jigsaw, found. That accounted for 27 percent of uses, feeding into fears over how deepfakes might influence elections globally this year.
An anonymous reader shares a report: Apple rejected overtures by Meta Platforms to integrate the social networking company's AI chatbot into the iPhone months ago, according to people with knowledge of the matter. The two companies aren't in discussions about using Meta's Llama chatbot in an AI partnership and only held brief talks in March, said the people, who asked not to be identified because the situation is private. The dialogue about a partnership didn't reach any formal stage, and Apple has no active plans to integrate Llama.
[...] Apple decided not to move forward with formal Meta discussions in part because it doesn't see that company's privacy practices as stringent enough, according to the people. Apple has spent years criticizing Meta's technology, and integrating Llama into the iPhone would have been a stark about-face.
Enlarge/ Michael Jackson in concert, 1986. Sony Music owns a large portion of publishing rights to Jackson's music. (credit: Getty Images)
Universal Music Group, Sony Music, and Warner Records have sued AI music-synthesis companies Udio and Suno for allegedly committing mass copyright infringement by using recordings owned by the labels to train music-generating AI models, reports Reuters. Udio and Suno can generate novel song recordings based on text-based descriptions of music (i.e., "a dubstep song about Linus Torvalds").
The lawsuits, filed in federal courts in New York and Massachusetts, claim that the AI companies' use of copyrighted material to train their systems could lead to AI-generated music that directly competes with and potentially devalues the work of human artists.
Like other generative AI models, both Udio and Suno (which we coveredseparately in April) rely on a broad selection of existing human-created artworks that teach a neural network the relationship between words in a written prompt and styles of music. The record labels correctly note that these companies have been deliberately vague about the sources of their training data.
French universities are becoming hotbeds for AI innovation, attracting investors seeking the next tech breakthrough. Ecole Polytechnique, a 230-year-old institution near Paris, stands out with 57% of France's AI startup founders among its alumni, according to Dealroom data analyzed by Accel. The school's approach combines STEM education with humanities and military training, producing well-rounded entrepreneurs. "AI is now instilling every discipline the same way mathematics did years ago," said Dominique Rossin, the school's provost. "We really push our students out of their comfort zone and encourage them to try new subjects and discover new areas in science," he added.
France leads Europe in AI startup funding, securing $2.3 billion and outpacing the UK and Germany, according to Dealroom.
Earlier this month Apple announced a partnership with OpenAI to bring ChatGPT to Siri.
"Now, the Wall Street Journal reports that Apple and Facebook's parent company Meta are in talks around a similar deal," according to TechCrunch:
A deal with Meta could make Apple less reliant on a single partner, while also providing validation for Meta's generative AI tech. The Journal reports that Apple isn't offering to pay for these partnerships; instead, Apple provides distribution to AI partners who can then sell premium subscriptions... Apple has said it will ask for users' permission before sharing any questions and data with ChatGPT. Presumably, any integration with Meta would work similarly.
OpenAI's 'Media Manager' Mocked, Amid Accusations of Robbing Creative Professionals
"Amid the hype surrounding Apple's new deal with OpenAI, one issue has been largely papered over," argues the Executive Director of America's writer's advocacy group, the Authors Guild.
OpenAI's foundational models "are, and have always been, built atop the theft of creative professionals' work."
[L]ast month the company quietly announced Media Manager, scheduled for release in 2025. A tool purportedly designed to allow creators and content owners to control how their work is used, Media Manager is really a shameless attempt to evade responsibility for the theft of artists' intellectual property that OpenAI is already profiting from.
OpenAI says this tool would allow creators to identify their work and choose whether to exclude it from AI training processes. But this does nothing to address the fact that the company built its foundational models using authors' and other creators' works without consent, compensation or control over how OpenAI users will be able to imitate the artists' styles to create new works. As it's described, Media Manager puts the burden on creators to protect their work and fails to address the company's past legal and ethical transgressions. This overture is like having your valuables stolen from your home and then hearing the thief say, "Don't worry, I'll give you a chance to opt out of future burglaries ... next year...."
AI companies often argue that it would be impossible for them to license all the content that they need and that doing so would bring progress to a grinding halt. This is simply untrue. OpenAI has signed a succession of licensing agreements with publishers large and small. While the exact terms of these agreements are rarely released to the public, the compensation estimates pale in comparison with the vast outlays for computing power and energy that the company readily spends. Payments to authors would have minimal effects on AI companies' war chests, but receiving royalties for AI training use would be a meaningful new revenue stream for a profession that's already suffering...
We cannot trust tech companies that swear their innovations are so important that they do not need to pay for one of the main ingredients — other people's creative works. The "better future" we are being sold by OpenAI and others is, in fact, a dystopia. It's time for creative professionals to stand together, demand what we are owed and determine our own futures.
The Authors Guild (and 17 other plaintiffs) are now in an ongoing lawsuit against OpenAI and Microsoft. And the Guild's executive director also notes that there's also "a class action filed by visual artists against Stability AI, Runway AI, Midjourney and Deviant Art, a lawsuit by music publishers against Anthropic for infringement of song lyrics, and suits in the U.S. and U.K. brought by Getty Images against Stability AI for copyright infringement of photographs."
They conclude that "The best chance for the wider community of artists is to band together."
The Roddenberry Foundation — named for Star Trek creator Gene Roddenberry — "announced Tuesday that this year's biennial award would focus on artificial intelligence that benefits humanity," reports the Los Angeles Times:
Lior Ipp, chief executive of the foundation, told The Times there's a growing recognition that AI is becoming more ubiquitous and will affect all aspects of our lives. "We are trying to ... catalyze folks to think about what AI looks like if it's used for good," Ipp said, "and what it means to use AI responsibly, ethically and toward solving some of the thorny global challenges that exist in the world...."
Ipp said the foundation shares the broad concern about AI and sees the award as a means to potentially contribute to creating those guardrails... Inspiration for the theme was also borne out of the applications the foundation received last time around. Ipp said the prize, which is "issue-agnostic" but focused on early-stage tech, produced compelling uses of AI and machine learning in agriculture, healthcare, biotech and education. "So," he said, "we sort of decided to double down this year on specifically AI and machine learning...."
Though the foundation isn't prioritizing a particular issue, the application states that it is looking for ideas that have the potential to push the needle on one or more of the United Nations' 17 sustainable development goals, which include eliminating poverty and hunger as well as boosting climate action and protecting life on land and underwater.
The Foundation's most recent winner was Sweden-based Elypta, according to the article, "which Ipp said is using liquid biopsies, such as a blood test, to detect cancer early."
"We believe that building a better future requires a spirit of curiosity, a willingness to push boundaries, and the courage to think big," said Rod Roddenberry, co-founder of the Roddenberry Foundation. "The Prize will provide a significant boost to AI pioneers leading these efforts." According to the Foundation's announcement, the Prize "embodies the Roddenberry philosophy's promise of a future in which technology and human ingenuity enable everyone — regardless of background — to thrive."
"By empowering entrepreneurs to dream bigger and innovate valiantly, the Roddenberry Prize seeks to catalyze the development of AI solutions that promote abundance and well-being for all."
"University of Zurich researchers have discovered that our brains process natural human voices and "deepfake" voices differently," writes Slashdot reader jenningsthecat.
From the University's announcement:
The researchers first used psychoacoustical methods to test how well human voice identity is preserved in deepfake voices. To do this, they recorded the voices of four male speakers and then used a conversion algorithm to generate deepfake voices. In the main experiment, 25 participants listened to multiple voices and were asked to decide whether or not the identities of two voices were the same. Participants either had to match the identity of two natural voices, or of one natural and one deepfake voice.
The deepfakes were correctly identified in two thirds of cases. "This illustrates that current deepfake voices might not perfectly mimic an identity, but do have the potential to deceive people," says Claudia Roswandowitz, first author and a postdoc at the Department of Computational Linguistics.
The researchers then used imaging techniques to examine which brain regions responded differently to deepfake voices compared to natural voices. They successfully identified two regions that were able to recognize the fake voices: the nucleus accumbens and the auditory cortex. "The nucleus accumbens is a crucial part of the brain's reward system. It was less active when participants were tasked with matching the identity between deepfakes and natural voices," says Claudia Roswandowitz. In contrast, the nucleus accumbens showed much more activity when it came to comparing two natural voices.
The complete paper appears in Nature.
Multiple AI companies are ignoring Robots.txt files meant to block the scraping of web content for generative AI systems, reports Reuters — citing a warning sent to publisher by content licensing startup TollBit.
TollBit, an early-stage startup, is positioning itself as a matchmaker between content-hungry AI companies and publishers open to striking licensing deals with them. The company tracks AI traffic to the publishers' websites and uses analytics to help both sides settle on fees to be paid for the use of different types of content... It says it had 50 websites live as of May, though it has not named them. According to the TollBit letter, Perplexity is not the only offender that appears to be ignoring robots.txt. TollBit said its analytics indicate "numerous" AI agents are bypassing the protocol, a standard tool used by publishers to indicate which parts of its site can be crawled.
"What this means in practical terms is that AI agents from multiple sources (not just one company) are opting to bypass the robots.txt protocol to retrieve content from sites," TollBit wrote. "The more publisher logs we ingest, the more this pattern emerges."
The article includes this quote from the president of the News Media Alliance (a trade group representing over 2,200 U.S.-based publishers). "Without the ability to opt out of massive scraping, we cannot monetize our valuable content and pay journalists. This could seriously harm our industry."
Reuters also notes another threat facing news sites:
Publishers have been raising the alarm about news summaries in particular since Google rolled out a product last year that uses AI to create summaries in response to some search queries. If publishers want to prevent their content from being used by Google's AI to help generate those summaries, they must use the same tool that would also prevent them from appearing in Google search results, rendering them virtually invisible on the web.
Slashdot reader DevNull127 writes:
A free and open source ChatGPT clone — named LibreChat — lets its users choose which AI model to use, "to harness the capabilities of cutting-edge language models from multiple providers in a unified interface". This means LibreChat includes OpenAI's models, but also others — both open-source and closed-source — and its website promises "seamless integration" with AI services from OpenAI, Azure, Anthropic, and Google — as well as GPT-4, Gemini Vision, and many others. ("Every AI in one place," explains LibreChat's home page.) Plugins even let you make requests to DALL-E or Stable Diffusion for image generations. (LibreChat also offers a database that tracks "conversation state" — making it possible to switch to a different AI model in mid-conversation...)
Released under the MIT License, LibreChat has become "an open source success story," according to this article, representing "the passionate community that's actively creating an ecosystem of open source AI tools." And its creator, Danny Avila, says in some cases it finally lets users own their own data, "which is a dying human right, a luxury in the internet age and even more so with the age of LLM's." Avila says he was inspired by the day ChatGPT leaked the chat history of some of its users back in March of 2023 — and LibreChat is "inherently completely private".
From the article:
With locally-hosted LLMs, Avila sees users finally getting "an opportunity to withhold training data from Big Tech, which many trade at the cost of convenience." In this world, LibreChat "is naturally attractive as it can run exclusively on open-source technologies, database and all, completely 'air-gapped.'" Even with remote AI services insisting they won't use transient data for training, "local models are already quite capable" Avila notes, "and will become more capable in general over time."
And they're also compatible with LibreChat...
The artificial intelligence revolution will demand more electricity, warns the Washington Post. "Much more..."
They warn that the "voracious" electricity consumption of AI is driving an expansion of fossil fuel use in America — "including delaying the retirement of some coal-fired plants."
As the tech giants compete in a global AI arms race, a frenzy of data center construction is sweeping the country. Some computing campuses require as much energy as a modest-sized city, turning tech firms that promised to lead the way into a clean energy future into some of the world's most insatiable guzzlers of power. Their projected energy needs are so huge, some worry whether there will be enough electricity to meet them from any source... A ChatGPT-powered search, according to the International Energy Agency, consumes almost 10 times the amount of electricity as a search on Google. One large data center complex in Iowa owned by Meta burns the annual equivalent amount of power as 7 million laptops running eight hours every day, based on data shared publicly by the company...
[Tech companies] argue advancing AI now could prove more beneficial to the environment than curbing electricity consumption. They say AI is already being harnessed to make the power grid smarter, speed up innovation of new nuclear technologies and track emissions.... "If we work together, we can unlock AI's game-changing abilities to help create the net zero, climate resilient and nature positive works that we so urgently need," Microsoft said in a statement.
The tech giants say they buy enough wind, solar or geothermal power every time a big data center comes online to cancel out its emissions. But critics see a shell game with these contracts: The companies are operating off the same power grid as everyone else, while claiming for themselves much of the finite amount of green energy. Utilities are then backfilling those purchases with fossil fuel expansions, regulatory filings show... heavily polluting fossil fuel plants that become necessary to stabilize the power grid overall because of these purchases, making sure everyone has enough electricity.
The article quotes a project director at the nonprofit Data & Society, which tracks the effect of AI and accuses the tech industry of using "fuzzy math" in its climate claims. "Coal plants are being reinvigorated because of the AI boom," they tell the Washington Post. "This should be alarming to anyone who cares about the environment."
The article also summarzies a recent Goldman Sachs analysis, which predicted data centers would use 8% of America's total electricity by 2030, with 60% of that usage coming "from a vast expansion in the burning of natural gas. The new emissions created would be comparable to that of putting 15.7 million additional gas-powered cars on the road."
"We all want to be cleaner," Brian Bird, president of NorthWestern Energy, a utility serving Montana, South Dakota and Nebraska, told a recent gathering of data center executives in Washington, D.C. "But you guys aren't going to wait 10 years ... My only choice today, other than keeping coal plants open longer than all of us want, is natural gas. And so you're going see a lot of natural gas build out in this country."
Big Tech responded by "going all in on experimental clean-energy projects that have long odds of success anytime soon," the article concludes. "In addition to fusion, they are hoping to generate power through such futuristic schemes as small nuclear reactors hooked to individual computing centers and machinery that taps geothermal energy by boring 10,000 feet into the Earth's crust..."
Some experts point to these developments in arguing the electricity needs of the tech companies will speed up the energy transition away from fossil fuels rather than undermine it. "Companies like this that make aggressive climate commitments have historically accelerated deployment of clean electricity," said Melissa Lott, a professor at the Climate School at Columbia University.
Slashdot reader DevNull127 writes:
A free and open source ChatGPT clone — named LibreChat — is also letting its users choose which AI model to use, "to harness the capabilities of cutting-edge language models from multiple providers in a unified interface". This means LibreChat includes OpenAI's models, but also others — both open-source and closed-source — and its website promises "seamless integration" with AI services from OpenAI, Azure, Anthropic, and Google — as well as GPT-4, Gemini Vision, and many others. ("Every AI in one place," explains LibreChat's home page.) Plugins even let you make requests to DALL-E or Stable Diffusion for image generations. (LibreChat also offers a database that tracks "conversation state" — making it possible to switch to a different AI model in mid-conversation...)
Released under the MIT License, LibreChat has become "an open source success story," according to this article, representing "the passionate community that's actively creating an ecosystem of open source AI tools." Its creator, Danny Avila, says it finally lets users own their own data, "which is a dying human right, a luxury in the internet age and even more so with the age of LLM's." Avila says he was inspired by the day ChatGPT leaked the chat history of some of its users back in March of 2023 — and LibreChat is "inherently completely private". From the article:
With locally-hosted LLMs, Avila sees users finally getting "an opportunity to withhold training data from Big Tech, which many trade at the cost of convenience." In this world, LibreChat "is naturally attractive as it can run exclusively on open-source technologies, database and all, completely 'air-gapped.'" Even with remote AI services insisting they won't use transient data for training, "local models are already quite capable" Avila notes, "and will become more capable in general over time."
And they're also compatible with LibreChat...
OpenAI CTO Mira Murati isn't worried about how AI could hurt some creative jobs, suggesting during a talk that some jobs were maybe always a bit replaceable anyway. From a report: "I think it's really going to be a collaborative tool, especially in the creative spaces," Murati told Darmouth University Trustee Jeffrey Blackburn during a conversation about AI hosted at the university's engineering department. "Some creative jobs maybe will go away, but maybe they shouldn't have been there in the first place," the CTO said of AI's role in the workplace. "I really believe that using it as a tool for education, [and] creativity, will expand our intelligence."
Enlarge/ Features like Image Playground won't arrive in Europe at the same time as other regions. (credit: Apple)
Three major features in iOS 18 and macOS Sequoia will not be available to European users this fall, Apple says. They include iPhone screen mirroring on the Mac, SharePlay screen sharing, and the entire Apple Intelligence suite of generative AI features.
In a statement sent to Financial Times, The Verge, and others, Apple says this decision is related to the European Union's Digital Markets Act (DMA). Here's the full statement, which was attributed to Apple spokesperson Fred Sainz:
Two weeks ago, Apple unveiled hundreds of new features that we are excited to bring to our users around the world. We are highly motivated to make these technologies accessible to all users. However, due to the regulatory uncertainties brought about by the Digital Markets Act (DMA), we do not believe that we will be able to roll out three of these features — iPhone Mirroring, SharePlay Screen Sharing enhancements, and Apple Intelligence — to our EU users this year.
Specifically, we are concerned that the interoperability requirements of the DMA could force us to compromise the integrity of our products in ways that risk user privacy and data security. We are committed to collaborating with the European Commission in an attempt to find a solution that would enable us to deliver these features to our EU customers without compromising their safety.
It is unclear from Apple's statement precisely which aspects of the DMA may have led to this decision. It could be that Apple is concerned that it would be required to give competitors like Microsoft or Google access to user data collected for Apple Intelligence features and beyond, but we're not sure.
An anonymous reader shares a report: Microsoft launched its range of Copilot Plus PCs earlier this week, and they all come equipped with the new dedicated Copilot key on the keyboard. It's the first big change to Windows keyboards in 30 years, but all the key does now is launch a Progressive Web App (PWA) version of Copilot. The web app doesn't even integrate into Windows anymore like the previous Copilot experience did since last year, so you can't use Copilot to control Windows 11 settings or have it docked as a sidebar anymore. It's literally just a PWA. Microsoft has even removed the keyboard shortcut to Copilot on these new Copilot Plus PCs, so WINKEY + C does nothing.
Amazon is planning a major revamp of its decade-old money-losing Alexa service to include a conversational generative AI with two tiers of service and has considered a monthly fee of around $5 to access the superior version, Reuters reported Friday, citing people with direct knowledge of the company's plans. From the report: Known internally as "Banyan," a reference to the sprawling ficus trees, the project would represent the first major overhaul of the voice assistant since it was introduced in 2014 along with the Echo line of speakers. Amazon has dubbed the new voice assistant "Remarkable Alexa," the people said. Amazon has also considered a roughly $10-per-month price, the report added.
On Thursday, Anthropic announced Claude 3.5 Sonnet, its latest AI language model and the first in a new series of "3.5" models that build upon Claude 3, launched in March. Claude 3.5 can compose text, analyze data, and write code. It features a 200,000 token context window and is available now on the Claude website and through an API. Anthropic also introduced Artifacts, a new feature in the Claude interface that shows related work documents in a dedicated window.
So far, people outside of Anthropic seem impressed. "This model is really, really good," wrote independent AI researcher Simon Willison on X. "I think this is the new best overall model (and both faster and half the price of Opus, similar to the GPT-4 Turbo to GPT-4o jump)."
As we've written before, benchmarks for large language models (LLMs) are troublesome because they can be cherry-picked and often do not capture the feel and nuance of using a machine to generate outputs on almost any conceivable topic. But according to Anthropic, Claude 3.5 Sonnet matches or outperforms competitor models like GPT-4o and Gemini 1.5 Pro on certain benchmarks like MMLU (undergraduate level knowledge), GSM8K (grade school math), and HumanEval (coding).
Google introduced AI Overviews in search results shortly after Google I/O in May, but it wasn’t first to the AI search game. It had already given Gemini the ability to search the internet, and Meta and other competing AI companies had done similarly with their own models. One of the biggest players in this field was Perplexity, which markets itself as a “conversational search engine”—basically another chatbot with internet access, but with even more of a focus on summaries and current events. Unfortunately, Perplexity is now finding itself in hot water after breaking rules and, like Google, returning wrong answer after wrong answer.
On June 11, Forbes published an article accusing Perplexity of stealing its content for quickly rewriting original articles without sourcing, and passing them off as its own. The AI company went as fair as to adapt Forbes’ reporting to podcast form. Shortly after, Wired ran an exposé on Perplexity, accusing it of “bullshitting” and breaking a widely held internet rule (more on that shortly). Now, we’re learning a lot more about what kind of recent data an AI might be able to train on going forward, and why AIs often make so many mistakes when trying to sum up current events.
Perplexity is accused of breaking a longstanding internet rule
Bots aren’t anything new on the internet. Before AI scraped websites for training material, search engines scraped websites to determine where to place them in search results. This led to a standard called the Robots Exclusion Protocol, which allows developers to lay out which parts of their site they don’t want bots to access. Perplexity says it follows this rule, but, spurred on by the Forbes story and an accusation of rule breaking from developer Robb Knight, Wired conducted its own investigation. What it discovered wasn't flattering to Perplexity.
“Wired provided the Perplexity chatbot with the headlines of dozens of articles published on our website this year, as well as prompts about the subjects of Wired reporting,” Wired’s article reads. According to the investigation, the bot then returned answers “closely paraphrasing Wired stories,” complete with original Wired art. Further, it would summarize stories “inaccurately and with minimal attribution.”
Examples include the chatbot inaccurately accusing a police officer of stealing bicycles, and, in a test, responding to a request to summarize a webpage containing a single sentence with a wholly invented story about a young girl going on a fairy tale adventure. Wired concluded Perplexity’s summaries were the result of the AI flagrantly breaking the Robots Exclusion Protocol, and that its inaccuracies likely stemmed from an attempt to sidestep said rule.
According to both Knight and Wired, when users ask Perplexity questions that would require the bot to summarize an article protected by the Robots Exclusion Protocol, a specific IP address running what is assumed to be an automated web browser would access the websites bots are not supposed to scrape. The IP address couldn’t be tracked back to Perplexity with complete certainty, but its frequent association with the service raised suspicions.
In other cases, Wired recognized traces of its metadata in Perplexity’s responses, which could mean the bot may not be reading articles themselves, but accessing traces of it left in URLs and search engines. These wouldn’t be protected by the Robots Exclusion Protocol, but are so light on information that they’re more likely to lead to AI hallucinations—hence the problem with misinformation in AI search results.
Both of these issues presage a battle for the future of AI in search engines, from both ethical and technical standpoints. Even as artists and other creators argue over AI’s right to scrape older works, accessing writing that is just a few days old puts Perplexity at further legal risk.
Perplexity CEO Aravind Srinivas issued a statement to Wired that said “the questions from Wired reflect a deep and fundamental misunderstanding of how Perplexity and the Internet work.” At the same time, Forbes this week reportedly sent Perplexity a letter threatening legal action over “willful infringement” of its copyrights.
It’s only been a few months since Anthropic debuted Claude 3, but the company is ready to take the next step—at least for one of its models. Enter Claude 3.5 Sonnet. As the middle-ground for Anthropic’s large language models (LLMs), Claude Sonnet is a good option for those who want access to a powerful but affordable AI chatbot, and with Claude 3.5 Sonnet, the company says it's making its middle offering even better.
According to the announcement, Claude 3.5 Sonnet is up to two times faster at processing than Claude 3 Opus, previously the most powerful model the company offered (Opus will be getting an update to take back its top spot). Anthropic claims that Claude 3.5 is “ideal for complex tasks,” and that it shows improvement in writing with nuance, humor, and following complex instructions. Claude 3.5 reportedly solved 64 percent of the problems it was given, outperforming the 38 percent record previously set by Claude 3 Opus. That is, indeed, a marked improvement.
Credit: Anthropic
You can see Anthropic's full list of how Claude 3.5 Sonnet compares to other LLMs across different areas in the image above. Based on the data shown, it appears to outperform OpenAI’s newest model, GPT-4o, in almost every category. However, exactly how well those benchmarks will play out in real-world usage remains to be seen.
Coinciding with the launch of Claude 3.5 Sonnet is Artifacts, a new feature that essentially creates a separate window in your Claude window that can showcase your documents, code, and other AI-generated content in a visual space in real time. Anthropic says that this will make collaborating through Claude much easier for teams. Eventually, it hopes to allow entire organizations to use Claude to securely centralize its knowledge in one shared space and then access it through the chatbot. This will likely be similar to what Google has been doing with its Gemini AI offerings in Google Workspace.
Anthropic isn’t stopping here, either. The AI company says it plans to release updated versions of the Opus and Haiku LLMs later this year. The company also noted that it is exploring features like Memory, which would allow Claude to remember a user’s specific preferences and interaction history to help make their experiences even more personal. ChatGPT already utilizes a memory system, so it isn’t surprising to see Anthropic leaning that way with Claude, too.
If you’re interested in trying Claude 3.5 Sonnet for yourself, you can access it directly from the Claude website or through the Claude iOS app. Claude Pro and Team subscribers will still get access to higher rates with the new model. Developers can also utilize Anthropic’s API directly, though they’ll need to pay for tokens to do so.
It's one of the world's worst-kept secrets that large language models give blatantly false answers to queries and do so with a confidence that's indistinguishable from when they get things right. There are a number of reasons for this. The AI could have been trained on misinformation; the answer could require some extrapolation from facts that the LLM isn't capable of; or some aspect of the LLM's training might have incentivized a falsehood.
But perhaps the simplest explanation is that an LLM doesn't recognize what constitutes a correct answer but is compelled to provide one. So it simply makes something up, a habit that has been termed confabulation.
Figuring out when an LLM is making something up would obviously have tremendous value, given how quickly people have started relying on them for everything from college essays to job applications. Now, researchers from the University of Oxford say they've found a relatively simple way to determine when LLMs appear to be confabulating that works with all popular models and across a broad range of subjects. And, in doing so, they develop evidence that most of the alternative facts LLMs provide are a product of confabulation.
A cinema in London has cancelled the world premiere of a film with a script generated by AI after a backlash. From a report: The Prince Charles cinema, located in London's West End and which traditionally screens cult and art films, was due to host a showing of a new production called The Last Screenwriter on Sunday. However the cinema announced on social media that the screening would not go ahead. In its statement the Prince Charles said: "The feedback we received over the last 24hrs once we advertised the film has highlighted the strong concern held by many of our audience on the use of AI in place of a writer which speaks to a wider issue within the industry."
Directed by Peter Luisi and starring Nicholas Pople, The Last Screenwriter is a Swiss production that describes itself as the story of "a celebrated screenwriter" who "finds his world shaken when he encounters a cutting edge AI scriptwriting system ... he soon realises AI not only matches his skills but even surpasses him in empathy and understanding of human emotions." The screenplay is credited to "ChatGPT 4.0." OpenAI launched its latest model, GPT-4o, in May. Luisi told the Daily Beast that the cinema had cancelled the screening after it received 200 complaints, but that a private screening for cast and crew would still go ahead in London.
Enlarge/ Ilya Sutskever physically gestures as OpenAI CEO Sam Altman looks on at Tel Aviv University on June 5, 2023. (credit: Getty Images)
On Wednesday, former OpenAI Chief Scientist Ilya Sutskever announced he is forming a new company called Safe Superintelligence, Inc. (SSI) with the goal of safely building "superintelligence," which is a hypothetical form of artificial intelligence that surpasses human intelligence, possibly in the extreme.
"We will pursue safe superintelligence in a straight shot, with one focus, one goal, and one product," wrote Sutskever on X. "We will do it through revolutionary breakthroughs produced by a small cracked team."
Sutskever was a founding member of OpenAI and formerly served as the company's chief scientist. Two others are joining Sutskever at SSI initially: Daniel Levy, who formerly headed the Optimization Team at OpenAI, and Daniel Gross, an AI investor who worked on machine learning projects at Apple between 2013 and 2017. The trio posted a statement on the company's new website.
Anthropic launched Claude 3.5 Sonnet on Thursday, claiming it outperforms previous models and OpenAI's GPT-4 Omni. The AI startup also introduced Artifacts, a workspace for users to edit AI-generated projects. This release, part of the Claude 3.5 family, follows three months after Claude 3. Claude 3.5 Sonnet is available for free on Claude.ai and the Claude iOS app, while Claude Pro and Team plan subscribers can access it with significantly higher rate limits.
Anthropic plans to launch 3.5 versions of Haiku and Opus later this year, exploring features like web search and memory for future releases.
Anthropic also introduced Artifacts on Claude.ai, a new feature that expands how users can interact with Claude. When a user asks Claude to generate content like code snippets, text documents, or website designs, these Artifacts appear in a dedicated window alongside their conversation. This creates a dynamic workspace where they can see, edit, and build upon Claude's creations in real-time, seamlessly integrating AI-generated content into their projects and workflows, the startup said.
Perplexity AI, a billion-dollar "AI" search startup, has come under scrutiny for its data collection practices and accuracy of its chatbot responses. Despite claiming to respect website operators' wishes, Perplexity appears to scrape content from sites that have blocked its crawler, using an undisclosed IP address, a Wired investigation found. The chatbot also generates summaries that closely paraphrase original reporting with minimal attribution. Furthermore, its AI often "hallucinates," inventing false information when unable to access articles directly. Perplexity's CEO, Aravind Srinivas, maintains the company is not acting unethically.
Ilya Sutskever, co-founder of OpenAI who recently left the startup, has launched a new venture called Safe Superintelligence Inc., aiming to create a powerful AI system within a pure research organization. Sutskever has made AI safety the top priority for his new company. Safe Superintelligence has two more co-founders: investor and former Apple AI lead Daniel Gross, and Daniel Levy, known for training large AI models at OpenAI. From a report: Researchers and intellectuals have contemplated making AI systems safer for decades, but deep engineering around these problems has been in short supply. The current state of the art is to use both humans and AI to steer the software in a direction aligned with humanity's best interests. Exactly how one would stop an AI system from running amok remains a largely philosophical exercise.
Sutskever says that he's spent years contemplating the safety problems and that he already has a few approaches in mind. But Safe Superintelligence isn't yet discussing specifics. "At the most basic level, safe superintelligence should have the property that it will not harm humanity at a large scale," Sutskever says. "After this, we can say we would like it to be a force for good. We would like to be operating on top of some key values. Some of the values we were thinking about are maybe the values that have been so successful in the past few hundred years that underpin liberal democracies, like liberty, democracy, freedom."
Sutskever says that the large language models that have dominated AI will play an important role within Safe Superintelligence but that it's aiming for something far more powerful. With current systems, he says, "you talk to it, you have a conversation, and you're done." The system he wants to pursue would be more general-purpose and expansive in its abilities. "You're talking about a giant super data center that's autonomously developing technology. That's crazy, right? It's the safety of that that we want to contribute to."
Shubham Sharma reports via VentureBeat: Chinese AI startup DeepSeek, which previously made headlines with a ChatGPT competitor trained on 2 trillion English and Chinese tokens, has announced the release of DeepSeek Coder V2, an open-source mixture of experts (MoE) code language model. Built upon DeepSeek-V2, an MoE model that debuted last month, DeepSeek Coder V2 excels at both coding and math tasks. It supports more than 300 programming languages and outperforms state-of-the-art closed-source models, including GPT-4 Turbo, Claude 3 Opus and Gemini 1.5 Pro. The company claims this is the first time an open model has achieved this feat, sitting way ahead of Llama 3-70B and other models in the category. It also notes that DeepSeek Coder V2 maintains comparable performance in terms of general reasoning and language capabilities.
Founded last year with a mission to "unravel the mystery of AGI with curiosity," DeepSeek has been a notable Chinese player in the AI race, joining the likes of Qwen, 01.AI and Baidu. In fact, within a year of its launch, the company has already open-sourced a bunch of models, including the DeepSeek Coder family. The original DeepSeek Coder, with up to 33 billion parameters, did decently on benchmarks with capabilities like project-level code completion and infilling, but only supported 86 programming languages and a context window of 16K. The new V2 offering builds on that work, expanding language support to 338 and context window to 128K -- enabling it to handle more complex and extensive coding tasks. When tested on MBPP+, HumanEval, and Aider benchmarks, designed to evaluate code generation, editing and problem-solving capabilities of LLMs, DeepSeek Coder V2 scored 76.2, 90.2, and 73.7, respectively -- sitting ahead of most closed and open-source models, including GPT-4 Turbo, Claude 3 Opus, Gemini 1.5 Pro, Codestral and Llama-3 70B. Similar performance was seen across benchmarks designed to assess the model's mathematical capabilities (MATH and GSM8K). The only model that managed to outperform DeepSeek's offering across multiple benchmarks was GPT-4o, which obtained marginally higher scores in HumanEval, LiveCode Bench, MATH and GSM8K. [...]
As of now, DeepSeek Coder V2 is being offered under a MIT license, which allows for both research and unrestricted commercial use. Users can download both 16B and 236B sizes in instruct and base avatars via Hugging Face. Alternatively, the company is also providing access to the models via API through its platform under a pay-as-you-go model. For those who want to test out the capabilities of the models first, the company is offering the option to interact. with Deepseek Coder V2 via chatbot.
An anonymous reader quotes a report from MIT Technology Review: Meta has created a system that can embed hidden signals, known as watermarks, in AI-generated audio clips, which could help in detecting AI-generated content online. The tool, called AudioSeal, is the first that can pinpoint which bits of audio in, for example, a full hourlong podcast might have been generated by AI. It could help to tackle the growing problem of misinformation and scams using voice cloning tools, says Hady Elsahar, a research scientist at Meta. Malicious actors have used generative AI to create audio deepfakes of President Joe Biden, and scammers have used deepfakes to blackmail their victims. Watermarks could in theory help social media companies detect and remove unwanted content. However, there are some big caveats. Meta says it has no plans yet to apply the watermarks to AI-generated audio created using its tools. Audio watermarks are not yet adopted widely, and there is no single agreed industry standard for them. And watermarks for AI-generated content tend to be easy to tamper with -- for example, by removing or forging them.
Fast detection, and the ability to pinpoint which elements of an audio file are AI-generated, will be critical to making the system useful, says Elsahar. He says the team achieved between 90% and 100% accuracy in detecting the watermarks, much better results than in previous attempts at watermarking audio. AudioSeal is available on GitHub for free. Anyone can download it and use it to add watermarks to AI-generated audio clips. It could eventually be overlaid on top of AI audio generation models, so that it is automatically applied to any speech generated using them. The researchers who created it will present their work at the International Conference on Machine Learning in Vienna, Austria, in July.
Enlarge/ Screen capture of a Runway Gen-3 Alpha video generated with the prompt "A giant humanoid, made of fluffy blue cotton candy, stomping on the ground, and roaring to the sky, clear blue sky behind them." (credit: Runway)
On Sunday, Runway announced a new AI video synthesis model called Gen-3 Alpha that's still under development, but it appears to create video of similar quality to OpenAI's Sora, which debuted earlier this year (and has also not yet been released). It can generate novel, high-definition video from text prompts that range from realistic humans to surrealistic monsters stomping the countryside.
Unlike Runway's previous best model from June 2023, which could only create two-second-long clips, Gen-3 Alpha can reportedly create 10-second-long video segments of people, places, and things that have a consistency and coherency that easily surpasses Gen-2. If 10 seconds sounds short compared to Sora's full minute of video, consider that the company is working with a shoestring budget of compute compared to more lavishly funded OpenAI—and actually has a history of shipping video generation capability to commercial users.
Gen-3 Alpha does not generate audio to accompany the video clips, and it's highly likely that temporally coherent generations (those that keep a character consistent over time) are dependent on similar high-quality training material. But Runway's improvement in visual fidelity over the past year is difficult to ignore.
For the vast majority of compatible PCs, Microsoft’s Windows 11 24H2 update still isn’t officially available as anything other than a preview (a revised version of the update is available to Windows Insiders again after briefly being pulled early last week). But Microsoft and most of the other big PC companies are releasing their first wave of Copilot+ PCs with Snapdragon X-series chips in them today, and those PCs are all shipping with the 24H2 update already installed.
For now, this means a bifurcated Windows 11 install base: one (the vast majority) that’s still mostly on version 23H2 and one (a tiny, Arm-powered minority) that’s running 24H2.
Although Microsoft hasn’t been specific about its release plans for Windows 11 24H2 to the wider user base, most PCs should still start getting the update later this fall. The Copilot+ parts won’t run on those current PCs, but they’ll still get new features and benefit from Microsoft’s work on the operating system’s underpinnings.
Login
