I don't mean to be alarmist, but I do think it's time to start assuming everything you see online is fake. The internet is full of content produced by real people, of course (this article included). But AI-generated media is getting so realistic, that it almost puts you at a disadvantage to presume the content you're scrolling past on your feeds is legitimate. Don't skip this article because you know what AI content looks like—the current stuff your algorithm delivers to your social media feeds is easy to spot if you know what you're looking for. But even if you can identify AI slop the second it hits your eyeballs, you need to know you're not ready for the next wave of AI-generated videos. That wave isn't just on its way—it's already here. AI content is already fooling peopleMost of us are acutely aware of the "AI video" look: This "tragic" video of a cat parent saving their kitten by throwing it out of a burning airplane is obvious AI slop to most who watch it. You probably know Trump isn't working this construction site, and you most assuredly can understand this family of cat farmers is, in fact, AI-generated.But there are the videos that aren't so obvious, especially to those of us not quite so in tune with AI, or technology in general. You might know this video of babies dancing in a circle is AI, but plenty of the people in the comments didn't (assuming they aren't bots, either). You might also be able to tell that this family of pets isn't really watching a bird investigate a toy alligator, but, again, plenty can't. And there is no end to the America's Got Talent videos that feature "realistic" yet impossible visuals—that still capture the hearts of hundreds of thousands, if not millions of people. (I weep.)But I'm not writing this piece today because I'm concerned about how many of these "believable" AI videos are tricking way too many people into thinking they're real. I am worried about that, but those worries pale in comparison to my new fears.So far, most of the AI videos taking over social media feeds rely mainly on their visuals and background sounds to sell their alleged authenticity. You'll notice none of the characters in any of these videos actually speak. If they do, it's immediately off-putting, with out of sync lip movements and, typically, robotic voices. It's been easier for AI creators to put the emphasis on the realism of the people and animals in their videos, and hope you're wowed enough by a baby dancing with a lion to not think, "this is bullshit, right?"Even OpenAI's Sora video model, which shocked me with its quality in February of last year, was working off of its realistic visuals. A video of woman "filming" her reflection through a train window too real for comfort, but Sora wasn't spitting out fully-rendered conversations. If you see such a scene on your feeds, you probably assume, of course, it's a real video—or at least one generated by humans.AI video is about to change completelySomething happened this week that only made me more pessimistic about the future of truth on the internet. During this week's Google I/O event, Google unveiled Veo 3, its latest AI video model. Like other competitive models out there, Veo 3 can generate highly realistic sequences, which Google showed off throughout the presentation. Sure, not great, but also, nothing really new there. But Veo 3 isn't just capable of generating video that might trick your eye into thinking its real: Veo 3 can also generate audio to go alongside the video. That includes sound effects, but also dialogue—lip-synced dialogue. In order to demonstrate Veo 3's audio/video capabilities, Google showed off a clip of an old sailor at sea. The video quality is sharp and realistic, and the words the man speaks are synced to his lip movements. Of course, knowing the video is AI, you notice quirks that give away the game (to my eye, this looks like a high quality animation more than a live action shot) but I am quite confident this video would fool a lot of fans of fake AGT videos. But even this clip wasn't what inspired my newfound fears—it was the videos that users started making once they got their hands on Veo 3. PetaPixel has a great roundup of some of the "best" Veo 3 videos people have made so far, but I'll highlight some of the ones that should scare you most. This clip shows a streamer playing Fortnite. Everything, including the game footage, was generated with Google's AI: This Tweet is currently unavailable. It might be loading or has been removed. This clip shows three concerts that never happened, featuring musicians and crowds that do not exist. The music isn't good, but that's not the point. The music, from the vocals to the instrumentals, was generated entirely by the AI, and then synced to lips, drums, guitars, and strings: This Tweet is currently unavailable. It might be loading or has been removed. But this clip is, without a doubt, the one that should sound the alarm for each and every one of us. Someone generated a fake video of a fake car show, featuring fake interviews with fake attendees. It's far from perfect, but any AI quirks are totally overshadowed by the surface-level realism here. Not only would the AI's Got Talent fans buy this, I would buy this, especially if I wasn't looking out for it: This Tweet is currently unavailable. It might be loading or has been removed. It's the visuals; it's the dialogue; it's the crowds; it's the lighting; it's the candid laughter at "mistakes;" it's the sound of the mic being "bumped" into. Congratulations on noticing the dialogue often doesn't make sense, or that the people in the background defy the laws of physics—you won't notice it when it hits mid-scroll on TikTok or Instagram. Even Veo 2, which isn't as powerful as Veo 3, now offers tools for realism, like the ability to dictate how you want the camera to move. And both models are available in Flow, Google's AI video editor of sorts. Creators now have the ability to generate highly realistic AI content that feels like it was filmed in-person, and the tech is only getting better. This Tweet is currently unavailable. It might be loading or has been removed. Google's best AI video generator tools cost $250 a month through its new AI Ultra subscription plan. That's expensive, but not out of reach for plenty of people interested in making AI-generated content. But the $20 per month plan, AI Pro, still comes with Veo 2 and Flow access. The rate limits are lower, but I wouldn't be shocked to see some realistic slop come out of those limitations, too. It's time to be a full-time skepticNone of this tech is perfect. I'm not here to tell you that everything Veo 3 spits out is indistinguishable from real content, or that the videos are absent any of the usual AI tells. In fact, there's clearly something up with Veo 3's training data: As 404 Media reports, the model continuously generates the same weird "dad joke" whenever you ask for a generation of a comedian performing standup. What I'm saying is, it's time to turn on your bullshit detectors and keep them active full time. When engaging with videos on the internet—especially short-form algorithmic clips—you might be safer operating under the assumption the content is fake from the jump, and require proof beyond a reasonable doubt that what you're seeing wasn't generated with a simple prompt and a $250 budget. That feels extreme, but after what I've seen this week, I don't really see another way to engage with this content going forward.We're in scary territory now. Today, it's demos of musicians and streamers. Tomorrow, it's a politician saying something they didn't; a suspect committing the crime they're accused of; a "reporter" feeding you lies through the "news."I hope this is as good as the technology gets. I hope AI companies run out of training data to improve their models, and that governments take some action to regulate this technology. But seeing as the Republicans in the United States passed a bill that included a ban on state-enforced AI regulations for ten years, I'm pretty pessimistic on that latter point. In all likelihood, this tech is going to get better, with zero guardrails to ensure it advances safely. I'm left wondering how many of those politicians who voted yes on that bill watched an AI-generated video on their phone this week and thought nothing of it.

Back in my day, the phrase used to be “there’s an app for that,” and that’s still the case, though with one major amendment: now, it’s “there’s an AI app for that.” In fact, there’s even an AI app for making apps—buckle up, kiddos, things are about to get meta. Let me explain: Rork, which I stumbled across while scrolling X, is—if we are to drink the Kool-Aid—the app to end all apps. The font from which all other apps may flow. The cold fusion of coding. Alright, I’m exaggerating, but it’s exactly what I alluded to: an app that makes apps, which is like a hat on a hat if the first hat actually made the second hat. To make things even more meta, Rork used an AI ad with Google’s new Veo 3 video generator to promote its tool. Is your head spinning yet? Mine kind of is. When I say Rork makes apps, I mean it really makes the damn thing (at least I think it does since I wouldn’t know a functional piece of code if it sat on my chest and suffocated me like a sleep paralysis demon). But on the surface, it does the whole thing. I went to the web version of Rork to try it out (there’s no mobile app that I’m aware of), and it seemingly took my text prompt, “I want to make an app that matches me with similar-sized people in my area to fight. Like Tinder but for fisticuffs,” and ran with it. Once I punched the prompt in (pun intended), Rork got to work (thinking for a while as AI does) and then used its corresponding large language model (Anthropic’s Claude 4 model) to start drawing everything up. And I mean everything—colors, features, parameters, basically every aspect of an app that you might need to launch. And the conjuring doesn’t stop there. Once everything is devised, Rork’s interface splits everything off into packages if you want to look at the code (that is, if you’re capable of reading it, unlike me), and then it does my favorite part—it generates a usable preview that you can test on your phone or another device. After the AI had coded everything, I was able to scan a QR code and generate a preview using ExpoGo, a tool that lets you deploy code in a preview mode. So, without further ado, ladies and gentlemen, I present to you: FightMatch, Tinder for kicking ass. © Rork / Screenshot by Gizmodo It’s worth noting that I tried to make this even more meta by prompting Rork to make an app that uses generative AI to make images or video—an AI app that generates AI—but it ran into some issues that I wasn’t able to fully wrap my head around. Per Rork, they were “critical errors,” and even when I clicked the “fix” button, it wouldn’t budge. No AI app inception today, folks, sorry. On one hand, as someone with no coding experience, I’m impressed. Rork, as promised, was able to take my very simple text prompt (Tinder for fighting) and write up all the code to make it happen in about a minute or so. Again, a coder I am not, but that feels pretty extraordinary from a sheer idea to preview perspective. I’m fairly certain whatever Rork and Claude generated wouldn’t be enough to push to an app store right away, both from a technical and aesthetic perspective, but as a first draft, it’s at least serviceable, if very far from perfect. Also, if I’m being honest, I was looking for more of a Fight Club-type app over MMA, but I suppose Claude played this one safe. There’s obviously vast potential here to expedite app creation, but just like with every generative tool like this, there’s also potential for something less exciting—slop. Like I wrote earlier this week, tools like Google’s Veo 3 and Flow are impressive technical feats, but they also feel primed to further bloat an already overwhelming bucket of AI slop. There’s always that question: do we need more apps or do we need better apps? I’m a proponent of the latter philosophy, but if there’s one thing I’ve come to expect in the tech world, it’s more. But hey, if I get rich quick with FightMatch, I can’t really complain, can I? And if you disagree, swipe right, and let’s settle this the old-fashioned way.

Even at first glance, there’s something off about the body on the street. The white sheet it’s under is a little too clean, and the officers’ movements are totally devoid of purpose. “We need to clear the street,” one of them says with a firm hand gesture, though her lips don’t move. It’s AI, alright. But here’s the kicker: my prompt didn’t include any dialogue.Veo 3, Google’s new AI video generation model, added that line all on its own. Over the past 24 hours I’ve created a dozen clips depicting news reports, disasters, and goofy cartoon cats with convincing audio — some of which the model invented all on its own. It’s more than a little creepy and way more sophisticated than I had imagined. And while I don’t think it’s going to propel us to a misinformation doomsday just yet, Veo 3 strikes me as an absolute AI slop machine.AI generated video: Made with Veo 3Google introduced Veo 3 at I/O this week, highlighting its most important new capability: generating sound to go with your AI video. “We’re entering a new era of creation,” Google’s VP of Gemini, Josh Woodward, explained in the keynote, calling it “incredibly realistic.” I wasn’t completely sold, but then, a few days later, I had Veo 3 generate a video of a news anchor announcing a fire at the Space Needle. All it took was a basic text prompt, a few minutes, and an expensive subscription to Google’s AI Ultra plan. And you know what? Woodward wasn’t exaggerating. It’s realistic as hell.I tried the news anchor prompt after seeing what Alejandra Caraballo, a clinical instructor at Harvard Law School’s Cyberlaw Clinic, was able to produce. One of her clips features a news anchor announcing the death of US Secretary of Defense Pete Hegseth. He is not dead, but the clip is incredibly convincing. A post including a string of videos with AI-generated characters protesting the prompts used to create them has 50,000 upvotes on Reddit. The scenes include disasters, a woman in a hospital bed using a breathing tube, and a character being threatened at gunpoint — all with spoken dialogue and realistic background sounds. Real lighthearted stuff!Maybe I’m being naive, but after playing around with Veo 3 I’m not quite as concerned as I was at first. For starters, the obvious guardrails are in place. You can’t prompt it to create a video of Biden tripping and falling. You can’t have a news anchor announce the assassination of the president, or even generate a video of a T-shirt-and-chain-wearing tech company CEO laughing while dollar bills rain down around him. That’s a start.That said, you can generate some troubling shit. Without any clever workarounds I prompted Veo 3 to create a video of the Space Needle on fire. Starting with my own photo of Mount Rainier, I generated a video of it erupting with smoke and lava. Coupled with a clip of a news anchor announcing said disaster, I can see how you could seed some mischief real easily with this tool.AI generated video: Made with Veo 3Here’s the better news: it doesn’t seem like a ready-made deepfake machine. I gave it a couple of photos of myself and asked it to generate a video with specific dialogue and it wouldn’t comply. I also asked it to bring a pair of giant boots in a photo to life and have them walk out of the scene; it managed one boot stomping across the sidewalk with some comical crunching noises in the background.AI generated video: Made with Veo 3I had an easier time generating videos when my prompts were less specific, which is how I confirmed something my colleague Andrew Marino pointed out: Veo 3 is excellent at creating the kind of lowest-common-denominator YouTube content aimed at kids.If you’ve never been subjected to the endless pit of garbage on YouTube Kids, let me enlighten you. Imagine watching the worst 3D rendering of a monster truck driving down a ramp, landing in a vat of colored paint. Next to it, another monster truck drives down another ramp into another vat of paint — this time, a different color. Now watch that again. And again. And again. There are hours of this stuff on YouTube designed to mesmerize toddlers. These videos are usually harmless, just empty calories designed to rack up views that make Cocomelon look like Citizen Kane. In about 10 minutes with Veo 3, I threw together a clip following the same basic formula — complete with jaunty background music. But the clip that’s even more troubling to me is the two cartoon cats on a pier.AI generated video: Made with Veo 3I thought it would be funny to have the cats complain to each other that the fish aren’t biting. In just a couple of minutes, I had a clip complete with two cats and some AI-generated dialogue that I never wrote. If it’s this easy to make a 10-second clip, stretching it out to a seven-minute YouTube video would be trivial. In its current form, clips revert to Veo 2 when you try to extend them into longer scenes, which removes the audio. But the way that Google has been pushing these tools forward relentlessly, I can’t imagine it’ll be long before you can edit a full feature-length video with Veo 3.Honestly, I wonder if this sort of use for AI-generated video is a feature and not a bug. Google showed us some fancy AI-generated video from real filmmakers, including Eliza McNitt, who is working with Darren Aronofsky on a new film with some AI-generated elements. And sure, AI video could be an interesting tool in the right hands. But I think what we’re most likely to see is a proliferation of the kind of bland imagery that AI is so good at generating — this time, in stereo.See More:

Even at first glance, there’s something off about the body on the street. The white sheet it’s under is a little too clean, and the officers’ movements are totally devoid of purpose. “We need to clear the street,” one of them says with a firm hand gesture, though her lips don’t move. It’s AI, alright. But here’s the kicker: my prompt didn’t include any dialogue.Veo 3, Google’s new AI video generation model, added that line all on its own. Over the past 24 hours I’ve created a dozen clips depicting news reports, disasters, and goofy cartoon cats with convincing audio — some of which the model invented all on its own. It’s more than a little creepy and way more sophisticated than I had imagined. And while I don’t think it’s going to propel us to a misinformation doomsday just yet, Veo 3 strikes me as an absolute AI slop machine.AI generated video: Made with Veo 3Google introduced Veo 3 at I/O this week, highlighting its most important new capability: generating sound to go with your AI video. “We’re entering a new era of creation,” Google’s VP of Gemini, Josh Woodward, explained in the keynote, calling it “incredibly realistic.” I wasn’t completely sold, but then, a few days later, I had Veo 3 generate a video of a news anchor announcing a fire at the Space Needle. All it took was a basic text prompt, a few minutes, and an expensive subscription to Google’s AI Ultra plan. And you know what? Woodward wasn’t exaggerating. It’s realistic as hell.I tried the news anchor prompt after seeing what Alejandra Caraballo, a clinical instructor at Harvard Law School’s Cyberlaw Clinic, was able to produce. One of her clips features a news anchor announcing the death of US Secretary of Defense Pete Hegseth. He is not dead, but the clip is incredibly convincing. A post including a string of videos with AI-generated characters protesting the prompts used to create them has 50,000 upvotes on Reddit. The scenes include disasters, a woman in a hospital bed using a breathing tube, and a character being threatened at gunpoint — all with spoken dialogue and realistic background sounds. Real lighthearted stuff!Maybe I’m being naive, but after playing around with Veo 3 I’m not quite as concerned as I was at first. For starters, the obvious guardrails are in place. You can’t prompt it to create a video of Biden tripping and falling. You can’t have a news anchor announce the assassination of the president, or even generate a video of a T-shirt-and-chain-wearing tech company CEO laughing while dollar bills rain down around him. That’s a start.That said, you can generate some troubling shit. Without any clever workarounds I prompted Veo 3 to create a video of the Space Needle on fire. Starting with my own photo of Mount Rainier, I generated a video of it erupting with smoke and lava. Coupled with a clip of a news anchor announcing said disaster, I can see how you could seed some mischief real easily with this tool.AI generated video: Made with Veo 3Here’s the better news: it doesn’t seem like a ready-made deepfake machine. I gave it a couple of photos of myself and asked it to generate a video with specific dialogue and it wouldn’t comply. I also asked it to bring a pair of giant boots in a photo to life and have them walk out of the scene; it managed one boot stomping across the sidewalk with some comical crunching noises in the background.AI generated video: Made with Veo 3I had an easier time generating videos when my prompts were less specific, which is how I confirmed something my colleague Andrew Marino pointed out: Veo 3 is excellent at creating the kind of lowest-common-denominator YouTube content aimed at kids.If you’ve never been subjected to the endless pit of garbage on YouTube Kids, let me enlighten you. Imagine watching the worst 3D rendering of a monster truck driving down a ramp, landing in a vat of colored paint. Next to it, another monster truck drives down another ramp into another vat of paint — this time, a different color. Now watch that again. And again. And again. There are hours of this stuff on YouTube designed to mesmerize toddlers. These videos are usually harmless, just empty calories designed to rack up views that make Cocomelon look like Citizen Kane. In about 10 minutes with Veo 3, I threw together a clip following the same basic formula — complete with jaunty background music. But the clip that’s even more troubling to me is the two cartoon cats on a pier.AI generated video: Made with Veo 3I thought it would be funny to have the cats complain to each other that the fish aren’t biting. In just a couple of minutes, I had a clip complete with two cats and some AI-generated dialogue that I never wrote. If it’s this easy to make a 10-second clip, stretching it out to a seven-minute YouTube video would be trivial. In its current form, clips revert to Veo 2 when you try to extend them into longer scenes, which removes the audio. But the way that Google has been pushing these tools forward relentlessly, I can’t imagine it’ll be long before you can edit a full feature-length video with Veo 3.Honestly, I wonder if this sort of use for AI-generated video is a feature and not a bug. Google showed us some fancy AI-generated video from real filmmakers, including Eliza McNitt, who is working with Darren Aronofsky on a new film with some AI-generated elements. And sure, AI video could be an interesting tool in the right hands. But I think what we’re most likely to see is a proliferation of the kind of bland imagery that AI is so good at generating — this time, in stereo.See More:

Even if the United States were to impose a 25 per cent tariff on iPhones manufactured in India, the total production cost would still be much lower if compared with manufacturing the devices in the U.S, according to a report by Global Trade Research Initiative (GTRI).This comes amid a statement by U.S. President Donald Trump, threatening to impose 25 per cent tariffs on iPhones if Apple decides to make it in India. However, the GTRI report showed that manufacturing in India remains cost-effective, despite such duties.The report breaks down the current value chain of a $1,000 (roughly Rs. 83,400) iPhone, which involves contributions from over a dozen countries. Apple retains the largest share of the value, about $450 (roughly Rs. 37,530) per device, through its brand, software, and design.It also added that the U.S. component makers, such as Qualcomm and Broadcom, add $80 (roughly Rs. 6,672), while Taiwan contributes $150 (roughly Rs. 12,510) through chip manufacturing. South Korea adds $90 (roughly Rs. 7,506) via OLED screens and memory chips, and Japan supplies components worth $85 (roughly Rs. 7,089), mainly through camera systems. Germany, Vietnam, and Malaysia account for another $45 (roughly Rs. 3,753) through smaller parts.GTRI stated that China and India, despite being major players of iPhone assembly, earn only around $30 (roughly Rs. 2,502) per device. This is less than 3 per cent of the total retail price of an iPhone.The report argues that manufacturing iPhones in India is still economically viable even if a 25 per cent tariff is applied.This is mainly because of the sharp difference in labour costs between India and the U.S. In India, assembly workers earn approximately $230 (roughly Rs. 19,182) per month, while in the U.S. states like California, labour costs could soar to around $2,900 (roughly Rs. 2,41,860) per month due to minimum wage laws, a 13-fold increase.As a result, assembling an iPhone in India costs about $30 (roughly Rs. 2,502), while the same process in the U.S. would cost around $390 (roughly Rs. 32,526). In addition to this Apple gets the benefit of production-linked incentive (PLI) on iPhone manufacturing in India from government.If Apple were to shift production to the U.S., its profit per iPhone could fall drastically from $450 (roughly Rs. 37,530) to just $60 (roughly Rs. 5,004), unless retail prices are significantly increased.The GTRI report highlighted how global value chains and labour cost differences make India a competitive option for manufacturing, even in the face of potential U.S. trade restrictions.(This story has not been edited by NDTV staff and is auto-generated from a syndicated feed.)

Google's new AI video tool, Veo 3, is being used to create hyperrealistic videos that are now flooding the internet, terrifying viewers "with a sense that real and fake have become hopelessly blurred," reports Axios. From the report: Unlike OpenAI's video generator Sora, released more widely last December, Google DeepMind's Veo 3 can include dialogue, soundtracks and sound effects. The model excels at following complex prompts and translating detailed descriptions into realistic videos. The AI engine abides by real-world physics, offers accurate lip-syncing, rarely breaks continuity and generates people with lifelike human features, including five fingers per hand. According to examples shared by Google and from users online, the telltale signs of synthetic content are mostly absent. In one viral example posted on X, filmmaker and molecular biologist Hashem Al-Ghaili shows a series of short films of AI-generated actors railing against their AI creators and prompts. Special effects technology, video-editing apps and camera tech advances have been changing Hollywood for many decades, but artificially generated films pose a novel challenge to human creators. In a promo video for Flow, Google's new video tool that includes Veo 3, filmmakers say the AI engine gives them a new sense of freedom with a hint of eerie autonomy. "It feels like it's almost building upon itself," filmmaker Dave Clark says. Read more of this story at Slashdot.

There are two main types of Sims 4 players. There are the folks who are here for the simulation side of things, with all the fun drama that entails. And then there are those who are here to build dope houses. Regardless of which of those you are, there's one cheat that is absolutely essential (though it'll matter much more for the builders): bb.moveobjects.This cheat is among the most-used Sims 4 cheats aside from the ones that give you money. Once you've used it, it's not tough to understand why it's so beloved--the moveobjects cheat makes building a realistic house much easier, and generally allows you to play a lot more items than there would normally be room for. It's so fundamentally important that EA has actually incorporated certain aspects of the cheat into the base game. At this point, EA treats it more like a tool for advanced users than a real cheat.Table of Contents [hide]How to turn on bb.moveobjectsHow to turn on bb.moveobjectsTo activate this cheat, press ctrl+shift+C on PC, or hold down all four shoulder buttons if you're playing on a console, to bring up the command console at the top left of your screen, and then type "bb.moveobjects on". There is no need to activate cheats generally with the separate "testingcheats true" command or to do anything else. Moveobjects works by itself.Continue Reading at GameSpot

The Fresh Prince of Pasta Google’s Will Smith double is better at eating AI spaghetti … but it’s crunchy? Veo 3 is a major leap in AI video synthesis, but the sound effects need more cooking time. Benj Edwards – May 23, 2025 1:57 pm | 58 A still frame of "Will Smith eating spaghetti" in Google's Veo 3 AI video generator. A still frame of "Will Smith eating spaghetti" in Google's Veo 3 AI video generator. Story text Size Small Standard Large Width * Standard Wide Links Standard Orange * Subscribers only   Learn more On Tuesday, Google launched Veo 3, a new AI video synthesis model that can do something no major AI video generator has been able to do before: create a synchronized audio track. While from 2022 to 2024, we saw early steps in AI video generation, each video was silent and usually very short in duration. Now you can hear voices, dialog, and sound effects in eight-second high-definition video clips. Shortly after the new launch, people began asking the most obvious benchmarking question: How good is Veo 3 at faking Oscar-winning actor Will Smith at eating spaghetti? First, a brief recap. The spaghetti benchmark in AI video traces its origins back to March 2023, when we first covered an early example of horrific AI-generated video using an open source video synthesis model called ModelScope. The spaghetti example later became well-known enough that Smith parodied it almost a year later in February 2024. Here's what the original viral video looked like: One thing people forget is that at the time, the Smith example wasn't the best AI video generator out there—a video synthesis model called Gen-2 from Runway had already achieved superior results (though it was not yet publicly accessible). But the ModelScope result was funny and weird enough to stick in people's memories as an early poor example of video synthesis, handy for future comparisons as AI models progressed. AI app developer Javi Lopez first came to the rescue for curious spaghetti fans earlier this week with Veo 3, performing the Smith test and posting the results on X. But as you'll notice below when you watch, the soundtrack has a curious quality: The faux Smith appears to be crunching on the spaghetti. On X, Javi Lopez ran "Will Smith eating spaghetti" in Google's Veo 3 AI video generator and received this result. It's a glitch in Veo 3's experimental ability to apply sound effects to video, likely because the training data used to create Google's AI models featured many examples of chewing mouths with crunching sound effects. Generative AI models are pattern-matching prediction machines, and they need to be shown enough examples of various types of media to generate convincing new outputs. If a concept is over-represented or under-represented in the training data, you'll see unusual generation results, such as jabberwockies. We tried running the prompt ourselves in Veo 3 and "Will Smith" was blocked by Google's content filters, but upon requesting "A black man eating spaghetti," we got a similar crunchy sound result (either Lopez had early test access without filters, or perhaps one of his prompts got around the filter, which happens sometimes if you try different variations). AI-generated video from Google Veo 3: "A black man eating spaghetti." Veo 3 is remarkable in its ability to generate coherent dialog and music, and it's already led to many interesting examples you can find on X. Some are wildly realistic. Not content to merely enjoy a video of a man eating very al dente noodles, we thought we'd see if the man could sing and eat at the same time with the prompt, "A man singing an English language comedy opera about spaghetti at a kitchen table while eating it." AI-generated video from Google Veo 3: "A man singing an English language comedy opera about spaghetti at a kitchen table while eating it." We have come quite a long way since 2023, and AI video generators are only going to get more realistic and capable over time. If it weren't for the celebrity filter on Veo 3 at the moment, we could likely make Smith appear to do the singing—and to do just about anything else, which is the potentially troublesome part of AI video. The cultural singularity is upon us. AI-generated video from Veo 3: A man in a shark costume rapping about Ars Technica on stage. Speaking of which, we recently ran our own extended battery of video generation tests in Veo 3 and will have those for you soon in a separate feature. But for now, we thought we'd briefly update you on the Fresh Prince of noodle consumption. Bon appétit! Benj Edwards Senior AI Reporter Benj Edwards Senior AI Reporter Benj Edwards is Ars Technica's Senior AI Reporter and founder of the site's dedicated AI beat in 2022. He's also a tech historian with almost two decades of experience. In his free time, he writes and records music, collects vintage computers, and enjoys nature. He lives in Raleigh, NC. 58 Comments

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More It was a big week for AI announcements following events from Microsoft, Google, and Anthropic. But OpenAI is finishing things out with news of its own. And no, we’re not just talking about its $6.5 billion acquisition of Jony Ive’s design team to lead a new hardware effort, “io” at OpenAI. Today, the company upgraded its Operator autonomous web browsing and cursor controlling agent within ChatGPT from using the prior GPT-4o multimodal large language model to the newer and more powerful o3 reasoning model. The update, released globally today, May 23, 2025, is available as a “research preview” to paying subscribers of OpenAI’s $200 USD-monthly ChatGPT Pro plan. Basically, that is OpenAI’s way of saying it’s not a fully “sanded down” or perfected product yet — it may still have kinks and issues. But with rival Google offering its own top tier AI subscription bundle for a price of nearly $250 USD regularly (currently running a discount down to $125 for the first three months) to access its latest Gemini multimodal, Imagen image generation, and Veo video generation models, suddenly OpenAI’s ChatGPT Pro plan seems more affordable by comparison. What is OpenAI’s Operator and what is it for? Operator first debuted in January 2025 as OpenAI’s initial step into semi-autonomous agents, specifically Computer Using Agents (CUAs). The idea is to go beyond the chatbot interface of ChatGPT and allow OpenAI’s powerful AI models to start taking more actions on behalf of the user. Thus, Operator was designed to autonomously point, click, scroll, and type to complete web-based tasks such as booking dinner reservations, compiling shopping lists, or ordering event tickets. This agentic capability allows it to complete user tasks directly through a browser interface, from booking reservations to gathering online data. For safety, privacy and security purposes, Operator didn’t use any existing web browser on a user’s PC or Mac. Instead, it ran in a cloud-hosted virtual browser accessible via a standalone site—operator.chatgpt.com—where users could input requests and observe the agent perform tasks in real time. It combined vision, reasoning, and interaction capabilities based on GPT-4o, marking a new direction for OpenAI in agentic AI. The product was launched as a research preview for ChatGPT Pro subscribers and featured built-in safety measures like user confirmations, Watch Mode, and restrictions on high-risk web platforms. It was also being tested in enterprise contexts, including travel planning and civic services, demonstrating its potential across both consumer and business environments. o3 offers improved accuracy, structure, and success rates With this update, OpenAI aims to enhance performance across several key dimensions. The new o3-based Operator demonstrates improved persistence and accuracy during browser interactions. In practical terms, this means it is more likely to complete user tasks successfully and with less need for correction or repetition. Moreover, users can expect responses that are clearer, more structured, and more comprehensive. In comparative evaluations, the new model shows a distinct preference advantage over its predecessor. Human preference studies reveal that users favor the o3 model for its style, comprehensiveness, and clarity. It also performs strongly in instruction following and efficiency, though results for factual correctness are more balanced between versions. Performance on third-party evaluation benchmarks reflects these enhancements. On the OSWorld benchmark that measures completion of browser-based tasks, the o3 model scores 42.9 compared to 38.1 for the previous version. However, OpenAI notes that due to limitations in the automated grading system, the actual performance gain could be closer to 20 percentage points! On WebArena, the new model achieved a score of 62.9, up from 48.1. The most dramatic improvement appears on the GAIA benchmark, where the o3 model scores 62.2, vastly surpassing the prior model’s 12.3. Side-by-side task comparisons further illustrate these gains. In one example involving a restaurant booking request, the new model provided a clearer and more detailed list of available reservations, including locations, Michelin ratings, and seating notes, presented in a well-formatted table. The previous version, while functional, delivered less information in a less organized manner, according to an image included with the new o3 Operator release notes: The o3 model also inherits the safety measures introduced with earlier versions, with further fine-tuning for its role as an agentic system. OpenAI has integrated enhanced training against harmful task execution, prompt injection vulnerabilities, and mistakes involving user intent. Evaluations show that the model now confirms 94% of sensitive actions before executing them, with 100% confirmation in financial transactions. Prompt injection susceptibility has also decreased from 23% to 20%. Notably, the o3 Operator maintains a cautious boundary on certain high-risk web interactions, such as email or financial platforms, where it may require user supervision via Watch Mode or explicitly refuse to proceed. These measures are part of a layered approach to safety that combines model-level robustness with real-time monitoring. While the upgrade to Operator marks a technical improvement, it also reflects OpenAI’s ongoing commitment to responsible AI deployment. The system’s ability to take real-world actions introduces new risks, and the development team continues to refine its safety protocols accordingly. According to OpenAI’s updated o3 system card documentation, the model remains below high-risk capability thresholds in categories such as biological and chemical misuse and has no native coding environment or terminal access, further reducing potential misuse vectors. Operator remains a research preview and is accessible only to ChatGPT Pro users. The Responses API version of Operator will continue to be based on the GPT-4o model, at least for now. Implications for enterprise technical decision-makers The upgraded Operator stands to significantly enhance the workflows of professionals in AI engineering, orchestration, data management, and IT security. For those building or maintaining machine learning models, the model’s improved accuracy and structured outputs reduce the overhead of test validation and troubleshooting. In orchestration contexts, it offers a practical, reliable tool for automating browser-based components of complex pipelines. Data engineers can delegate manual web interactions—such as data verification and scraping—with more confidence, freeing time for higher-level optimization work. Security professionals, meanwhile, gain a safer way to simulate user behavior in audits and incident response exercises, thanks to the model’s layered safety mechanisms. Across these disciplines, the o3-based Operator introduces both a capability upgrade and a risk mitigation framework, making it a practical addition to the modern technical toolkit. Daily insights on business use cases with VB Daily If you want to impress your boss, VB Daily has you covered. We give you the inside scoop on what companies are doing with generative AI, from regulatory shifts to practical deployments, so you can share insights for maximum ROI. Read our Privacy Policy Thanks for subscribing. Check out more VB newsletters here. An error occured.

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More It was a big week for AI announcements following events from Microsoft, Google, and Anthropic. But OpenAI is finishing things out with news of its own. And no, we’re not just talking about its $6.5 billion acquisition of Jony Ive’s design team to lead a new hardware effort, “io” at OpenAI. Today, the company upgraded its Operator autonomous web browsing and cursor controlling agent within ChatGPT from using the prior GPT-4o multimodal large language model to the newer and more powerful o3 reasoning model. The update, released globally today, May 23, 2025, is available as a “research preview” to paying subscribers of OpenAI’s $200 USD-monthly ChatGPT Pro plan. Basically, that is OpenAI’s way of saying it’s not a fully “sanded down” or perfected product yet — it may still have kinks and issues. But with rival Google offering its own top tier AI subscription bundle for a price of nearly $250 USD regularly (currently running a discount down to $125 for the first three months) to access its latest Gemini multimodal, Imagen image generation, and Veo video generation models, suddenly OpenAI’s ChatGPT Pro plan seems more affordable by comparison. What is OpenAI’s Operator and what is it for? Operator first debuted in January 2025 as OpenAI’s initial step into semi-autonomous agents, specifically Computer Using Agents (CUAs). The idea is to go beyond the chatbot interface of ChatGPT and allow OpenAI’s powerful AI models to start taking more actions on behalf of the user. Thus, Operator was designed to autonomously point, click, scroll, and type to complete web-based tasks such as booking dinner reservations, compiling shopping lists, or ordering event tickets. This agentic capability allows it to complete user tasks directly through a browser interface, from booking reservations to gathering online data. For safety, privacy and security purposes, Operator didn’t use any existing web browser on a user’s PC or Mac. Instead, it ran in a cloud-hosted virtual browser accessible via a standalone site—operator.chatgpt.com—where users could input requests and observe the agent perform tasks in real time. It combined vision, reasoning, and interaction capabilities based on GPT-4o, marking a new direction for OpenAI in agentic AI. The product was launched as a research preview for ChatGPT Pro subscribers and featured built-in safety measures like user confirmations, Watch Mode, and restrictions on high-risk web platforms. It was also being tested in enterprise contexts, including travel planning and civic services, demonstrating its potential across both consumer and business environments. o3 offers improved accuracy, structure, and success rates With this update, OpenAI aims to enhance performance across several key dimensions. The new o3-based Operator demonstrates improved persistence and accuracy during browser interactions. In practical terms, this means it is more likely to complete user tasks successfully and with less need for correction or repetition. Moreover, users can expect responses that are clearer, more structured, and more comprehensive. In comparative evaluations, the new model shows a distinct preference advantage over its predecessor. Human preference studies reveal that users favor the o3 model for its style, comprehensiveness, and clarity. It also performs strongly in instruction following and efficiency, though results for factual correctness are more balanced between versions. Performance on third-party evaluation benchmarks reflects these enhancements. On the OSWorld benchmark that measures completion of browser-based tasks, the o3 model scores 42.9 compared to 38.1 for the previous version. However, OpenAI notes that due to limitations in the automated grading system, the actual performance gain could be closer to 20 percentage points! On WebArena, the new model achieved a score of 62.9, up from 48.1. The most dramatic improvement appears on the GAIA benchmark, where the o3 model scores 62.2, vastly surpassing the prior model’s 12.3. Side-by-side task comparisons further illustrate these gains. In one example involving a restaurant booking request, the new model provided a clearer and more detailed list of available reservations, including locations, Michelin ratings, and seating notes, presented in a well-formatted table. The previous version, while functional, delivered less information in a less organized manner, according to an image included with the new o3 Operator release notes: The o3 model also inherits the safety measures introduced with earlier versions, with further fine-tuning for its role as an agentic system. OpenAI has integrated enhanced training against harmful task execution, prompt injection vulnerabilities, and mistakes involving user intent. Evaluations show that the model now confirms 94% of sensitive actions before executing them, with 100% confirmation in financial transactions. Prompt injection susceptibility has also decreased from 23% to 20%. Notably, the o3 Operator maintains a cautious boundary on certain high-risk web interactions, such as email or financial platforms, where it may require user supervision via Watch Mode or explicitly refuse to proceed. These measures are part of a layered approach to safety that combines model-level robustness with real-time monitoring. While the upgrade to Operator marks a technical improvement, it also reflects OpenAI’s ongoing commitment to responsible AI deployment. The system’s ability to take real-world actions introduces new risks, and the development team continues to refine its safety protocols accordingly. According to OpenAI’s updated o3 system card documentation, the model remains below high-risk capability thresholds in categories such as biological and chemical misuse and has no native coding environment or terminal access, further reducing potential misuse vectors. Operator remains a research preview and is accessible only to ChatGPT Pro users. The Responses API version of Operator will continue to be based on the GPT-4o model, at least for now. Implications for enterprise technical decision-makers The upgraded Operator stands to significantly enhance the workflows of professionals in AI engineering, orchestration, data management, and IT security. For those building or maintaining machine learning models, the model’s improved accuracy and structured outputs reduce the overhead of test validation and troubleshooting. In orchestration contexts, it offers a practical, reliable tool for automating browser-based components of complex pipelines. Data engineers can delegate manual web interactions—such as data verification and scraping—with more confidence, freeing time for higher-level optimization work. Security professionals, meanwhile, gain a safer way to simulate user behavior in audits and incident response exercises, thanks to the model’s layered safety mechanisms. Across these disciplines, the o3-based Operator introduces both a capability upgrade and a risk mitigation framework, making it a practical addition to the modern technical toolkit. Daily insights on business use cases with VB Daily If you want to impress your boss, VB Daily has you covered. We give you the inside scoop on what companies are doing with generative AI, from regulatory shifts to practical deployments, so you can share insights for maximum ROI. Read our Privacy Policy Thanks for subscribing. Check out more VB newsletters here. An error occured.