“What’s surprising about these large language models is how much they know about how the world works simply from reading all the stuff that they can find,” says Chris Manning, a professor at Stanford who specializes in AI and language.
But GPT and its ilk are essentially very talented statistical parrots. They learn how to re-create the patterns of words and grammar that are found in language. That means they can blurt out nonsense, wildly inaccurate facts, and hateful language scraped from the darker corners of the web.
Amnon Shashua, a professor of computer science at the Hebrew University of Jerusalem, is the cofounder of another startup building a bilingual AI model based on this approach. He knows a thing or two about commercializing AI, having sold his last company, Mobileye, which pioneered using AI to help cars spot things on the road, to Intel in 2017 for $15.3 billion.
Shashua’s new company, AI21, which came out of stealth last week, has developed an AI algorithm, called Jurassic-1, that demonstrates striking language skills in both English and Hebrew.
In demos, Jurassic-1 can generate paragraphs of text on a given subject, dream up catchy headlines for blog posts, write simple bits of computer code, and more. Shashua says the model is more sophisticated than GPT-3, and he believes that future versions of Jurassic may be able to build a kind of common-sense understanding of the world from the information it gathers.
Other efforts to re-create GPT-3 reflect the world’s—and the internet’s—diversity of languages. In April, researchers at Huawei, the Chinese tech giant, published details of a GPT-like Chinese language model called PanGu-alpha (written as PanGu-α). In May, Naver, a South Korean search giant, said it had developed its own language model, called HyperCLOVA, that “speaks” Korean.
Jie Tang, a professor at Tsinghua University, leads a team at the Beijing Academy of Artificial Intelligence that developed another Chinese language model called Wudao (meaning “enlightenment”) with help from government and industry.
The Wudao model is considerably larger than any other, meaning that its simulated neural network is spread across more cloud computers. Increasing the size of the neural network was key to making GPT-2 and -3 more capable. Wudao can also work with both images and text, and Tang has founded a company to commercialize it. “We believe that this can be a cornerstone of all AI,” Tang says.
Such enthusiasm seems warranted by the capabilities of these new AI programs, but the race to commercialize such language models may also move more quickly than efforts to add guardrails or limit misuses.
Perhaps the most pressing worry about AI language models is how they might be misused. Because the models can churn out convincing text on a subject, some people worry that they could easily be used to generate bogus reviews, spam, or fake news.
Musser says research suggests that it won’t be possible to use AI to catch disinformation generated by AI. There’s unlikely to be enough information in a tweet for a machine to judge whether it was written by a machine.
More problematic kinds of bias may be lurking inside these gigantic language models, too. Research has shown that language models trained on Chinese internet content will reflect the censorship that shaped that content. The programs also inevitably capture and reproduce subtle and overt biases around race, gender, and age in the language they consume, including hateful statements and ideas.
Similarly, these big language models may fail in surprising or unexpected ways, adds Percy Liang, another computer science professor at Stanford and the lead researcher at a new center dedicated to studying the potential of powerful, general-purpose AI models like GPT-3.
Am I a weirdo for liking in-person meetings? Sure, they can go on too long, be scheduled far too often and distract from your core work. But even before the pandemic made them an exotic memory, I always loved the energy of being in a room with other people, finding the same creative wavelength, working towards a common goal. There’s no video chat app that can replicate that. So, what about VR?
We’ve already seen companies like Spatial take a stab at virtual meetings, where you can chat with others, inspect objects and even explore 3D environments from the comfort your home with a VR headset. Now Facebook’s Oculus is entering the fray with Horizon Workrooms, an ambitious attempt to capture the best aspects of in-person meetings for Oculus Quest 2 users.
To be clear, Workrooms isn’tFacebook Horizon, its long-awaited multiplayer VR playground. But it is part of the company’s overall vision for the Horizon universe, one entirely devoted to collaborating in meeting rooms and classrooms. And despite being so focused, it’s still a showpiece for Facebook’s VR ambitions, tapping into the company’s expressive avatars, spatial sound, hand tracking and mixed-reality capabilities. You’ll even be able to stream your PC desktop in VR to take notes or goof off during meetings (just like real life!).
After spending an hour in a Horizon Workrooms demo, it’s clear that Facebook still has plenty of work left to do. There were a few connection issues, avatar glitches and at one point I got booted out entirely. But when it was running smoothly, it was the closest I’ve felt to being in an live meeting since March of last year.
But let me start from the beginning. After creating a Workrooms account, I installed the PC companion app and the Workrooms Quest 2 app. Once I booted it up in VR, I was surprised that it recommended putting my controllers down and enabling the Quest 2’s hand-tracking. Outside of demos for that feature, I can’t think of any Oculus apps that have done the same. After that, I customized a goofy avatar — an idealized version of myself with purple hair and no beer belly — and mapped out the edge of my desk so Workrooms can accurately place me in front of virtual tables.
Facebook
As soon as I hopped into my Horizon Workrooms demo, it was instantly clear why Oculus recommended hand-tracking. I waved to a PR rep, and as I struggled to unmute myself, I was able to throw a thumbs up to confirm I was hearing her. I shouted hello to CNET’s Scott Stein, who was sitting a few seats away, and his audio sounded appropriately far off. We all spent way too much time staring at our hands and testing the limits of the Quest 2’s finger tracking capabilities (for the record, rolling your hands over each other like a wannabe raver can easily scramble things).
Even though I was in a swanky virtual room, with a stylized deer head wearing VR goggles on the wall, and ceiling-tall windows overlooking expansive lake and mountain rage somewhere in the Pacific Northwest, sitting desk-side with others felt vaguely normal. We made small talk about our families and our vastly different locations (including folks in California, New Jersey, London and myself in Atlanta). I could see who was working on Macs and PCs, based on the virtual accessories in front of them. And along the wall there were also video feeds from other Facebook reps calling in from their PCs, which reminded me of the enormous video conferencing screens in many meeting rooms.
To see my keyboard and mouse, I hit a button to enable pass-through mode, which projected grey-scale footage of my actual desk inside of Workrooms’ VR environment. It wasn’t crystal clear, but it was enough to touch-type notes in Evernote by streaming my PC desktop. If you’re running an Apple Magic Keyboard or a specific Logitech model, you’ll also be able to see your keyboard in VR, so you won’t have to deal with the pass-through video.
Andrew Bosworth, Facebook’s head of VR and AR, said during the demo that he believes there’s "something missing" with video calling, and that there’s a strong pull towards VR collaboration apps as a solution. Of course, that’s exactly what you’d expect Facebook’s VR lead to say. But he has a point.
Video calls are a fantastic way for seeing people and maintaining social connections. But when it comes to getting work done, they can get old fast. Everyone’s audio comes in from one channel; it’s hard to make eye contact in large group chats; and the entire session just lives on your computer or phone screen. There’s no sense of actually being in a space with others. Bosworth says that Facebook has been using Workrooms internally for about six months, and he’s noticed some clear benefits. In particular, he can remember specific meetings better thanks to the immersion of sitting beside people in VR with realistic spatial audio.
Facebook
Of course, virtual reality can also make meetings awkward in entirely new ways. I quickly learned that I couldn’t look closely at someone’s avatar if they were sitting nearby, because it actually felt like I was staring at them. I couldn’t reach for a cup of water on my real-world desk, because it was sitting in front of my neighbor’s virtual space. I also had to be careful about where I placed my hands, because the Quest 2’s hand tracking could make things look… suspect, if they’re under your desk.
All of that awkwardness culminated when Mark Zuckerberg crashed our demo and sat in the virtual chair beside me. He sketched out his overall vision for Horizon as a stepping stone for his metaverse ambitions. But as he was speaking, I was also trying hard to keep my virtual cool: Don’t stare! Don’t be weird with your hands! Pay attention and be respectful! I thought I was doing pretty well in the heat of the moment, but a recording of our session looked like I was performing some sort of interpretive dance beside the world’s fifth richest man. (Thankfully, that footage isn’t meant for public consumption.)
Zuckerberg only stopped by for a few minutes, but the fact he made an appearance at all is telling. Workrooms isn’t some side project: It’s a significant component of what he wants Facebook to become. You can see that in the app’s more ambitious features, like a whiteboard that can grow as long as you’d like. You can sketch out thoughts by holding your Oculus controller straight up, so that you can use the end like a massive pen. And if you want to stretch your legs, you can assign a blank wall in your room to serve as your life-sized whiteboard. You can also change the size and layout of your workspace at will, allowing you to move from a circular conference table to something that resembles a classroom.
Facebook
Perhaps because Zuckerberg wants Workrooms to be a success, the company is also making it more accessible to people who don’t want Facebook accounts. You can join a session via video chat by creating a Workrooms account, which is separate from Oculus and Facebook logins. If you want to enjoy the experience in VR, though, you’ll need a Facebook account to use the Quest 2. Maybe letting people peek at the 2D Workrooms experience could encourage them to go virtual.
The company is also being up-front about security and privacy, saying that it won’t use conversations or materials from Workrooms sessions to inform Facebook ads. You’ll also be able to report individuals and entire groups if they’re harassing you or violating community standards. That was a major concern when I demoed Facebook Horizons last year; the idea of a VR playground sounds great, but how do you make it safe for everyone?
If anything, Horizon Workrooms tells us that Zuckerberg’s metaverse ambitions are more than just marketing fluff. He paid $2 billion for Oculus, after all, and Facebook has continually invested in VR development and content. You don’t spend that sort of money if you don’t have a vision for the future.
When he stopped by our demo, Zuckerberg told us he used to scribble code in his notebook during middle school. As soon as he got home, he would type it up with the hopes of building an immersive world to hang out with friends. Workrooms isn’t the perfect VR collaboration solution yet, but it could be the start of what the younger Zuckerberg envisioned. After all, the metaverse won’t be built in a day.
Don’t expect an easy getaway if one of Boston Dynamics’ Atlas robots ever chases you down. The Hyundai-owned firm has shared a video (below) of the humanoid bots successfully completing a parkour routine in an obstacle course for the first time. The pair of Atlas machines leapt gaps, vaulted beams and even coordinated a backflip, all without missing a beat — they might be more graceful than you are.
The routine took "months" of development, according to the company, and served as a useful test of the robots’ ability to maintain their balance while switching behaviors and coordinating actions. This isn’t just canned behavior, either. As with other recent tests, Atlas now uses visuals to adapt its movement to the course.
Don’t worry that the robots will chase you down any time soon, though. The vault in particular has about a 50 percent failure rate, and there’s still a slight chance of failure at each step. Even a fist pump at the end didn’t go smoothly, Boston Dynamics said. The company still has to refine movements that are limited by the very nature of the robots themselves, such as the lack of a spine and the relatively weak arm joints.
As it stands, Atlas isn’t a production robot like Spot. It’s a research model meant to push the limits of robotics. Boston Dynamics does, however, envision this parkour practice leading to future helper robots that can handle a wide variety of tasks with human-like dexterity. Let’s just hope they stay on our good side.
Pokémon Go developer Niantic Labs has acquired Scaniverse, the company announced on Tuesday. According to creator Keith Ito, the iOS 3D scanning app will remain available on the App Store, with all the features previously part of its $17 annual subscription now included for free. What’s more, Niantic will continue to support the software with future feature updates.
Much like its other recent acquisition, it sounds like Niantic hopes Scaniverse will help in its goal to build a 3D map of the world. As part of the deal, Ito will join the company’s augmented reality engineering team. “Together, we plan on taking 3D scanning to new heights by combining Niantic’s dynamic mapping with Scaniverse’s LiDAR reconstruction,” he said.
Meanwhile, the company notes, “Scaniverse will make multi-OS scanning easier and more accessible to the Niantic Explorer community who contribute to this effort every day.” It says that will allow it to create more immersive AR experiences in Pokémon Go, Harry Potter: Wizards Unite and its upcoming games.
If you subscribe to the 2TB / $10-per month (or higher) Google One plan, then you get a little something extra over those who do not: Access to Google’s mobile VPN. This feature has, until now, been limited to customers in the US, but now Google is letting a far greater number of countries get their hands on the goodies. As of today, the Google One VPN is available to Android users in Mexico, Canada, the UK, France, Germany, Spain and Italy.
This particular product is designed to encrypt all of your mobile traffic when you’re out and about to prevent sniffers from getting access to your personal data. The VPN itself is baked into the Google One app, and the company says it will encrypt and protect you no matter what app you’re using. Plus, if you’re unsure how exactly to get connected, you can schedule a Pro Session for an expert to coach you through the basics.
At the same time, Google is reminding folks that the VPN has seen a number of feature upgrades since its launch, including the ability to block internet traffic when the VPN is inactive. You can also bypass the VPN in specific situations or with specific mobile apps, and set timed controls for when you need to be free from its protective embrace. Google added that the VPN has been independently audited and found to be robust enough for most folks to trust it.
Sony has announced that after months of negotiations, the acquisition of anime streaming service Crunchyroll is complete. The service joins fellow anime entertainment company Funimation as an official subsidiary of Sony.
The buyout is worth a whopping $1.175 billion, according to a press release. With its newfound resources, Crunchyroll can broaden its reach by distributing its content through various Sony channels like PlayStation. Similarly, Funimation will also take advantage of Sony’s deep resource bag. It remains unclear when the two anime streaming networks will ever merge.
Chairman, president, and CEO of Sony Group Corporation, Kenichiro Yoshida, said the company is excited to welcome Crunchyroll to Sony.
“Anime is a rapidly growing medium that enthralls and inspires emotion among audiences around the globe,” Yoshida said. “The alignment of Crunchyroll and Funimation will enable us to get even closer to the creators and fans who are the heart of the anime community. We look forward to delivering even more outstanding entertainment that fills the world with emotion through anime.”
Tony Vinciquerra, chairman and CEO of Sony Pictures Entertainment Inc., said the buyout adds “tremendous value to Sony’s existing anime business,” reiterating that the company wants to create the ultimate anime experience.
“With the addition of Crunchyroll, we have an unprecedented opportunity to serve anime fans like never before and deliver the anime experience across any platform they choose, from theatrical, events, home entertainment, games, streaming, linear TV–everywhere and every way fans want to experience their anime,” Vinciquerra said. “Our goal is to create a unified anime subscription experience as soon as possible.”
While the goal for Sony seems to be to create a unified subscription experience, it remains to be seen how Crunchyroll’s acquisition impacts Funimation. With Sony now owning the two networks, though, competition amongst streaming services for where to watch anime is slowly dwindling. It’s possible that Crunchyroll could either be linked alongside Funimation or be folded with the business, making its offerings all Funimation all the time.
Got a news tip or want to contact us directly? Email news@gamespot.com
SpaceX just dipped its toes into the Internet of Things. SpaceNewsreports that SpaceX has acquired Swarm Technologies, a startup best known for smallsats that power IoT services. The two companies aren’t shy about how this will help Swarm — this gives the younger company the resources it needs to take on other smallsat operators, no to mention "synergies" that come with a company used to designing and launching satellites.
Swarm’s commercial mini-satellite network just went live in February of this year, but it already had enough funding to finish rolling out an array of 150 satellites. It’s a small company with roughly 30 employees as of March.
It’s not clear if or how SpaceX will alter its strategy as a result of the deal, although an FCC filing suggested Elon Musk’s outfit would primarily benefit from Swarm’s "intellectual property and expertise." In other words, this might be an acquihire rather than an expansion of SpaceX’s strategy.
There are areas where the businesses might mesh well. SpaceX is busy deploying a giant constellation of relatively small Starlink internet satellites. The firm also recently began launching rideshare missions full of cubesats and other small payloads. Swarm would at least serve as a customer for those rideshare missions. It might also help SpaceX refine or expand its smallsat efforts, no matter how much IoT plays a role.
