 |
|
|
|||||
Duolingo has been through a lot of changes over the past few years. What was once solely a language-learning app has grown into a social media marketing machine, a destination for math and music lessons, and now an online chess tutor. In June, Duolingo launched a Duolingo Chess course to teach beginners the basic rules and moves by allowing them to play against an AI tutor named Oscar. This month, the company is taking the course further, launching a multiplayer version of the game where users can compete against one another. Duolingo, which is on track to surpass $1 billion in revenue this year, has 48 million daily active users and 11 million paying subscribers. Though the company doesnt reveal user numbers for the free chess course, at the Fast Company Innovation Festival in September, cofounder and CEO Luis von Ahn said that it had already notched millions and millions of users within three months. Even more remarkable: Von Ahn greenlit the development of the chess course only nine months before it launched. The story of how Duolingo created its hit chess course is the clearest illustration of how AI is transforming the company. Fast Company spoke with two of the principals behind the chess course. As they make clear, AI is not replacing engineers, but it is giving them a head start. Vibe coding has become a phenomenon, but we don’t have any high-profile examples of a hit product that relied on it. Until now. Chess as education Von Ahn admits to being skeptical about offering chess lessons when the idea was first pitched to him a couple of years ago. He didnt want Duolingo to turn into an app for games. But a conversation with Guatemalas minister of education changed his mind. (Von Ahn was born and raised in Guatemala.) The minister was so frustrated with the education system in her country that she said she was thinking about sending every student a chess board so that at least theyd learn how to play. That helped von Ahn see the game through an educational lens. He greenlit the idea and handed it to the pair of employees who initially pitched it: Neither of them knew how to play chess, and neither of them were engineers, von Ahn says. But they didnt need those skills to get started. They simply vibe coded a prototype. AI has been an important part of Duolingos workflow strategy. Earlier this spring, von Ahn sent a memo to his staff outlining plans to make Duolingo an AI-first company to remove bottlenecks and inspire employees to focus on creative work and real problems rather than repetitive tasks. The memo sparked a backlash. On social media, people accused von Ahn of using AI to eliminate jobs. At Fast Companys Innovation Festival, von Ahn cleared the air, saying that the goal isnt to save money. “We have never done a layoff. We have not laid off a single full-time employee, he said. The goal is not to replace human employees. The goal is to do a lot more and get closer to our mission. How to vibe code the game of chess The development of Duolingos chess course kicked off in September 2024 with Edwin Bodge, senior staff product manager, and Tyler Murphy, Duolingos chief product designer. Bodge and Murphy, who had worked together to launch the math and music products for Duolingo, had grown interested in chess when they were brainstorming ideas for a strong third subject. They saw the correlation between learning chess and language learning. To begin the prototyping, they tried a combination of products, including the popular AI-generated coding platform Cursor. Using Cursor, they created an initial version of the chess experience. Bodge says the first steps in prototyping were relatively easy. They began by describing a chess board on the Cursor platform, then added other elements, such as a character (which eventually became the AI tutor, Oscar), a progress bar, and the actual lessons and puzzles. But as Bodge and Murphy layered on more curriculum, things became much more complicated. They started to have a difficult time visualizing the course, so Bodge created multiple chess boards on Cursor to play through each version of the prototype. Soon, they had something fully formed enough to show to colleagues. It was a pretty janky prototype, Bodge admits. But we were able to put it into people’s hands and say, Well, here’s the curriculum we’ve been writing. Here’s all this stuff that we’ve been teaching and how we think it’s going to work. And from that moment, it just became so much more natural. Bodge notes that there are still some aspects of the prototype that are incorporated into the final product, including the curriculum and some of the AI-generated code that he and Murphy developed. That core of the vibe code and prototype is very much in the production app, he says. Building the final product Within a few months, Bodge and Murphy began handing off the project to Duolingo engineers to create the code that would bring the course to life. Sammi Siegel, staff software engineer, was instrumental in building the final product. Thanks to the prototypes that Bodge and Murphy had developed, there was a pretty strong vision for what [the course] should look like, Siegel says. The first few months were implementation, heads down, and coding everything all at once. Siegel says the chess game was the first time the team had launched a new subject within the existing Duolingo app. (The math and music classes were built to be stand-alone apps, and then eventually merged into Duolingo.) We were building [a course] within the current system, so we were able to prove how much faster that [process] was, Siegel explains. Siegels team was also helped by the amount of open-source work that exists around chess. She and her team were able to take advantage of publicly available chess engines, which analyze chess positions and the best moves. And then theres the visual simplicity of chess. Chess is pretty simple, Siegel says. It’s just a board and a guy telling you what he wants you to do on the board, so that made the implementation phase shorter. We were able to get a lot further with a very simple interface. (To build the chessboard, Siegel says the team used a software called Rive, which is cross-platform compatible and can be built in one go for iOS and Android.) Duolingo released the course in beta in April and launched it to the public in June. Now its ready for the next step: introducing player-versus-player mode, which entails a complex coding process. Siegel says creating real-life, player-versus-player functionality is tougher than building matches against a predictable bot, like Oscar. Youre coordinating two humans who think, pause, and play at their own pace, she explains. With Oscar, we can guarantee that youll get matched with a botthats pretty close to your skill level. Siegel adds that keeping both devices in sync is a challenging engineering problem, noting, Weve had to invest in anti-cheating systems and thoughtful handling of disconnections to ensure every match feels fair and stable. A new generation of online chess players By incorporating chess into Duolingo, the company is introducing the game to people who may have never considered learning it before. We talk internally about chess as a very male-dominated game, so it’s interesting to see how we can widen the scope of folks who have access to that kind of education, Siegel says. Bodge says when they first started developing the product, they had a hypothesis that it would appeal to a lot of people. But they wanted to make sure that it felt accessible and engaging, rather than elite and academic. Thats why, in classic Duolingo style, they use an animated characterOscarto play the role of the easygoing tutor. There are a lot of people who we’ve basically activated as chess players, Bodge says. Were ushering in a new generation of players.
Category:
E-Commerce
Black Friday isn’t what it once was. Less than 15 years ago, it was fairly common for people to wake up at ridiculously early hours to drive to a store, where they would stand in line, waiting for the doors to open in order to grab the best deals. Those people still exist, but not in the numbers they used to, thanks to the convenience of online shopping (and the early start to holiday deals). But as artificial intelligence becomes more entrenched in people’s habits, it could have an increasingly large role in Black Friday (and Cyber Monday). And 2025 could be something of a test case for the technology. The average consumer is expected to spend $1,595 on holiday gifts this year, according to Deloitte. That’s 10% less than 2024, a figure that highlights the importance shoppers will be placing on bargains this year. And a growing number of consumers will be relying on AI to help them find those deals. Some 33% of the people Deloitte spoke with in its 2025 Holiday Retail Survey said they plan to use AI as part of their holiday shoppingdouble the number who did last year. Many say the tech could assist them with inspiration and product discovery. That could benefit retailers who have already embraced AI in their recommendation engines, as well as those planning to roll it out. “Consumer adoption of gen AI shows that expectations are shifting toward personalization and efficiency,” Deloitte wrote. “Shoppers now expect instant recommendations tailored to their preferences, budgets, and recipients, raising the bar for retailers digital experiences. To meet holiday shoppers expectations, retailers could consider embedding AI-powered gift finders, style assistants, or deal copilots directly into their sites or apps.” A separate study from marketing automation platform Klaviyo found that 56% of consumers say theyll use AI tools during Black Friday and Cyber Monday. AI can do a lot more than help people think of creative gifts, of course. Gen AI models like ChatGPT can research prices and recommend the best deal, in some cases even making the purchase for you. And a growing number of people are taking advantage of that. Traffic from AI platforms to retail sites during Prime Days and other sales in July was up by 4,700%, according to Adobe Analytics. And the company is predicting an increase in AI usage of between 515% and 550% this holiday season, compared to 2024. Thinking of enlisting a GenAI to help you find deals? Here’s how you’ll want to go about it. Make your list. Check it twice Chatbots aren’t a lot of good without specifics. You’ll need to know exactly what you’re looking to buy if you’re planning to use AI for price comparisons. Using the broadest example, telling ChatGPT you’re looking for the best price on, say, a Barbie or a blender is akin to calling a Best Buy or GameStop and saying you want to know their best price for a game console. The $60 no-name brand that has a Tetris clone might technically be the correct answer, but that does you no good if you really wanted a PlayStation 5. Set the AI loose Ask your chatbot to find the best deals for your specific product. Again, details matter, so be sure to offer as much granularity as you can about the product. (To go back to the PS5 example, do you want the PS5 with a disc drive? What amount of internal storage do you want? Do you want a PS5 Slim or Pro or some other model?) It’s also worth asking the chatbot to suggest additional ways to save, such as applicable cash-back apps (like Rakuten), promo codes, or coupons. Fact check the results Prices change all the time during the holiday season, so just because ChatGPT says Store X is the best deal, you’ll still want to check that site’s website to verify the amount your AI assistant quoted is still valid. At the very least, using AI to help you shop will quickly eliminate some options and, ideally, free up some of your time, letting you spend less of November and December hunting for deals and more enjoying the season.
Category:
E-Commerce
From my earliest days as a journalist, Ive always prized my dictaphone. It sounds quaint now, but I actually remember excitedly keeping up with advancements in the field. Sonys ICD-TX50 was a particular revelation for me in 2012, with its tiny OLED display and worlds-thinnest 6.4mm frame. There was no sleeker way to show up to Tokyo press conferences. In recent years, though, my dictaphone collection has taken on a new, less physical form. Googles Pixel phones have been a revelation for journalists, offering real-time, on-device transcription through the Recorder app. Ive often found myself bringing a Pixel along to a press event even if I wasnt actively using it as a phone at the timethe ability to get an automatic transcript once your recording is done has been an incredible timesaver. But now a new wave of hardware, buoyed by the AI revolution, has been changing my habits once again. While the jury is out on a lot of hyped AI devices, some startups have locked in on voice recording as one of the few serious use cases. These arent like the Humane AI Pin or the much-mocked Friend pendanttheyre gadgets with a focused, real-world use case. {"blockType":"creator-network-promo","data":{"mediaUrl":"https:\/\/images.fastcompany.com\/image\/upload\/f_webp,q_auto,c_fit\/wp-cms-2\/2025\/03\/multicore_logo.jpg","headline":"Multicore","description":"Multicore is about technology hardware and design. It's written from Tokyo by Sam Byford. To learn more visit multicore.blog","substackDomain":"https:\/\/www.multicore.blog","colorTheme":"salmon","redirectUrl":""}} The devices Ive been using a couple of recorders from Plaud for several monthsthe Plaud Note and the Plaud NotePinand testing them against a product from a newer Japanese startup called Notta. At least for now, Ive found that both companies devices offer a significant upgrade over the Pixel for working journalists. The Plaud Note is an incredibly thin audio recorder that can magnetically attach to an iPhone or a newer Pixel through an included leather case. Theres a simple record button, as well as a switch for a mode that lets you record phone calls through audio vibrations. (Obviously, you should make sure of the legality in your jurisdiction before using this feature.) The NotePin, meanwhile, is a small pill-shaped device that attaches to various magnetic accessories like a clip or a wristband. Its designed for hands-free use, as opposed to holding the Plaud Note out toward a subject or resting it on a desk. Both devices feel very well-built, with solid metal cases. Power and recording status is indicated on each through a tiny LED in the A of the Plaud logo. The Note is a little easier to use than the NotePin thanks to its tactile round power button; the NotePin relies on an invisible capacitive surface that you have to hold down until you hear a vibration, which isnt quite as convenient in situations where youre scrambling to start a recording. The NotePin does have a usability advantage of its own through its magnetic USB-C charging adapter, however. The Note is so thin that it requires a proprietary pogo-pin cable, which Im not sure I trust myself not to lose at some point. Overall Plaud has designed some attractive hardware, but given that every phone has a microphone built in, the app is what sets the product apart. Once your phone is paired to one of the devices, transferring audio files is simple and quick over Wi-Fi, and then you have to upload them to Plauds cloud servers for processing. This can take several minutes depending on the length of the file, though its still much faster than manually transcribing. What Ive really appreciated in the chaotic aftermath of press conferences is how Plaud automatically generates a summary of the recording and identifies speakers, so you can dial right into the most important parts and verify the quotes for yourself. The auto setting works well, but you can also choose from specific templates and AI models like GPT-5, Gemini 2.5 Pro, and so on. Plauds free plan offers 300 transcription minutes a month. You can top up with 600 extra minutes for $12.99, or theres a $100 annual Pro plan with 1,200 minutes a month. The unlimited plan will set you back $240 a year. Meanwhile, the Plaud Note and NotePin each cost $159 themselves. I think this is reasonable pricing for anyone using this as a professional service. I did find Plauds app to be a little fussy in use, however, requiring a fair amount of interaction on the part of the user when it comes to uploading files and selecting how they ought to be processed. Despite its access to several AI services, I also thought the lack of in-app translation was an obvious miss. The Notta option Thats what got me interested in a Japanese startup called Notta, which is making a very similar product to the Plaud Note called the Notta Memo. I suspect each company may have related supply chain contacts in China, because the resemblance really is uncannyeven the magnetic leather cases and charging cables are near-identical. The Notta Memo does add a tiny monochrome OLED screen that shows useful information like recording status, and more importantly reminds me of my Sony dictaphone. (Plaud has since released a Note Pro model with a similar display, though I havent yet tested it.) Personally I think the Notta Memo looks sleeker, with a black finish and a geometric, quilted diamond texture. While Nottas app is similar to Plauds, I also preferred its design. Files begin transcribing automatically after transfer, and you can also translate them right within the app itself, which saves an extra step when youre on the move. I used the Notta Memo while reporting at some recent sports events in Tokyo and the process of recording directly on the device, uploading the file and getting a translated, annotated transcript that I could check myself was seamless. Notta slightly undercuts Plaud with subscription pricing, charging $98 a year for 1,800 minutes a month or $200 for the unlimited plan, and the Notta Memo device itself is $149. Nottas free plan only offers 120 minutes a month, though, and doesnt support longer recordings or transcript translation, which complicates the comparison for anyone unsure of how much theyll use it. Worth the cost Overall, any one of these devices is going to be a serious investment. But for working journalists or other professionalsmaybe lawyers or consultantswhere voice recordings are a daily need, I actually think theyre all quite compelling. If you have a Pixel phone, sure, you might get by with the built-in transcripts most of the time. But the advantages of having dedicated hardware are real. You can press the record button at any time regardless of what youre doing on your phone. The NotePin is a unique form factor that expands the situations in which you can rely on being able to record. Call recordings, meanwhile, straight-up arent supported in phone operating systems. The hype around AI has led to a lot of hardware startups that havent been able to live up to it. But one thing AI is indisputably very good at is understanding language, and these voice recorders are perfect for the task. If theres a use case for AI-first hardware right now, its this. {"blockType":"creator-network-promo","data":{"mediaUrl":"https:\/\/images.fastcompany.com\/image\/upload\/f_webp,q_auto,c_fit\/wp-cms-2\/2025\/03\/multicore_logo.jpg","headline":"Multicore","description":"Multicore is about technology hardware and design. It's written from Tokyo by Sam Byford. To learn more visit multicore.blog","substackDomain":"https:\/\/www.multicore.blog","colorTheme":"salmon","redirectUrl":""}}
Category:
E-Commerce
All news |
||||||||||||||||||
|
||||||||||||||||||