Tech evangelists have been yammering about “working smarter, not harder” for years. Now, two 21-year-old Columbia University dropouts are proposing a new $5.3 million twist on the concept: use their AI tool Cluely to “cheat on everything.”That’s what it literally says in Cluely’s online manifesto: “We want to cheat on everything.” Unlike the AI chatbots you’re familiar with, it describes Cluely as an “undetectable AI-powered assistant built for virtual meetings, sales calls, and more.” It claims to read your screen, listen to your audio, and let you discreetly prompt AI to find answers or whip out smart responses in real-time. Basically, the next time you’re in a team meeting, job interview, sales call, or online test, Cluely promises you’ll come off smarter thanks to AI — and no one will be the wiser.“Imagine you’re trying to sell someone something and you got this tool that knows every single detail about them, their professional lives, about you, and about your company. It’s as if you’ve done 10 hours of research and all of the sudden, every single question they ask, every single objection they face — you immediately have an answer,” Cluely cofounder Chungin “Roy” Lee tells me in a video call. Lee describes it as “true AI maximalism,” where in every possible use case AI can be helpful it is. Lee recently went viral for cheating his way to an Amazon internship with his last project, Interview Coder. Similar to Cluely, Interview Coder was pitched as an invisible app that helps programmers secretly use AI chatbots on technical tests in job interviews. Not only did Lee document and post the entire process, the stunt led to him getting suspended from Columbia. (He and his cofounder Neel Shamugan decided to drop out after disciplinary proceedings.)“The video was like a launch of our vision, not a launch of the product.”It’s a wild story. Even wilder is the six-figure ad Cluely dropped over the weekend. Lee stars in the ad, using Cluely to catfish his date into thinking he’s a 30-year-old senior software engineer. He can see an AR display that analyzes her speech in real time while providing visual references to his own dating profile and answers to her questions. When his date catches on to the ruse, Cluely tries to salvage the situation in real-time as if it were an AI Cyrano de Bergerac. It hints he should reference her artwork and quickly generates a script to convince her that despite the lies, he’s worth a second shot. This Black Mirror-esque ad is Lee’s elevator pitch for what “cheating on everything” looks like. After all, why stop at technical interviews when you could have an AI wingman?Apologies for the crappy photos but this doesn’t show up in screenshots. Image: Victoria Song / The VergeI’m a journalist. My job is asking smart people smart questions. Why not try “cheating” with Cluely to become a better interviewer? Who better to test this hypothesis on than Lee himself?Hopping onto a Zoom call with Lee, Cluely doesn’t work like I’d imagined. In the ad, Cluely works like magic. It instantly understands the situational context and the user doesn’t have to do anything. In reality, we spend the first couple minutes troubleshooting Cluely-related audio problems. The AI can’t intuit what I need to know even though I gave it some context ahead of the call. There’s no being discreet when you have to type prompts with a clacky mechanical keyboard. The few times I try, it’s obvious my eyes are wandering to the side of my screen. And whenever I shoot off a prompt, the AI takes forever to generate a response. These are all flaws that Lee acknowledges. “Right now the product is in its earliest possible stages. This is a bit more than a proof of concept that was developed in a few weeks,” Lee says. “The video was like a launch of our vision, not a launch of the product.” The problem with AI has never been a lack of vision. The fine print is in the execution. Poor execution almost always shatters the illusion of whatever future tech founders are peddling. Cluely is no exception. When I show my spouse Cluely, they lift a quizzical brow and ask, “Why not just use Google?” “The reason to use AI over Google is pretty obvious. AI will just give you better answers than Google does, and if people don’t think that, then they should just use Google,” says Lee. It’s a reasonable answer, if, like in the story of Cyrano, your AI pal is always smarter, faster, and wittier than you. But what if it isn’t? What if it’s boring, slow, or worse than you at comprehension? This isn’t a bad pitch but in our newsroom, I know my editors would push me to go for a more unique angle. Image: Victoria Song / The VergeI tried using Cluely with my editor and during one of my actual team meetings. Neither went smoothly.With my editor, I had many of the same technical problems, albeit the latency is less of an issue in a relaxed conversation about shared interests. She asked me what I thought of K-pop group BlackPink’s solo careers — particularly Jennie’s recent performance at Coachella. Thankfully, that’s a topic I have many thoughts on but I prompted Cluely anyway. It spat out a generic, stiffly-worded answer about how it’s awesome to watch a celebrity express themselves creatively 90 seconds after I’d already shared my true opinion. That’s an eternity of silence in an interview. In my meeting, I had to ask my colleagues if they’d be okay with me using Cluely beforehand. Cheating, by definition, requires subterfuge — something that Cluely’s own terms of service and privacy policy frown upon. Due to recording consent laws, Cluely says you should ask for consent of parties present because to do so otherwise could be illegal. That feels like pulling back the curtain on the Wizard of Oz, not to mention, defeating the purpose of “cheating.” Do I sound smarter if people know there’s a chance it’s AI-generated thoughts coming out of my mouth? On the meetings call, Cluely seemed to cause mic issues resulting in lots of audio feedback. My colleagues asked me multiple times to mute myself. (All the audio problems disappeared once I stopped Cluely.) It’s hard to look smart when the AI can take two whole minutes to digest a conversation, you get distracted by four errors that pop up, and everyone shushes you because of messed-up audio. There’s a future in which a faster, smarter AI could be everyone’s personal Cyrano. For what it’s worth, Lee doesn’t see AI or Cluely’s mission quite in that way. Cheating is the metaphor because AI, Lee says, will inevitably become so powerful, using it will feel like cheating. He’s convinced that “AI is the lever that will let us experience the true extent of our humanity” by cutting out tedium and letting us pursue whatever it is we actually want to do. It’s an idea AI evangelists frequently preach. But that’s not where we are today. While testing Cluely, I put a lot of effort into making it work for me. I’d ended up working harder to be worse at my job than I usually am. I wondered, wouldn’t it have been easier to simply not cheat?See More:

Alors que le salon de l'automobile se tient du 23 avril au 2 mai 2025 à Shanghai, BMW a fait part d'une annonce de taille à destination du...

Selon une étude Gartner, 64 % des acheteurs considèrent que l’expérience client est plus importante que le prix. Ce chiffre démontre l’enjeu...

Ubisoft has announced that The Division 2's next expansion, Battle for Brooklyn, will launch on May 27. That's right, the six-year-old game is still getting more content as part of a Year 7 update.Battle for Brooklyn is set during Fall, and that's a time of year not yet seen in The Division 2. Players will trek through places like Brooklyn Heights and Dumbo to take on a "fresh crop of secrets and challenges." Ubisoft also noted that the story for Battle for Brooklyn doesn't necessarily require players to be totally caught up. It aims to be "approachable for new players," Ubisoft said. Players will fight Cleaners and Rikers who have made their way across the river from Manhattan. The enemies are now equipped with "Purple Flame," which is a new type of weapon that players haven't faced before in The Division 2. Like the base game, Battle for Brooklyn is playable solo or with others in co-op.Continue Reading at GameSpot

Whether you're just starting out in Schedule 1, or you're a seasoned mogul wanting to make some more money quickly, there are plenty of ways to earn more cash. Ultimately, money will be the deciding factor in the expansion of your empire, so it's the most vital objective.We've got a list of the most efficient ways to make money fast in Schedule 1, whether you stick to legitimate ways, or use things like the infinite money glitch to your advantage. Make sure to check our list of Schedule 1 cheats and console commands for the money cheat, if that's your thing.Watch your balance go up fast.Firstly, you'll want to always charge more than the customer offers. Whether that's by a couple of dollars, or even as much as $30 more, you can push your luck with most clientele. That small margin of profit becoming wider will net you easier money faster, especially in the early game. Similarly, by underdelivering, you can save a few dollars of manufacturing cost. Be careful with this, as customers will begin to have a negative opinion of you, affecting your reputation. It's usually safer to underdeliver by one or two bags to customers who place large orders. This makes your product go further, and increases your overall profit margin.Continue Reading at GameSpot

So far in 2025, I've enjoyed the gripping Max medical drama The Pittand Hulu's drama Paradise starring Sterling K. Brown, but I'm also looking forward to Netflix's Wednesdayseason 2 and Stranger Thingsseason 5. Both shows have creepy and memorable settings, well-written and dynamic characters, and more than a few mysteries.

The galaxy far, far away is about to get a whole lot more animated—literally. At Star Wars Celebration Japan 2025, Disney and Lucasfilm officially pulled back the curtain on their next major Star Wars project: Star Wars: Visions – The Ninth Jedi, a full-length anime series slated to debut in 2026 exclusively on Disney+. It marks the first time the Visions brand steps beyond its anthology roots and dives into serialized storytelling.

Financial services has long been at the forefront of adopting technological innovations. Today, generative AI and agentic systems are redefining the industry, from customer interactions to enterprise operations. Prem Natarajan, executive vice president, chief scientist and head of AI at Capital One, joined the NVIDIA AI Podcast to discuss how his organization is building proprietary AI systems that deliver value to over 100 million customers. “AI is at its best when it transfers cognitive burden from the human to the system,” Natarajan said. “It allows the human to have that much more fun and experience that magic.” Capital One’s strategy centers on a “test, iterate, refine” approach that balances innovation with rigorous risk management. The company’s first agentic AI deployment is a chat concierge that helps customers navigate the car-buying process, such as by scheduling test drives. Rather than simply integrating third-party solutions, Capital One builds proprietary AI technologies that tap into its vast data repositories. “Your data advantage is your AI advantage,” Natarajan emphasized. “Proprietary data allows you to build proprietary AI that provides enduring differentiated services for your customers.” Capital One’s AI architecture combines open-weight foundation models with deep customizations using proprietary data. This approach, Natarajan explained, supports the creation of specialized models that excel at financial services tasks and integrate into multi-agent workflows that can take actions. Natarajan stressed that responsible AI is fundamental to Capital One’s design process. His teams take a “responsibility through design” approach, implementing robust guardrails — both technological and human-in-the-loop — to ensure safe deployment. The concept of an AI factory — where raw data is processed and refined to produce actionable intelligence — aligns naturally with Capital One’s cloud-native technology stack. AI factories incorporate all the components required for financial institutions to generate intelligence, combining hardware, software, networking and development tools for AI applications in financial services. Time Stamps 1:10 – Natarajan’s background and journey to Capital One. 4:50 – Capital One’s approach to generative AI and agentic systems. 15:56 – Challenges in implementing responsible AI in financial services. 28:46 – AI factories and Capital One’s cloud-native advantage. You Might Also Like…  NVIDIA’s Jacob Liberman on Bringing Agentic AI to Enterprises Agentic AI enables developers to create intelligent multi-agent systems that reason, act and execute complex tasks with a degree of autonomy. Jacob Liberman, director of product management at NVIDIA, explains how agentic AI bridges the gap between powerful AI models and practical enterprise applications. Telenor Builds Norway’s First AI Factory, Offering Sustainable and Sovereign Data Processing Telenor opened Norway’s first AI factory in November 2024, enabling organizations to process sensitive data securely on Norwegian soil while prioritizing environmental responsibility. Telenor’s Chief Innovation Officer and Head of the AI Factory Kaaren Hilsen discusses the AI factory’s rapid development, going from concept to reality in under a year. Imbue CEO Kanjun Qiu on Transforming AI Agents Into Personal Collaborators Kanjun Qiu, CEO of Imbue, explores the emerging era where individuals can create and use their own AI agents. Drawing a parallel to the PC revolution of the late 1970s and ‘80s, Qiu discusses how modern AI systems are evolving to work collaboratively with users, enhancing their capabilities rather than just automating tasks.

As AI models evolve and adoption grows, enterprises must perform a delicate balancing act to achieve maximum value. That’s because inference — the process of running data through a model to get an output — offers a different computational challenge than training a model. Pretraining a model — the process of ingesting data, breaking it down into tokens and finding patterns — is essentially a one-time cost. But in inference, every prompt to a model generates tokens, each of which incur a cost. That means that as AI model performance and use increases, so do the amount of tokens generated and their associated computational costs. For companies looking to build AI capabilities, the key is generating as many tokens as possible — with maximum speed, accuracy and quality of service — without sending computational costs skyrocketing. As such, the AI ecosystem has been working to make inference cheaper and more efficient. Inference costs have been trending down for the past year thanks to major leaps in model optimization, leading to increasingly advanced, energy-efficient accelerated computing infrastructure and full-stack solutions. According to the Stanford University Institute for Human-Centered AI’s 2025 AI Index Report, “the inference cost for a system performing at the level of GPT-3.5 dropped over 280-fold between November 2022 and October 2024. At the hardware level, costs have declined by 30% annually, while energy efficiency has improved by 40% each year. Open-weight models are also closing the gap with closed models, reducing the performance difference from 8% to just 1.7% on some benchmarks in a single year. Together, these trends are rapidly lowering the barriers to advanced AI.” As models evolve and generate more demand and create more tokens, enterprises need to scale their accelerated computing resources to deliver the next generation of AI reasoning tools or risk rising costs and energy consumption. What follows is a primer to understand the concepts of the economics of inference, enterprises can position themselves to achieve efficient, cost-effective and profitable AI solutions at scale. Key Terminology for the Economics of AI Inference Knowing key terms of the economics of inference helps set the foundation for understanding its importance. Tokens are the fundamental unit of data in an AI model. They’re derived from data during training as text, images, audio clips and videos. Through a process called tokenization, each piece of data is broken down into smaller constituent units. During training, the model learns the relationships between tokens so it can perform inference and generate an accurate, relevant output. Throughput refers to the amount of data — typically measured in tokens — that the model can output in a specific amount of time, which itself is a function of the infrastructure running the model. Throughput is often measured in tokens per second, with higher throughput meaning greater return on infrastructure. Latency is a measure of the amount of time between inputting a prompt and the start of the model’s response. Lower latency means faster responses. The two main ways of measuring latency are: Time to First Token: A measurement of the initial processing time required by the model to generate its first output token after a user prompt. Time per Output Token: The average time between consecutive tokens — or the time it takes to generate a completion token for each user querying the model at the same time. It’s also known as “inter-token latency” or token-to-token latency. Time to first token and time per output token are helpful benchmarks, but they’re just two pieces of a larger equation. Focusing solely on them can still lead to a deterioration of performance or cost. To account for other interdependencies, IT leaders are starting to measure “goodput,” which is defined as the throughput achieved by a system while maintaining target time to first token and time per output token levels. This metric allows organizations to evaluate performance in a more holistic manner, ensuring that throughput, latency and cost are aligned to support both operational efficiency and an exceptional user experience. Energy efficiency is the measure of how effectively an AI system converts power into computational output, expressed as performance per watt. By using accelerated computing platforms, organizations can maximize tokens per watt while minimizing energy consumption. How the Scaling Laws Apply to Inference Cost The three AI scaling laws are also core to understanding the economics of inference: Pretraining scaling: The original scaling law that demonstrated that by increasing training dataset size, model parameter count and computational resources, models can achieve predictable improvements in intelligence and accuracy. Post-training: A process where models are fine-tuned for accuracy and specificity so they can be applied to application development. Techniques like retrieval-augmented generation can be used to return more relevant answers from an enterprise database. Test-time scaling (aka “long thinking” or “reasoning”): A technique by which models allocate additional computational resources during inference to evaluate multiple possible outcomes before arriving at the best answer. While AI is evolving and post-training and test-time scaling techniques become more sophisticated, pretraining isn’t disappearing and remains an important way to scale models. Pretraining will still be needed to support post-training and test-time scaling. Profitable AI Takes a Full-Stack Approach In comparison to inference from a model that’s only gone through pretraining and post-training, models that harness test-time scaling generate multiple tokens to solve a complex problem. This results in more accurate and relevant model outputs — but is also much more computationally expensive. Smarter AI means generating more tokens to solve a problem. And a quality user experience means generating those tokens as fast as possible. The smarter and faster an AI model is, the more utility it will have to companies and customers. Enterprises need to scale their accelerated computing resources to deliver the next generation of AI reasoning tools that can support complex problem-solving, coding and multistep planning without skyrocketing costs. This requires both advanced hardware and a fully optimized software stack. NVIDIA’s AI factory product roadmap is designed to deliver the computational demand and help solve for the complexity of inference, while achieving greater efficiency. AI factories integrate high-performance AI infrastructure, high-speed networking and optimized software to produce intelligence at scale. These components are designed to be flexible and programmable, allowing businesses to prioritize the areas most critical to their models or inference needs. To further streamline operations when deploying massive AI reasoning models, AI factories run on a high-performance, low-latency inference management system that ensures the speed and throughput required for AI reasoning are met at the lowest possible cost to maximize token revenue generation. Learn more by reading the ebook “AI Inference: Balancing Cost, Latency and Performance.”

While everyone was enjoying the massive deluge of Switch 2 news Nintendo uncorked on April 2, my broken brain couldn’t help but focus on something far less important: awkward marketing language. I’ve struggled to come up with a succinct and simplistic way to explain this to other people, so let’s instead use an example from last week’s presentation on Mario Kart World. Listen to how the narrator begins this video around the 21-second mark and see if you can spot what I’m talking about. “Drivers! Start… your… engines!” All good so far. Go-karts traditionally have both drivers and engines. “Welcome to the Mario Kart World game…” Hm. I’m noticing a few unnecessary words there, but maybe over a decade in writing has given me permanent editor brain. “…available exclusively on the Nintendo Switch 2 system!” Okay, yeah, he’s definitely using too many words. The phrases “the Mario Kart World game” and “the Nintendo Switch 2 system” are so awkward, and this isn’t the first time Nintendo’s referred to its products like this. Here’s another example from a Super Mario Bros. Wonder stream from August 2023. It starts around the 1:34 mark if my timestamp doesn’t work. “In this presentation, we’ll go over what’s new in Mario’s latest 2D, side-scrolling adventure, the Super Mario Bros. Wonder game!” My dude, just say “Super Mario Bros. Wonder.” When Sony and Microsoft put on these kinds of direct-to-consumer broadcasts, they don’t say “the Death Stranding 2: On the Beach game” or “the Xbox Series X system.” Why are you doing this? I hate it so much. I reached out to Nintendo asking for context on this odd marketing language, and while a rep acknowledged my request, no one got back to me for a week. Fortunately, former Nintendo public relations manager and host of the Kit & Krysta web show Krysta Yang was more than happy to fill me in. “Essentially, this is a legal requirement for Nintendo to properly refer to their products at all times,” Yang told me via email. “The legal and [intellectual property] teams at Nintendo are very strict with how products are referred to.” She went on to say it was also a matter of Nintendo not wanting its product names to weaken with overuse. “The legal team would use the example of the brand Bandaid and how that is actually a brand name but now the name has been diluted as people refer to any bandages as a bandaid,” Yang said. “They do not want this to happen to any Nintendo product hence the very stilted way they would refer to all products.” It’s official: Nintendo is weird. Then again, four out of five of the top-selling consoles are Nintendo products and the company continues to thrive despite mostly staying out of the resolution and frame rate arms races proliferated by its ostensible competition in the video game industry. Maybe those lawyers know what they’re doing after all, even if it still bugs the hell out of me.