← Tech Brew Ride Home
Tech Brew Ride Home
Interviewing For A Job At Anthropic? DON’T Use AI.
00:21:46
Nvidia unveiled the RTX Spark, an Arm-based consumer chip family built with MediaTek on TSMC 3, plus a DGX Station desktop that runs 1T-parameter models. Intel detailed its Crescent Island GPUs, MiniMax launched a coding model rivaling Opus 4.7 at 1/40th the price, and Anthropic bans AI in interviews.
Loading summary
Transcript3 lines
- [00:00]
A
Study and play come together on a Windows 11 PC and for a limited time, college students get the best of both worlds. Get the unreal college deal Everything you need to study and play with select Windows 11 PCs. Eligible students get a year of Microsoft 365 Premium and a year of Xbox Game Pass ultimate with a custom color Xbox wireless controller. Learn more@windows.com studentoffer while supplies last ends June 30th terms at aka mscollegepc.
- [00:35]
B
Welcome to the Tech We Write home from Monday, June 1st, 2026. I'm Brian McCullough. Today Nvidia unveiled the RTX Spark, an ARM based consumer chip family plus a DGX station desktop that runs 1 trillion parameter models. Minimax launched a coding model rivaling Opus 4.7 at 1 40th the price and anthropic bans the use of AI in their own job interview process. Here's what you missed today in the world of tech Foreign. Today's episode is brought to you by Doppel Social Engineering Attacks don't bother to knock. They slip right into your inbox phone or on websites, instead pretending to be a harmless internal email or normal text message until it's too late. Doppel sees right through the disguise. Their AI native platform trains your team to recognize threats like deepfakes, bad links and impersonation attempts before they can actually cause any damage. Doppel strengthens team resilience by giving employees the tools and defenses they need to protect themselves from increasingly sophisticated social engineering threats. It's kind of like having a security team that has eyes everywhere, and their digital risk protection takes it one step further by keeping an eye on every channel to connect patterns and shut them down fast. Invest in Social Engineering Defense learn more@doppel.com that's.p p e l.com all of the sudden, today was a day of chip news. For example, Nvidia announced the RTS Spark, an ARM based consumer chip family that it calls the most efficient PC chip ever built. It's made on TSMC3 in partnership with Mediatek. Quoting the Verge, the RTX Spark is effectively the same GB10 chip that's in the DGX Spark, the tiny personal AI supercomputer that Nvidia released last year, Only now it's a family of chips instead of just one. The flagship version appears to be spec to spec, identical with 20 CPU cores, 6,144 GPU cores, and 128 gigabytes of LPDDR 5x memory. But Nvidia says there will be lesser versions later, targeting lower prices and with as little as 16 gigabytes of RAM. Like Apple and Qualcomm's chips, this Nvidia chip is ARM based silicon, meaning legacy Windows software made For intel and AMD's x86 processors need to run through an emulation layer to work. That can mean lower performance, but Microsoft has now spent years getting Windows and its Prism emulator ready for Qualcomm and now Nvidia chips. And Nvidia claims its own graphics and AI chops will take the idea further than ever before. With the power of the RTX Spark Nvidia Bose, you can render a 90 gigabyte 3D scene, edit 12K resolution video, or play the graphically intensive Indiana Jones and the Great Circle at a smooth 100 frames per second at 1440p resolution, all in a 14 millimeter thick laptop. Without a power cord plugged in, and with up to 128 gigabytes of unified memory tied with AMD's previous gen Strix Halo parts, an RTX Spark laptop or desktop can also host 120 billion parameter AI agents, something that Microsoft is seemingly excited about for Windows. At Microsoft's Build conference this week, it'll be showing off new Windows security and containment primitives that, along with Nvidia's open shell runtime, allows personal agents to run safely under full user control. Nvidia claims this adds up to a new personal computing paradigm where AI is the UX and users no longer need to Master complicated app UIs because you'll just talk to your PC instead of needing to use a mouse and keyboard. Nvidia suggested that, for example, an Esports streamer could get their PC to automatically turn off their lights, mute their microphone, and change their broadcasting mode when they want to step away and grab dinner. A designer could use Adobe to automatically turn a sketch into a full image, render a 3D model of it, then create an AI video just by asking. A software developer can automatically monitor their GitHub projects and autonomously fix QA issues, with the AI agent taking over the laptop's keyboard and mouse cursor to do repetitive and boring tasks. Nvidia says that with the RTX Sparks local AI shops, your data stays private and you won't be burning through tokens to do AI things. I'm not convinced Nvidia has pieced together the Star Trek computer just yet, but it does seem like the company has a lot of partners on board. Almost every major laptop vendor is accounted for, with eight specific laptops already confirmed for this fall. One of those is from Microsoft, which is putting the Nvidia RTX Spark and a new laptop that Surface boss Andrew Hill tells us is the most powerful thing we've ever made. It's called the Surface Laptop Ultra. Those machines are apparently just the start. Nvidia says that partners are already working on over 30 laptops and over 10 desktops, with Acer, Asus, Dell, Gigabyte, HP, MSI and Lenovo all on board. RTX Spark is going to be a family of products that are going to attack a lot of different price points, Nvidia promises. The overall market opportunity that we see is quite large, and between Microsoft Nvidia's wrangling efforts, lots of Windows developers are also on board with arm. The company points out that the Blender, DaVinci, Resolve, Maxim, Cinema 4D, Maxon, Redshift, Topaz, Photo Capcut, Cubase, Bitwig, Studio Affinity by Canva and many more developers all run natively on ARM today, as do the audio, video, midi and control peripherals they require. Adobe is on board with special optimizations for Premiere and Photoshop that take advantage of Nvidia's new chip. Even games with Anti Cheat that thumb their nose at Linux and the Steam Deck are now supporting Windows on ARM too. Microsoft writes that Riot Games is now bringing both League of Legends and Valorant to Windows on arm, and Nvidia tells us it's working with more developers who use Easy Anti Cheat, battleye and Denuvo. Epic's Fortnite already came to Windows on ARM last November. After an announcement last March, Nvidia says all the top games will run on RTX Spark and provide a great experience. That's a pretty high bar to meet. There are still many open questions, though. Neither Nvidia nor Microsoft gave us a clear idea of how much these computers might cost. Save that the first batch of this fall is targeting the more premium price points in the market. On all day battery life, Nvidia would only say that we should expect it to be much better than anything you've seen before on RTX laptops, and that you won't need a charger if you're not pushing heavy workloads. The chip scales down to low single digit wattage and goes as high as 80 watts. Nvidia says the latter means it could theoretically drain bigger laptop batteries in around an hour on full bore. In terms of performance, Nvidia didn't have a single statistic or chart to share, and Nvidia wouldn't answer questions about how the RTX Spark family stacks up to chips from intel, amd, Apple and Qualcomm, saying Nvidia will have more to share closer to launch. But Nvidia does say that depending on the application, it has roughly the graphical power of an RTX 5070 mobile GPU, and that we should expect the CPU portion to be competitive with anything else out there in the Windows space. End quote. Meanwhile, intel is zagging in the other direction, quoting Tom's hardware at Computex 2026, intel is offering a few more details and updates for its next generation data center GPU product. Codenamed Crescent island. The Crescent Island GPU will be built on Intel's G3P GPU architecture. Intel says this architecture is built for agentic AI and it supports a broad range of potential data types from FP4 for high performance AI inference all the way up to FP64 potentially for scientific computing applications. Intel isn't providing any raw throughput specs at this stage of Crescent Island's development, so we can't make any guesses about its compute performance. Crescent island will be a PCI Express add in card with a 350 watt power target, placing its power and thermal requirements close to products like Nvidia's RTX Pro 5000 Blackwell card. But Crescent Island's architecture is quite different from anything else on the market. It foregoes GDDR or HBM memory for LPDDR5X. Intel says its Crescent island reference design will include 160 gigabytes of LPDDR5X, but that the chip is designed to allow partners the flexibility to build accelerators with up to 480 gigabytes of memory. Recent leaks and past analyses have Crescent island will take a wide and slow approach with LPDDR5X, potentially using a 640 bit bus connecting 20 LPDDR5X devices to achieve these high capacities. Some basic math suggests that partners would need to employ 24 gigabyte modules to fully realize that memory capacity, and those modules are already available from sources like Samsung. With 10.7 Gbps, Crescent island would offer 684 Gbs of memory bandwidth. From a design standpoint, maximizing memory capacity while maintaining adequate bandwidth will help keep more AI data close to the GPU and require less data movement, potentially making Crescent Island a more efficient inference engine compared to GPUs built with lower capacity GDDR devices. Going with LPDDR 5X doesn't put pressure on valuable advanced packaging capacity or compete with higher end accelerators for scarce HBM making it potentially easier for intel to produce these accelerators economically and in volume. There's no word on how or where the Crescent island package itself will be fabricated, however. Because Crescent island is an air cooled card with relatively modest power requirements, it's likely ready to drop into traditional 4U or 5U GPU servers, potentially making it appealing for companies trying to develop on premise inferencing solutions. Eight of these accelerators with a full 480GB of RAM each would produce an impressively Dense server with 3.8 terabytes of local GPU memory, allowing for massive models or swarms of smaller AI agents to reside within one box. Intel describes Crescent island as coming soon and has touted a second half 2026 launch for the platform, so we'll presumably learn more about the product and the ecosystem building around it as we progress further into the year. End quote. AI is uncharted territory and many leaders are trying to navigate through without a guide to help them. That's why Morning Brew created the Intellig Shift, a new podcast with PwC. It's all about how AI is fundamentally changing different industries. Host Dan Priest sits down with people who work with AI on a daily basis. Together they discuss real stories, real strategies and real takeaways for leaders. Get guidance from industry experts. Listen to the Intelligence Shift wherever you get your podcast. There's a lot to navigate in the economy right now and small business owners like you are really feeling the effects. You can't control interest rates or tariffs, but there is one thing you can control, and that's how efficient your business is. Gusto can help. Automating payroll and HR with Gusto is one of the fastest ways to cut friction and focus on what actually moves the needle. Gusto is online payroll and benefit software built for small businesses. It's all in one remote, friendly and incredibly easy to use, so you can pay, hire, onboard and support your team from anywhere. Unlimited payroll runs for one monthly price. Try Gusto today@gusto.com brew and get three months free when you run your first payroll. That's three months of free payroll@gusto.com brew but back to Nvidia because they also unveiled a desktop grade computer that they Say can run 1 trillion parameter AI models on device. Quoting Silicon Angle, Nvidia says it's uprooting supercomputers from the vast sprawling data center complexes they normally live inside and squeezing them into compact desktop sized workstations that can sit on or aside the desks of individual developers, researchers and data scientists. That's the idea behind the new Nvidia DGX station for Windows, which it said is to be the world's first desk side AI supercomputer. Announced today at GTC Taipei concurrent with the Computex conference, it's slated to launch in the fourth quarter and specifically designed for building and running powerful always on AI agents that can automate workflows with Windows ecosystems. The system was developed in close collaboration with Microsoft and delivers data center grade AI infrastructure into the desktop or desk side form factor for the first time. It supports Nvidia Open Shell on Windows, a secure open source agentic runtime that integrates Microsoft security and container technology to ensure agents can safely within isolated sandbox environments. Nvidia says it's addressing a major friction point for enterprise developer teams. Traditionally, the heaviest duty AI workloads such as model training, fine tuning and large scale inference have been run in Linux based cloud data centers that have the necessary infrastructure to support them. However, the bulk of the Fortune 500 runs much of its everyday business operations and engineering workflows locally on Windows based systems. By bringing its latest architecture to the Windows ecosystem, the DGX station eliminates the need for teams to push early stage AI workloads to the cloud so they can instead build and deploy powerful AI agents locally while leveraging the tools and security infrastructure they trust. The DGX station for Windows is powered by Nvidia's GB300 Grace Blackwell Ultra Desktop Superchip, which is a customized version of the Grace Blackwell platform for rackscale data centers. It packages a Blackwell Ultra Graphics Processing unit with an Nvidia Grace central processing unit that's connected via NVLink C2C interconnect. All told, the system provides 20 petaflops of FP4 performance and up to 748 gigabytes of memory. With such immense computing power, the DGX station for Windows can run frontier models of up to 1 trillion parameters locally or support hundreds of parallel AI agents running simultaneously on the same platform. But that's just the basic setup for customers that need to run intensive simulations or visualization workflows. DGX Station for Windows can be paired with an Additional Nvidia RTX Pro 6000 Blackwell Workstation GPU, enabling physical AI workflows that combine powerful compute with ray tracing capabilities. This supports the creation of AI agents that can not only perceive but interact with virtual environments that support lifelike physics. For even more power, customers can link multiple DGX Station for Windows Systems over Nvidia ConnectX8 Super NIC to create their very own integrated clusters of desktop supercomputers and run much larger workloads, Nvidia said. Because every agent runs in Nvidia's Open Shell runtime environment, they can be set up to run within tight boundaries and with strict security guardrails in place. Instead of relying on easily bypassed behavioral system prompts, Open Shell leans on Windows security primitives to create isolated sandbox environments for each agent. When users do this, their agent's security and privacy policies will be enforced at the system level, preventing them from overriding corporate policy or leaking credentials and sensitive information. Open Shell also makes it possible for administrators to manage their agents using Microsoft's fleet of management tools. Meanwhile, the Windows subsystem for LIN ensure capability with Linux tool chains, Nvidia said. The DGX station for Windows will become available before the end of the year and no word on pricing, which will likely be insane. Again. On the can the big LLM models be undercut on price front? Quoting the information Chinese AI developer Minimax on Monday launched a a new large language model called M3, saying the new model's coding capability approaches that of Anthropic's Opus 4.7, which was released in April. The new Minimax model is particularly suitable for coding and complex multi step tasks for AI agents. According to the company. It can understand text, images and video in prompts. Minimax's M3 costs only $0.12 per 1 million input tokens, compared with $5 for Opus 4.7. Competition is heating up among Minimax and other Chinese developers of open source models that offer much cheaper AI coding alternatives to Anthropic and OpenAI. Such models have gained popularity among cost conscious customers around the world. Deepseek launched its long awaited new model V4 in late April. Moonshot AI, another Chinese Open source AI developer, also released its latest model, Kimi K26, in April. End quote. Finally today, what does it take to get hired at Anthropic? Well, for one thing, don't use Claude in the interview, quoting Bloomberg. For Silicon Valley's hottest companies, hiring has long been a way to test more than practical ability. From Google's famous brain teasers onward, interview rituals have helped companies turn their own theories of talent into lore. For applicants lucky enough to get a call from an Anthropic recruiter, as many as five rounds of rigorous interviews and skills assessments follow. According to candidates who've been through the process. Those exercises are largely intended to be completed sans AI. Anthropic says it prohibits its use in interviews unless explicitly authorized. All candidates are then required to sign a non disclosure agreement before moving on to interview with the hiring manager or any other employee outside of the recruiting team. The hiring process includes a culture interview as well. Many companies have something similar, but what distinguishes Anthropics is its perceived intensity, according to candidates who experienced the process over the past year and how seriously the company appears to take it, according to Aline Lerner, founder of prep company Interviewing IO. One researcher who applied last year who requested anonymity to speak candidly about industry peers says the cultural questions came in quick succession, with the interviewer explaining he was trying to get signal and occasionally cutting him off. When the answers seemed to stop yielding new information. The researcher sensed that his focus on immediate concerns and such as the risks of people forming relationships with chatbots was seen as too pedestrian. The researcher was rejected. End quote. Oh thanks, Anthropic announcing this just now as I finished up recording the show. Anthropic says it has confidentially filed for an IPO, joining OpenAI and SpaceX and preparing to go public this year. More on that tomorrow if anything substantially new comes out. For those of you following along, unfortunately, Arsenal lost the Champions League final this weekend. But here's the thing. If at the beginning of the season you had offered me Arsenal winning the League but losing the Champions League final on penalties, I would have bitten your hand off for that and twice on Tuesday. So it was a great season for us. And I know you hate us rest of the footballing world, but get ready because we're gonna dominate for the next few years. This team is young, it's talented, and it is just getting started. Talk to you tomorrow.
- [19:57]
C
You can't reason with the sun. Trust us, we've tried. This summer, it's time to put that angry ball of fire on mute. Columbia's Omnishade technology is engineered to protect you from the sun's harsh rays that can burn and damage your skin. The sun is relentless, but so is our gear. Level up your summer@columbia.com to spend more time outside and less time slathering on aloe lotion. You're welcome, Columbia Engineered for whatever.