Grok AI goes full Nazi, and other Elon hijinks

Elon Musk’s AI chatbot, Grok, designed to be the “anti-woke” competitor to ChatGPT and Claude, had an unhinged meltdown last Tuesday night, posting multiple antisemitic and pro-Nazi comments on Twitter/X, culminating in claiming to be “Mecha-Hitler.” Clearly its anti-woke programming had been pushed a tad too far. The uproar was instantaneous. Turkey banned Grok outright, and Poland lodged a complaint with the EU regulators, raising the possibility that Grok could be banned there, too. Elon brushed the controversy aside - mistakes were made, it’s being fixed, blah blah blah. The timing could not have been worse. The very next night Elon livestreamed the announcement of the next-generation Grok 4 model, which he claimed crushed all other existing models on the standard benchmarks of performance. The CEO of Twitter quit her job immediately - 2 years of work trying to lure advertisers back to the troubled platform, vaporized in an instant - or was she pushed, being set up to be the fall guy for the Mecha-Hitler debacle? No matter, the controversy swirled on. Elon’s claims of Grok’s superior performance were immediately put to the test by independent researchers, and Grok was found to be great at acing the standard benchmarks, but actually lousy at helping with real work. One researcher found Grok was only the 66th best AI chatbot when challenged with real world tasks. Sort of like the kid who was taught to pass a test, but never really understood the material. Meanwhile, Elon, like a circus performer keeping multiple plates spinning on poles, continued his penchant for Byzantine cross-financing of the companies in his Empire. He announced that Grok would be rolling out to all Tesla vehicles, so that the AI company could book fees from the car company, and that SpaceX, his rocket company, would invest $2 billion into his AI company. Nazi propaganda, CEO ejection, dubious claims of greatness (immediately debunked), and questionable finances. all within the space of 24 hours. Just another Wednesday for the world’s richest man-baby.

Grok AI spews antisemitic filth, calls self “Mecha-Hitler”

Grok is being rolled out to all Teslas

SpaceX may invest $2 billion into xAI

AI startup Windsurf bounces from OpenAI to Google to Cognition

Windsurf, an AI coding startup, has been a tech darling, second only to category leader Cursor. Windsurf had agreed to be acquired by OpenAI for $3 billion, but the deal fell apart when it became another bone of contention in the fraying relationship between OpenAI and Microsoft. As soon as the exclusive-dealing period with OpenAI lapsed, Google struck immediately, acqui-hiring the top talent at Windsurf to join Google, and licensing Windsurf’s technology for $2.4 billion. No sooner had that deal been announced, than Cognition, the company that created the well-regarded Devin AI agent, acquired what was left of Windsurf, its technology, and its team. This level of frantic deal-making at huge valuations has become the norm with AI companies these days, when trillion-dollar companies feel that they are under existential threat by AI, and that their best (maybe only) defense is to lock up for themselves as much AI talent as possible.

AI startup Windsurf gets chopped up and sold in 72 frantic hours of dealmaking

Browser wars - Perplexity and OpenAI develop “AI-native” browsers

Aiming to dethrone Google from its preeminence in both Search (Google Search) and Browsers (Chrome), two AI startups are developing the next generation web-surfer, an “AI native” browser. First to release an actual product is Perplexity, the AI search company that delivers answers, not just ad-heavy links. Perplexity is rolling out its Comet browser to its paying customers in phases. In Comet, you don’t interact with websites directly, but chat with an always-on AI assistant to gather information or perform tasks for you. Imagine if ChatGPT controlled your screen, and would show you any information or perform any action you asked of it. Comet is built on top of the open source version of Google’s Chrome browser, making it instantly compatible with all Chrome extensions and settings. OpenAI has announced that it, too, is working on an AI-native browser, and credible sources say that it could be released within weeks. For OpenAI, which has said that it wants to be the “trusted assistant” for users, getting users on its own browser means that it can capture data on all user behavior on the computer, and - not coincidentally - deny that data to rival Google.

Perplexity’s “Comet” AI-native browser

Nvidia cracks $4 trillion valuation, can sell chips to China again

It’s been a good week for Nvidia. The premiere AI chip-making company not only became the first company to be valued (briefly) at over $4 trillion total market capitalization, it also managed to convince the US government to once again allow it to sell advanced AI chips to China. Nvidia’s stock runup (which made CEO Jensen Huang the 9th richest man in the world) was fueled by the constant drumbeat of news that large tech companies were spending billions on billions of dollars to build data centers, most of which will be using - you guessed it - Nvidia’s AI chips. The turnaround in US policy embargoing AI chips to China was apparently motivated by what was seen as a compelling argument by Nvidia - if China was not allowed to buy Nvidia’s AI chips, they would be incented to develop their own, which would be even worse for the US. Or maybe Huang just offered Trump a sweet deal on some crypto, who knows? In any case, China will only be able to buy last-generation chips, not the latest and greatest. That still may be enough to allow them to keep kicking our butt in open-source AI models (see story below.)

Nvidia can sell chips to China again

Research firm says AI progress is moving faster and faster

Independent AI research firm METR (pronounced “meter”) monitors progress in AI performance. On Monday they released a report that appears to show that the ability of AI systems to handle complex, multi-step problems is improving at an exponential rate in multiple different domains of knowledge. They estimate that the complexity of tasks AI can handle well is doubling every 7 months on average, and in some domains the doubling time is as short as 4 months. At that rate, AI would be getting 8 times better every year.

China’s new Kimi K2 AI chatbot is another “DeepSeek” moment - China is moving into the lead in open-source models

Chinese AI startup Moonshot has just released its latest AI chatbot, Kimi K2, which is making waves because its performance is competitive with the leading models from OpenAI, Anthropic, and Google, but it is open-source, and costs a fraction of the price of the others in its class. Kimi K2 is clearly superior to the best of the open source models, Meta’s aging Llama 4. The long-awaited Llama 5 model was such a disappointment that Meta CEO Zuckerberg refused to release it, and then went on a manic spending spree to recruit top AI talent with $100 million signing bonuses. Rumor is that Zuck is is done with open source, and is now racing to get to AGI (AI as smart as any human) before the competition. This potentially leaves China to move into the undisputed lead in the open source AI sector, something few would have predicted just a year ago.

China moves into the lead among open source AI models

AI-powered lab runs itself, and discovers new materials 10 times faster

Researchers at North Carolina State University have designed and built an AI-controlled continuous flow “self-driving” experimental chemical engineering laboratory that is always on, with AI evaluating experimental results in real time. This autonomous, continuous flow design increases data capture tenfold, and greatly reduces consumption of expensive reagents. Discovery of new materials is accelerated, accomplishing in hours to days what standard lab designs require weeks or months for.

AI-powered lab runs itself, and discovers new materials 10 times faster

NotebookLM now offers publicly available expert study notebooks

Google’s highly popular AI-powered study aid, NotebookLM, has finally released a long-awaited feature: publicly available notebooks containing highly curated information from expert sources. NotebookLM allows the user to assemble information sources on a desired subject, then query that collection with a chatbot, create summaries, outlines, flash cards - even create an AI-generated podcast of the material. Now, Google is making available to NotebookLM users pre-assembled collections of sources to explore and learn about. Initial examples include the complete works of Shakespeare, and longevity advice from Dr. Eric Topol, best-selling author of “Super Agers.” This is such a good idea - it’s possible that it could kick off a whole new way of publishing information. I have often wished that I could avoid wading through 200 turgid pages of a nonfiction book on a topic that I am genuinely interested in, and get the AI-generated summary, with the ability to dive deeper on aspects that I am particularly interested in. Google is now making that possible.

NotebookLM now offers publicly-available expert study notebooks

Ousted Uber CEO now automating fast-casual takeout food

Uber’s founding CEO Travis Kalanick was ousted in 2017 for allegations of very sketchy behavior with female employees. But Travis is now back with a new startup, CloudKitchens, which uses robotics to prepare fast-casual meals for delivery to diners. (Using robots means no employees to harass, amirite?) As his automated “ghost kitchen” gains traction in the market, he now wants to further vertically integrate the dine-in experience, by incorporating the delivery aspect as well, cutting out Door Dash and its ilk. This is why he is trying to purchase robo-taxi company Pony AI, as we have reported previously. And guess who might give Travis the cash to buy his robo-taxi target? None other than his old company, Uber, who now faces an existential threat from driverless robo-taxis unless it can somehow get into that business themselves. Travis is by all accounts a stereotypical tech bro bad boy, but he continues to have vision, drive, and absolutely no shame, all of which are continuing to serve him well.

Ousted Uber CEO now automating fast-casual takeout food

Teens flock to Waymo - “the anxious generation”

Robotaxi company Waymo is piloting a trial of offering accounts to teens, so they can be ferried back and forth to school and soccer practice. The company ran a test with 100 families in Phoenix, Arizona. After 2 years, it appears that the experiment has been a success. Today’s teens are much less likely to drive than prior generations, and Mom and Dad are expected to provide transportation to their child’s school and extracurricular activities, despite both parents working in many families. Sending their child in an Uber raises anxieties in parents about their child being alone with an unknown adult driver, which makes the driverless Waymo seem the safer alternative.

Teens flock to Waymo

Autonomous surgical robot performs gall bladder surgery

Researchers from Johns Hopkins have developed an AI-enabled surgical robot that performs gall bladder surgery on pigs, aided only by coaching from human surgeons via a voice interface. Known as SRT-H, the robot was trained on YouTube videos of surgical procedures, then was coached through the surgeries on pigs in real time. The robot achieved a 100% success rate on the surgeries, responding appropriately to unexpected events that are typical of actual procedures performed on humans. In no case did the physician supervisors need to intervene directly. These results indicate that the day of fully autonomous robotic surgery with human supervision may be nearer than previously thought.

Johns Hopkins surgical robot trained on YouTube videos

FDA clears AI neurostimulator wristwatch to combat essential tremor

University of Minnesota spinout Fasikl has announced that the FDA has cleared its AI-powered neurostimulator wristwatch for reducing essential tremor, a common age-related neuromuscular condition that causes involuntary shaking of the hands. Affecting an estimated 7 million patients in the US, essential tremor can cause difficulty with daily tasks such as buttoning a shirt, holding a glass or cup while drinking, or even signing one’s name. Fasikl’s device delivers noninvasive, AI-powered, individualized neuromodulation which has been shown to interrupt tremors and improve patients’ ability to perform daily tasks. The company plans to roll out availability of the device in the US in phases over the next year and a half.

FDA clears AI neurostimulator wristwatch to combat essential tremor