Brian McCullough (0:04)

Welcome to the Techmeme Write Home for Thursday, July 10th, 2025. I'm Brian McCullough. Today Grok4 is out +grok4 voice grok4 code a bunch more okay, we know Zuck wants to catch up on AI, but it insider explains why they think Meta fell behind in AI in the first place. Are AI web browsers the next battlefield in the AI wars? And if your electricity bill goes up substantially this summer, you'll never guess what you can blame. You can probably guess. Here's what you missed today in tech XAI has rolled out Grok 4 trained on its Colossus supercomputer, featuring multimodal Tools, faster reasoning, Grok4Voice, Grok4Code, a new interface, and more. According to artificial analysis benchmarks, Grok 4 is now the leading AI model, a first for XAI as a company, though they also note that Grok 4's per token pricing is more expensive than Gemini 2.5 pros and O3's quoting Tom an hour after the live stream was supposed to start last night, July 9, Elon Musk and a few members of his XAI team introduced us to Grok 4 on Xai. The long winded announcement shared the news of multimodal features, faster reasoning and an upgraded interface, something Musk compared to an era of, quote, Big Bang intelligence. On paper, Grok 4 is Musk's most ambitious AI model yet. The model is expected to rival OpenAI's GPT4 and Anthropic's Claude 4 opus, both of which have recently dominated headlines for their Real time Speed reasoning and advanced vision. We've run out of test questions to ask, musk boasted during the launch, adding reality is the ultimate reasoning test. But Grok isn't just trying to compete, it's trying to survive a credibility crisis. The platform's unfiltered free speech approach has led to concerning outputs, including racist and biased content that circulated widely over the weekend. That's raised big questions about how much testing and guardrailing XAI has actually done, especially as it rushes toward real time human like interaction. Here's a rundown of what Grok 4 has advanced reason Grok 4 has been trained on Xai's Colossus supercomputer for advanced scientist grade reasoning. The model promises stronger logical reasoning and text generation. It uses the same model physicists use. A coding focused variant is also included. Developers will appreciate Grok4Code, a specialized model designed to write, debug and explain code more efficiently, echoing tools like GitHub, Copilot or GPT4 code interpreter. Multimodal capabilities Grok4 is expected to support not only text, but images and possibly video, with Musk mentioning that's one of their biggest weaknesses. Better multimodal capabilities would bring the model closer in competition with OpenAI's and Google's Gemini. Grok 4 may support video at some point. Voice features Grok 4 voice features a natural, human like voice with fewer interruptions. Real time Web Access like earlier Grok models, Grok 4 features deep search, a tool that pulls in live data from the web, especially from Musk's X platform. That means Grok can provide up to date results during chats, no separate tab or browser needed. Cultural Fluency and meme smarts Grok's biggest differentiators is its understanding of Internet culture. Grok 4 is being tuned to interpret memes, slang and humor with high accuracy, potentially making it one of the most online AI assistants yet, End quote and quoting the Verge during Wednesday's livestream, Musk said he's been at times kind of worried about AI's intelligence far surpassing that of humans and whether it will be, quote, bad or good for humanity. I think it'll be good, most likely it'll be good, musk said. But I've somewhat reconciled myself to the fact that even if it wasn't going to be good, I'd at least like to be alive to see it happen. I would expect Grok to discover new technologies that are actually useful no later than next year and maybe the end of this year, musk said. It might discover new physics next year. Let that sink in. Xai also launched Grok 4 Heavy, a multi agent version of Grok 4, offering increased performance, and unveiled a new $300 per month AI subscription plan called Super Grok Heavy. Still covering this whole AI talent free for all. I've honestly never in 25 years in this industry seen hiring in the tech industry go like this. Bloomberg says that Meta offered Rooming Pang a more than $200 million pay package over a several year period, in line with Meta's other AI hires. Apple apparently did not try to match the the compensation packages for hires at Meta's Superintelligence Labs or msl, are comprised of a base salary, a signing bonus and Meta shares, with the stock as the weightiest part of the package. The salary and bonus to join are often significant cash payments. In cases where a recruit would have to walk away from significant startup equity to join Meta, the signing bonus may be higher. To account for that lost opportunity, the people said for the stock portion, Meta tends to put in its contracts that the payouts are tied to specific metrics, such as Meta's stock growing by at least a certain percentage in a given year, the people said. In many cases, the new hires are agreeing to join on contracts that exceed a typical four year vesting schedule for stock, they added. In a separate report, Bloomberg says Mark Zuckerberg's hiring blitz for top AI talent is spurring other leading AI scientists to join Meta. Like if they assemble a true a team, a players might want to join that team, I guess. Of course money is a great motivator. But many of these researchers are already wealthy and their field is so ideologically charged and so close knit that they're motivated by the glory of being published in Nature or having a hand in the biggest new AI model just as much as they are by the prospect of yachts and mansions. Zuckerberg's public commitment to open source AI with his llama model has already attracted scientists who believe such systems can have a more democratizing impact if they're free for all. OpenAI made a similar bet early on, sharing much of its research freely for recruitment purposes, according to its then chief scientist Elias Suskever, before taking that work behind closed doors. Many researchers reckon AI can eventually solve intractable human problems like aging, climate change and cancer, and that overwhelmingly through history, technology has been a net good for humanity. But for many, the desire to build that technology first is even more powerful, a dynamic not so different to the field of cancer research, where scientists want to win the race to a cure as much as they want to find cures at all. The key catnip Zuckerberg is offering researchers is vast amounts of compute AI scientists need huge arrays of powerful AI chips and data farms to develop the most cutting edge models models, and with it, the chance of being first to cross the superintelligence threshold, which Zuckerberg likely cares about as much as anyone else in the field. End quote. Now a bit of counter to that. In an internal Meta essay, an outgoing Meta AI researcher says Meta's culture of fear, their words with performance reviews and rolling layoffs, is like a metastatic cancer. Quoting the information, one outgoing research scientist from Meta's generative AI group has come up with his own diagnosis of the company's AI problems, and his assessment isn't pretty. In a more than 2,000 word essay that has circulated inside Meta in recent days, the research scientist Tigman Blankvoort paints a bleak picture of cultural and organizational dysfunction inside Meta that he argues has stymied its progress in AI. I have yet to meet someone in Metagenai that truly enjoys being there, someone that feels like they want to stay in Meta for a long time because it's such a great place, wrote Blankfort, referring to the nearly 2,000 person group that develops Meta's flagship AI model, Llama. You'll be hard pressed to find someone that really believes in our AI mission. To most, it's not even clear what our mission is. The essay, a copy of which the information reviewed reflects the perspective of one individual. But the concerns of the author who worked on Llamata echo those other members of Meta's AI team have expressed in recent months about infighting, lack of focus and burnout in the group. Many of those current employees, as well as Meta alumni have been buzzing in recent days about the essay, which Blankfort posted in an internal Meta bulletin board system called Workplace, according to two people familiar with the essay. I've seen too many brilliant people who absolutely thrived in other companies come and go, blankvoort wrote. Mostly this is because of our culture. Many of the people that left hate Meta's generative AI group with vigorous intent. Instead of being motivated by our mission or an aspiration to build something great, many people I've talked to are motivated by the fear of getting fired, blankfort wrote. This attitude runs through the company like a metastatic cancer. It has led to a culture of every person for themselves, as well as, quote, land grabbing, project sniping, stealing work, he added. Another problem with Meta's AI efforts, according to Blankvort, is a lack of vision. Over the last two years, the company's generative AI group has accumulated a list of sometimes competing responsibilities, including llama, its Meta AI assistant, and other AI products and features across its social media apps and devices. In Contrast, much of OpenAI's attention is on its hit ChatGPT. Chatbot and anthropic has been hyper focused on developing the best AI for coding, resulting in an explosion of revenue growth this year. After sharing the essay, Meta's AI leadership read it and reached out to me right away and very positively so, blankfort said. They are very eager to improve. I believe the AI organization just tried to run too fast for too long and they have realized that this needs to turn around a bit for the org to be effective. End quote. Ever wonder what ChatGPT and Claude are actually doing with your conversations? Have you ever even stopped to think about that? Know Alexa listens to us and recommends products based on our conversations? Meta retargets us based on our browsing and engagement history. But now, in this new AI era, there's a new privacy problem to consider. Think about what else we tell these AI platforms. Our thoughts, our dreams, sensitive questions, business ideas, et cetera. They take all this information, tie it to your identity, and then sell it to various third parties and governments. ChatGPT literally has the former director of the NSA sitting on their board right now. That's why I've started using Venice AI who is sponsoring today's podcast. Venice AI is a generative AI platform that is private and permissionless. They utilize leading open source AI models to deliver text code and image generation to your web browser. There's no downloads, no installations of anything. Venice AI doesn't spy on you or censor the AI messages are encrypted and your conversation history is stored only in your browser. This is a cause I can get behind. If you want to join this too, and you want to use AI without fear of handing over your most intimate thoughts to a corporation or the government, you can get 20% off a pro plan using my link Venice AI and CodeTechmeme that is Venice AI, Techmeme and CodeTechmeme thus far, summer in New York City has been hella rainy and now the humidity is really kicking into high gear. Fortunately, Mack Weldon has everything I need to stay cool, comfortable and stylish regardless of the temperature. Their new Tech Linen line combines the easy charm of classic linen with coolmax technology to help you look and feel your best. These are genuinely breezy but solid looking clothes. Looking confident doesn't have to mean calling attention to your clothes. Mack Weldon balances classic pieces with updated details to keep you looking sharp. They're not flashy, just classic. Always in style and made from the world's most comfortable performance materials, Mack Weldon clothes are designed to fit your style and the demands of modern life. They look like regular clothes but feel like the latest in modern comfort. They're the go to choice for guys who want to look great without even trying to explore. Polos, sweaters, hoodies and more@mackweldon.com and get 25% off your first order with code Brian get your closet ready for summer. Go to mackweldon.com and get 25% off your first order of $125 or more with promo code Brian that's M A C K W E l d o n.com promo code b r I a n Sources say OpenAI plans an AI powered web browser with some user interactions within a ChatGPT native chat interface so that you would use that instead of clicking through to websites, quoting Reuters. The browser is slated to launch in the coming weeks, three of the people said, and aims to use artificial intelligence to fundamentally change how consumers browse the web. It will give OpenAI more direct access to a cornerstone of Google's success user data. If adopted by the 500 million weekly active users of ChatGPT, OpenAI's browser could put pressure on a key component of rival Google's ad money spicking Chrome is an important pillar of Alphabet's ad business, which makes up nearly three quarters of its revenue as Chrome provides user information to help Alphabet target ads more effectively and profitably, and gives Google a way to route search traffic to its own Engine. By default, OpenAI's browser is designed to keep some user interactions within a chatgpt like native chat interface instead of clicking through to websites, two of the sources said. The browser is part of a broader strategy by OpenAI to weave its services across the personal and work lives of consumers, one of the sources said. OpenAI has its work cut out for its Google Chrome, which is used by more than 3 billion people, currently holds more than two thirds of the worldwide browser market, according to web analytics firm Statcounter. Apple's second place, Safari, lags far behind with a 16% share. Last month, OpenAI said it had 3 million paying business users for ChatGPT. Perplexity, which has a popular AI search engine, launched an AI browser, Comet, on Wednesday capable of performing actions on a user's behalf. Two other AI startups, the browser company and Brave, have released AI powered browsers capable of browsing and summarizing the Internet. Chrome's role in providing user information to help Alphabet target ads more effectively and profitably has proven so successful that the Department of Justice has demanded its divestiture after a US Judge last year ruled that Google holds a unlawful monopoly on online search. OpenAI's browser is built atop Chromium, Google's own open source browser code, two of the sources said. Chromium is the source code for Google Chrome as well as many competing browsers, including Microsoft's Edge and opera. Last year, OpenAI hired two longtime Google vice presidents who were part of the original team that developed Google Chrome. The information was first to report their hires and that OpenAI previously considered building a browser. End quote from the AI job pocalypse may be hitting big tech first file A source says that during a presentation, Microsoft Chief Commercial Officer Judson Althoff said AI saved Microsoft more than $500 million last year in call centers and user satisfaction activities, quoting Bloomberg. The company is also starting to use AI to handle interactions with smaller customers, Altoff said. This effort is nascent but already generating tens of millions of dollars, he said. Microsoft declined to comment. At Microsoft, AI generated 35% of the code for new products, accelerating launch times, Altoff said. The company's GitHub Copilot is a leader in the market for AI coding tools and has 15 million users, Microsoft said in April. AI implementation has fueled replacement anxiety for many workers, particularly in the tech industry. Microsoft has announced cuts of about 15,000 employees this year with a wave of layoffs last week targeting customer facing roles like sales. Altoff stressed to employees that AI could make them more effective as sellers through the use of Microsoft's Copilot AI Assistant. Each salesperson is finding more leads, closing deals quicker and generating 9% more revenue, he said. Productivity gains from AI were were not a predominant factor in the job reductions of recent months, Microsoft's top lawyer Brad Smith said Wednesday during an event announcing a donation of over $4 billion in cash and technology to schools with a focus on spreading AI skills. End quote or how about a utility bill apocalypse? Reuters says Electricity bills may rise more than 20% in parts of PJM interconnections 13 state territory as the largest US power grid struggles to meet AI and data center demand Quote Electricity bills are projected to surge by more than 20% this summer in parts of PJM Interconnections territory, which covers 13 states from Illinois to Tennessee, Virginia to New Jersey. Serving 67 million customers in a region with the most data centers in the world, the governor of Pennsylvania is threatening to abandon the grid. The CEO has announced his departure, and the chair of PJM's board of managers and another board member were voted out. The upheaval at PJM started a year ago with a more than 800% jump in prices at its annual capacity auction. Rising prices out of the auction trickle down to everyday people's power bills. Now, PJM is barreling towards its next capacity auction on Wednesday, when prices may rise even further. The auction aims to avoid blackouts by establishing a rate at which generators agree to pump out electricity during the most extreme periods of stress on the grid, usually the hottest and coldest days of the year. High prices out of the auction should spur new power plant construction, but that hasn't happened quickly enough in PJM's region, as aging power plants continue to retire and data center demands explode. PJM has made the situation worse by delaying auctions and pausing the application process for new plants, according to more than a dozen power developers, regulators, energy attorneys and other experts. Interview interviewed by Reuters. We need speed from pjm, we need transparency from pjm, and we need to keep consumer costs down with pjm, pennsylvania governor Josh Shapiro told Reuters in an interview. I think they've taken some steps in that direction, which is really encouraging to me, and we're going to continue to work at it. Pjm says the supply and demand crunch has been caused largely by factors outside of its control, including state energy policies that closed fossil fuel fired power plants prematurely and data center growth. In Data Center Alley in Northern Virginia and other burgeoning hubs in the Mid Atlantic, prices will remain high as long as demand growth is outstripping supply. This is a basic economic policy, said PJM spokesman Jeffrey Shields. Right now we need every megawatt we can get, end quote. Related the Financial Times says U.S. power providers have sought regulatory approval for $29 billion in rate increases in the first half of this year, up 142% year on following, you guessed it, booming data center demand. Nothing more for you today. Talk to you tomorrow.