Tech Brew Ride Home - Tue. 06/03 – Framing The AI Debate

Episode Date: June 3, 2025

Elon Musk is suddenly fundraising everywhere. An attempt to solve the nomenclature problem around hacking groups. Is the solution to more energy for data centers already hidden inside the grid? And the final two pieces today are two different takes on the great AI debate, our entire civilization is having right now. Sponsors: Oracle.com/techmeme Links: Musk Taps Investors for Billions Days After Washington Exit (Bloomberg) 'Forest Blizzard' vs 'Fancy Bear' - cyber companies hope to untangle weird hacker nicknames (Reuters) Gridcare thinks more than 100 GW of data center capacity is hiding in the grid (TechCrunch) Walmart is supercharging revenue — but with fewer workers (Financial Times) My AI Skeptic Friends Are All Nuts (Thomas Ptacek) Learn more about your ad choices. Visit megaphone.fm/adchoices

On April 4th, 2023, around 2 in the morning, a man was found stabbed multiple times on a sidewalk in downtown San Francisco. Hey, who did this to you? What happened next turned the story into a political firestorm. Reports have identified the victim as Bob Lee, the founder of Cash App. From Bloomberg Podcasts, this is Foundering, the Killing of Bob Lee, beginning April 16. Welcome to the Tech meme right home for Tuesday, June 3, 2025. I'm Brian McCullough today. Elon Musk is suddenly fundraising everywhere.

An attempt to solve the nomenclosure problem around hacking groups is the solution to more energy for data centers already hidden inside the grid. And the final two pieces today are two different takes on the great AI debate our entire civilization is having right now. Here's what you missed today in the world of tech. A whole bunch of Elon funding news. All of the sudden, sources say Elon Musk is selling $5 billion and X-A-I debt at a double-digit interest rate via Morgan Stanley with demand already over $3.5 billion and commitments due by June 17th. Quoting Bloomberg, Musk appears eager to refocus on his array of

businesses after announcing last week that he would be stepping back from politics. He had spent months as a senior advisor and regular companion to President Donald Trump, for whom he campaigned in the 2024 election and was a top financial supporter. The debt package includes a floating rate term loan, a fixed-rate term loan and senior secured notes, said the people who are not authorized to share the information publicly. The proceeds will go toward general corporate purposes with commitments due June 17th. Early pricing discussions are seven percentage points over the benchmark rate for the floating rate term loan and roughly a 12% yield on the senior notes, different people with knowledge of the matter said. The debt sale has already garnered demand in excess of

$3.5 billion, they added, end quote. In addition, XAI has launched a $300 million tender offer at a $113 billion valuation validating XAI's price tag for when it acquired XAI. in March of 2025, this will let existing staff sell shares, quoting the FT. Musk launched XAI in 2023 to take on Sam Altman's Open AI and other big tech rivals. It quickly unveiled the Grock Chatbot and built a supercomputer cluster dubbed Colossus, one of the biggest AI Data Center projects in the U.S. The AI startup obtained a $45 billion valuation in a $5 billion private funding round late last year. Musk last year granted investors that backed his Twitter acquisition, including large venture

capital firms such as Sequoia Capital and Andreson Horowitz, 25% of the shares in XAI. The tie-up between the companies was discussed with input from only a few close Musk confidants, according to a number of backers of X and XAI, end quote. Also, Musk's Neurrelink raised a $650 million series E from ARC founders fund, G42, human capital, Sequoia, Thrive, and others reportedly at around a $9 billion pre-money valuation, quoting TechCrunch. Neurrelink last raised a $280 million Series D funding round in 2023, with an additional $43 million

trunch added months later. Since then, Norrelink's brain chip technology has made some significant leaps. The company says it has now conducted more human clinical trials implanting its brain chips in five individuals with severe paralysis. In May, Neurrelink received breakthrough device designation from the U.S. Food and Drug Administration, a program intended to speed up development, assessment, and review processes for experimental technologies, end quote. Microsoft, Google, CrowdStrike, and Palo Alto Networks plan to create a public glossary of state-sponsored hacking groups to ease unofficial alias confusion. Quoting Reuters, cybersecurity companies have long assigned coded names to hacking groups, as attributing hackers to a country or an organization can be difficult, and researchers need a way to describe who they are up against. Some names are dry and functional, like the APT1 hacking group exposed by cybersecurity firm Mandient or the TA4.

or 53 group tracked by ProofPoint. Others have more color and mystery, like the Earth Lamia group tracked by Trend Micro or the Equation group, uncovered by Kaspersky. Crowdstrikes evocative nicknames Cozy Bear for a set of Russian hackers or kryptonite panda for a set of Chinese ones have tended to be the most popular, and others have also adopted the same kind of offbeat monikers.

In 2016, for example, the company SecureWorks, now owned by Sophos, began using the name Iron Twilight for Russian hackers it previously tracked as TG4127. Microsoft itself recently revamped its nicknames, moving away from state element-themed ones like Rubidium to weather-themed ones like Lemon Sandstorm or Sangria Tempest. But the explosion of whimsical aliases has already led to overload. When the U.S. government issued a report about hacking attempts against the 2016 election, it sparked confusion by including 48 separate nicknames attributed to a grab bag of Russian hacking groups and malicious programs including Sofacy, Pondstorm, Chopstick, SAR team, and Onion Duke.

Michael Sikorsky, the chief technology officer for Palo Alto's threat intelligence unit, said the initiative was a game changer. Disparate naming conventions for the same threat actors create confusion at the exact moment defenders need clarity, he said. Juan Andreas Garrow Saad, executive director for intelligence and security and research at cybersecurity firm Sentinel One, was skeptical of the effort saying the cold reality of the cybersecurity industry was that companies hoarded information. Unless that change, he said, this is branding, marketing, fairy dust sprinkled on top of business realities. But Crowdstrike Senior Vice President

of Counter-Adversary Operations Adam Myers said the move had already delivered a win by helping his analysts connect a group Microsoft called Salt Typhoon with one CrowdStrike dubbed Operator Panda, end quote. We've discussed before how energy usage is exploding because of AI adoption, and this has put a strain on energy production generally. But what if a solution is already hidden inside the grid? Quoting TechCrunch. Hyper-scalers and data center developers are in a pickle. They all want to add computing power tomorrow,

but utilities frequently play hard to get, citing years-long waits for grid connections. All the AI data centers are struggling to get connected, amid Narayan, founder and CEO of GridCare, told TechCrunch, they're so desperate, they are looking for solutions which may or may not happen, certainly not in the five-year timelines they cite. This has led many data centers to pursue what's called behind the meter power sources. Basically, they build their own power plants, a costly endeavor that hints at just how desperate they are for electricity.

But Narayan knew there was plenty of slack in the system, even if utilities themselves haven't discovered it yet. He has studied the grid for the last 15 years, first as a Stanford researcher and then as a founder of another company. How do we create more capacity when everyone thinks there is no capacity on the grid, he said. Narayan said that GridCare, which has been operating in stealth, has already discovered several places where extra capacity exists and it's ready-to-play matchmaker between data centers and utilities. GridCare recently closed and oversubscribed 13 and a half million dollar seed round. The company told TechCrunch for Narayan and his colleagues at GridCare, the first step to finding untapped capacity was to map the existing grid. Then the company used generative AI to help forecast what changes might be implemented in the coming years. It also layers on other details, including the availability of fiber optic connections, natural gas, water, extreme weather, permitting, and community sentiment around data center construction and expansion.

We'll find out where the maximum bang is for the buck, Narihan said. At the same time, GridCare works with hyperscalers and data center developers to identify where they are looking to expand operations or build new ones. They have already told us what they're willing to do. We know the parameters under which they can operate, he said. That's when the matchmaking begins. grid care sells its services to data center developers, charging them a fee based on how many megawatts of capacity the startup can unlock for them. That fee is significant for us, but it's negligible for data centers, Nariang said.

For some data centers, the price of admission might be foregoing grid power for a few hours here and there, relying on-site backup power instead. For others, the path might be clearer if their demand helps greenlight a new grid-scale battery installation nearby. In the future, the winner might be the developer that is willing to pay more. Utilities have already approached GridCare inquiring about auctioning access to newfound capacity. Regardless of how it happens, Narayan thinks that GridCare can unlock more than 100 gigawatts of capacity using its approach. We don't have to solve nuclear fusion to do this, he said, end quote.

So look, the biggest debate in tech right now, the biggest debate in society really is AI, right? Is it hype? Is it real? Is it stealing jobs? Is it making jobs better? There's no way I can do this show without sharing all. all sides of this debate with you. I can't ignore that the debate is going on. I want to underline again. I'm not taking any sides. I'm just presenting the debate, so you're in the no. So the second half of the show today is two stories looking at parts of this debate from different sides. Let's start with this one from the Financial Times. It could be summed up like this. Walmart has 70,000 less employees than it did three years ago. During the same period,

revenue grew 36%. So as revenue increased, headcount has reduced and Walmart expects automation from robotics to AI to keep that trend of lower headcount continuing. Quote, Walmart executives are aiming to grow sales by 4% a year, but they do not expect to significantly increase headcount. The employee figures raise questions about the future of labor in the U.S. retail trade, which employs 1 in 10 American workers, and provides broad avenues of promotion for those without university degrees. About 1.6 million of Walmart's employees are in the U.S. a figure that has barely budged in the past decade. Wall Street analysts say the company's expansion without job creation reflects a hard push into e-commerce and the automation of cumbersome tasks from unloading

shipping pallets to updating shelf price labels. Artificial intelligence is poised to supercharge these efforts. Walmart executives say the technology investments mean new roles for workers, not fewer workers. Tasks will get automated, jobs will change, and many years from now will still employ a large number of people and be happy to do so. Chief Executive Doug McMillan said at an investor event in April this week. Walmart is hosting 13,000 employees and shareholders at its Associates Week event, an annual ritual in its home of Northwest Arkansas. But critics say workers are missing out. Net sales at Walmart U.S. have risen by 36% in the past five years, while average U.S. hourly wages have increased $28% to $18.25. $0. Walmart's jobless growth is a continuation of a pernicious trend that Walmart itself helped pioneer, squeezing more output from each hour of labor and growing sales faster than wages, said John Marshall,

Capital Strategies Director at Local 3,000 Division of the United Food and Commercial Workers, which ended an unsuccessful effort to unionize the company a decade ago. Longer term, the trajectory is very clear. I think most retailers want to automate a lot of different functions within their operations because labor is a very, very costly part of doing business, said Neil Sanders, a retail analyst at global data. We've seen Walmart really lean heavily on that, end quote. Walmart in April showed off labor-saving technologies to investors and media at two new warehouses outside Dallas, one a cold storage hub for foods, the other a fulfillment center to enable speedy deliveries for e-commerce customers, end quote. And then from a different

perspective on his blog, software developer Thomas Paycheck. Sorry if I miss pronouncing that, Thomas outlines what he says are the very real benefits of using LLMs to write code. Quote, some of the smartest people I know share a bone-deep belief that AI is a fad, the next iteration of NFT mania. I've been reluctant to push back on them because, well, they're smarter than me, but their arguments are unsurious and worth confronting. Extraordinarily talented people are doing work that LLMs already do better out of spite. All progress on LLMs could halt today, and LLMs would remain the second. most important thing to happen over the course of my career. Important caveat. I'm discussing only the

implications of LLMs for software development. For art, music, writing, I got nothing. I'm inclined to believe the skeptics in those fields. I just don't believe them about mine. First, we need to get on the same page. If you're trying and failing to use an LLM for code six months ago, you're not doing what most serious LLM-assisted coders are doing now. People coding with LLMs today use agents. agents get to poke around your code base on their own. They author files directly. They run tools. They compile code, run tests, and iterate on the results. They also pull in arbitrary code from the tree or from other trees online into their context windows. Run standard Unix tools to navigate the tree and extract information. Interact with Git. Run existing tooling like linters, formatters, and model checkers, and make essentially arbitrary tool calls that you set up through MCP. If you're making requests on a chat GPT page, and then pasting the resulting broken code into your editor, you're not doing what the AI boosters are doing. No wonder you're talking past each other.

LLMs can write a large fraction of all the tedious code you'll ever need to write. And most code on most projects is tedious. LLMs drastically reduce the number of things you'll ever need to Google. They look things up themselves. Most importantly, they don't get tired. They're immune to inertia. Think of anything you wanted to build but didn't. You tried to home in on some first steps,

If you'd been in the limerent phase of a new programming language, you'd have started writing, but you weren't, so you put it off for a day, a year, or your whole career. I can feel my blood pressure rising, thinking of all the bookkeeping and Googling and dependency drama of a new project. An LLM can be instructed to just figure all that S out. Often it will drop you precisely at that golden moment where the S almost works, and development means tweaking code and immediately seeing things work better, that dopamine hit is Y-I-Code. end quote. He goes on to address several points. It's a long essay. I encourage you to read it,

but essentially his points are this. If you're merging code without reading it, whether it came from an LLM or not, that's on you. Developers have always been responsible for what hits main. LLMs don't absolve you of that duty. They generate code you should read, tweak, and make your own. Complaints about probabilistic code. Miss the point. Code is deterministic. It's your job to understand and validate it regardless of how it was written. Yes, some of that code is mediocre. guess what? So is a lot of human written code. LLMs raise the floor, even if they don't raise the ceiling. You're still the one making decisions, shaping architecture, and fixing the important stuff. If your problem is that LLMs aren't good at Rust or your artisanal craft of beautiful

code, cool, enjoy the hand tools. But don't confuse a personal aesthetic with professional utility. Developers are here to solve problems, not sculpt masterpieces. And no, he doesn't care if it's not aGI.I. It just works for him. Let me quote from his conclusion. There's plenty of things I can't trust in LLM with. No LLM has any of access to prod here, but I've been first responder on an incident and fed 4-0, not 40, many 40, log transcripts, and watched it in seconds spot LVM metadata corruption issues on a host we've been complaining about for months. Am I better than an LLM agent at interrogating open search logs and honeycomb traces? No, no, I am not. To the consternation of many of my friends, I am not a radical

or a futurist. I'm a statist. I believe in the haphazard perseverance of complex systems, of institutions, of reversions to the mean. I write go in Python code. I'm not a Kool-Aid drinker, but something real is happening. My smartest friends are blowing it off. Maybe I persuade you, probably I don't, but we need to be done making space for bad arguments, end quote. Nothing more for you today. Talk to you tomorrow.