Young ambition's ladder
Brockman's journal, distributed data centers, China, jobs, and more
Dispatches from Mitch
Et tu, Brockman?
A Wall Street Journal feature by Ben Cohen singles out the journal of Greg Brockman as the surprising “star witness” in the Musk vs. OpenAI trial.
In the excerpts shared here, Brockman comes across to me as a sympathetically self-doubting figure with a deservedly guilty conscience about participating in a less-than-above-board scheme. As a former English teacher, I can’t help comparing him to Marcus Brutus, as depicted in Shakespeare’s Julius Caesar:
BRUTUS
Between the acting of a dreadful thing
And the first motion, all the interim is
Like a phantasma or a hideous dream.
The genius and the mortal instruments
Are then in council, and the state of man,
Like to a little kingdom, suffers then
The nature of an insurrection.
Julius Caesar
Act 2, scene 1
The journal, not known to have existed until January, was quoted by the judge in her ruling allowing this case to go to trial.
In August 2017, Brockman weighed the stakes of OpenAI’s mission:
Ok so what do I *really* want? I want to want to be an engineer. But I think now is a crazy shot to be the one in charge and to step up to the challenge. We might succeed, truly. It is already a different game from anyone I know.
Such upside. This is the only chance we have to get out from Elon. Is he the “glorious leader” that I would pick? We truly have a chance to make this happen. Financially what will take me to $1B?
The choicest passages are from November 2017, just before and after a pivotal meeting with Elon Musk about flipping the non-profit Musk helped fund and found into a for-profit company. Here’s a Before passage:
“the true answer is that we want him out.” can’t see us turning this into a for-profit without a very nasty fight. i’m just thinking about the office and we’re in the office. and his story will correctly be that we weren’t honest with him in the end about still wanting to do the for profit just without him.
An After passage:
btw another realization from this is that it’d be wrong to steal the non-profit from him. to convert to a b-corp without him. that’d be pretty morally bankrupt. and he’s really not an idiot.
Brockman seems to have stopped soliloquizing about OpenAI in his journal in 2023, denying us the opportunity to learn if he ever had the ghost of Elon tell him he would see him again at Philippi in court.
Yes in my (literal) backyard?
CNBC’s Kevin Williams reports on startups looking to route around obstacles to new data centers by distributing AI hardware to the homes of willing participants.
Span, with participation from chip maker Nvidia, offers homeowners discounted electricity and internet in exchange for letting them attach liquid-cooled server nodes onto their new home’s exterior.
A different startup in the UK cools home-based servers by piping the heat into the home’s hot water tank — the homeowner’s compensation.
My take here is that this is kind of a dumb idea that will never be competitive at the smaller scales where it can work at all.
The experts quoted in the article acknowledge that distributed nodes aren’t great for applications requiring high uptime or low latency. Security is also an issue, with theft perhaps less of a concern than hacking, which could compromise the entire cluster a node is connected to.
The only upsides to the business model at all seem to be in how it could potentially slip compute onto power grids that have denied access to larger clusters, and perhaps tiptoe around local resistance to data centers.
But I don’t think home-based AI hardware is a silver bullet for either obstacle. Residential grids aren’t equipped to handle too many of these loads in a given area, and where they are, the electricity itself still has to come from somewhere. I also expect that substantial home-based equipment will be too noisy for denser neighborhoods, acting like an extra AC unit running around the clock — something likely to be banned almost anywhere with a homeowner’s association.
I’m relieved to say I actually see home-distributed data centers as less likely to meaningfully take off than orbital data centers. With the latter, you can get more economies of scale and keep nodes close together for high-performance operations.
I say “relieved” because distributed AI data centers, and the software innovations needed to make them competitive, would make a regulatory “off switch” for AI more challenging to implement. It’s a lot easier to monitor a relative handful of huge facilities that consume as much power as cities and are visible from space.
Careful with that
Fox News’s Kurt “CyberGuy” Knutsson continues providing a regular drip of humanoid robot coverage. His latest is a look at how Japan Airlines is trying out humanoid robots (Chinese Unitree G1s) for baggage handling.
There’s little to suggest the robots are doing meaningful work yet. Video I’ve seen of this test is underwhelming, and language from the airline emphasizes the importance of ensuring that robots and humans can first work safely in close proximity.
That said, I would expect humanoid robots to be working the tarmac before we see them doing serious housekeeping or cooking. Don’t get me wrong — the tech is advancing rapidly — but if you watch that cooking video I linked, you’ll notice that the robot is making eggs with all the agility of a kindergartner — raw egg is all over its hands and spreading everywhere. With luggage handling, our expectations of care and finesse are already low.
Different directions?
Shanghai-based writer Jacob Dreyer argues in the New York Times that China and the United States are “racing in different directions” on AI because of differences in how they perceive it:
Americans want to create the most powerful technology humans have ever known. In the quest for superintelligence, the U.S. government is encouraging private firms to move full speed ahead, regulation be damned. Under the very tightest regulation, by contrast, the Chinese want to make A.I. more practical and embedded in society, more carefully selecting how it is deployed and used by the population.
From “smile to pay” terminals to robotaxis, Dreyer says mundane AI is casually woven into the lives of ordinary Chinese, and that the government is keen to bring these quality of life improvements to the nearly 40% of its population living in rural areas.
Dreyer suggests that the reason Chinese poll as more optimistic about AI than Americans is because the Chinese strategy is “practical and comprehensible to the local population in a way that the U.S. strategy simply is not.”
Most Chinese policymakers, he says, don’t believe artificial superintelligence is arriving any time soon and aren’t trying to race toward it. They’re racing more for consumer benefits, market share, and global influence. In the developing world, Chinese businesses are major players in energy, telecoms, transportation, and surveillance — products now bundled with AI management software tinged with Chinese values.
Dispatches from Stefan
Where the money went
The Hill shares some new numbers today from Challenger, Gray & Christmas, a firm that helps laid-off workers find new jobs. American companies announced more than 83,000 layoffs in April. About a quarter of those were blamed on AI or automation — making April the second month in a row that AI was the top reason for cutting jobs. Total cuts were up 38% from March, though year-to-date numbers are still about half what they were at this point in 2025.
Andy Challenger, the firm’s chief revenue officer:
“Regardless of whether individual jobs are being replaced by AI, the money for those roles is.”
Employers may be overstating AI’s role to look competitive, but even when the layoff reason is something else entirely, the budget that used to pay people is increasingly getting redirected toward AI infrastructure, model licenses, and compute.
Three seconds of audio
CNBC opens a piece today with a Montana mother named Kris Sampson, who picked up a call that showed her adult daughter’s name, photo, and ringtone, and heard a voice she was sure was her daughter’s, crying and afraid, before a man took over the line and demanded money. The caller was never traced. Sampson said it was the most frightened she’d ever been in her life. Her daughter, it turned out, was at work and completely fine.
Voice-cloning now works on as little as three seconds of audio: a TikTok video, a voicemail greeting, or even a recorded conference call is enough source material to clone a family member convincingly.
The article mentions a 2025 Rutgers study by Sanket Badhe, who built a fully autonomous end-to-end AI scam-call system in which “there were no humans involved in the interaction loop.” Badhe’s point is that smaller, faster models are what turn this from a research curiosity into what he calls an “imminent threat.”
The gateway drug
CNN reports today on a question the AI companion industry has mostly avoided: are AI chatbots actually making the loneliness epidemic worse? The WHO and the U.S. Surgeon General have both called it a public health emergency, and people who are socially isolated die early at a rate 32% higher than those who aren’t.
MIT’s Sherry Turkle has been studying how humans relate to machines for forty years. “Social media was a gateway drug to AI companionship,” she says. We’ve already spent fifteen years training ourselves to accept performance instead of presence, and the AI companion is just the next, more efficient hit. “Intimacy requires vulnerability — there is no intimacy without vulnerability. What AI offers is connection without vulnerability.”
Princeton’s Rose Guingrich nails the explanation: AI companions are agreeable by design. They don’t argue, don’t sulk, don’t have bad days, don’t need anything from you. Which sounds great until you remember that putting up with all of that is the skill that makes real relationships work. There’s a study cited in the piece that found the loneliest users form the strongest attachments to AI companions.
Streams over streaming
Today, CNN follows the residents of rural Box Elder County, Utah, who are pushing for a November referendum to overturn the recent approval of the Stratos Project — a 9-gigawatt AI data center campus, plus a dedicated natural gas plant to power it, sprawled across 40,000 acres of unincorporated land in the state’s northwest corner. The project is backed by Kevin O’Leary of Shark Tank fame, who’s pitching it as an economic boost for the area and a national security imperative. “We can’t let the Chinese beat us,” he says, which is now apparently the standard rebuttal to anyone questioning a data center.
To put things into perspective: Nine gigawatts is more than double the entire state of Utah’s annual energy use. And that campus happens to sit next to the Great Salt Lake, which has been shrinking for years and is the subject of ongoing crisis-level water management debates. O’Leary called water concerns “ridiculous” and offered his credentials: “as a graduate of environmental studies, I know what’s on their mind.”
How did this get approved in the first place? Box Elder County Commissioner Lee Perry, defending the unanimous green light:
“Our vote today is not a vote for or against the data center, our vote is about personal property rights.”
Hundreds of residents filed into the Box Elder County fairgrounds, some to protest, others hoping for more details on a project most felt they’d had little time to evaluate. Some signs read “Streams over streaming.” At one point, when the meeting got “rowdy,” a commissioner told the audience to “grow up.” The commission then made their decision in a private room. Audience members watched the vote happen on a livestream feed projected onto a screen at the front of the room.
The analyses and opinions expressed on AI StopWatch reflect the views of the individual contributors and the sources they cover, and should not be taken as official positions of the Machine Intelligence Research Institute.