Dire warnings from crackpots

Another AI guy has gone rogue while coincidentally building his personal brand. Jacob Coxon has resigned from Anthropic because, as he says on X, the everything including pedophilia app, the frontier labs aren’t “acting responsibly” and are “gambling with our lives.” After making lots of money from this as an employee of both OpenAI and Anthropic, Jacob can’t take it anymore.

I never really know how to respond to things like this because there are many valid reactions, some of which even initially seem to be contradictory. Despite warning against AI, Coxon is making the exact same sorts of bold yet ill-defined predictions that AI boosters do. He says we’re “racing straight to self-improving superintelligence,” which is dubious regardless of whether you believe this is good or bad — and anyway, still elides the basic point of what the hell “superintelligence” really means. Not that the guy really ought to pause in a Twitter thread to define this, but somebody, somewhere, should.

He also claims that “the people building AI earnestly believe that it could kill us all by the end of the decade.” That they believe this, I have no doubt. But could it actually do that? How? I’m not asking rhetorically. What are the mechanisms by which even a self-improving superintelligence could kill us all? What could we do to stop them?

Like all of the most lurid predictions about AI, I would bet that Silicon Valley crackpots see the extermination of (most of) humanity as a feature and not a bug. Imagine a system so powerful that it could end life on earth. Now imagine that you, a dork-ass loser with a punchable face, have created that system. What does it make you? It makes you God, right?

I don’t mean to completely rule out the possibility of AI causing a mass casualty event. Just because something is stupid doesn’t mean it can’t also be dangerous. Some of our biggest dumbasses have been responsible for some of our most painful tragedies. There’s really no reason to think that what we’re calling AI can’t wreak havoc with the best of them. It’s just that it’s probably less “evil genius” stuff and more “dog that’s been trained to run between tank treads with a bomb on its back.”

Still, there’s one more point to make about relative risk. Coxon asserts that “no other human activity poses this level of danger.” That’s an ironic thing to type while inhaling wildfire smoke, but it’s emblematic of the SV way of thinking. Hypothetical future risks are far more pressing than the risks that we are facing today. The people who think it’s easier to make the planet Mars habitable than the planet Earth will of course think that their superintelligent terminator machines are more dangerous than climate change, pandemic, war, and all the other blights presently afflicting us. You have to remember that, in the end, these guys just aren’t very smart. Or maybe it’s more accurate to say they’ve disappeared all the way up their own assholes.

And they’re dragging the rest of us with them.

Tech predictions I won't bother to follow up on

Nvidia's Jensen Huang says 'AGI has arrived' and congratulates OpenAI

Greg Brockman, the company's president, said on a call with reporters on Thursday: "Welcome to the AGI era."

AGI stands for artificial general intelligence, a fuzzy concept that nonetheless has become a core goal for leading frontier AI labs.

OpenAI defines it as "highly autonomous systems that outperform humans at most economically valuable work." The company pitched Astra as a major research breakthrough and a shift in what's possible to delegate to AI. It's rolling out to customers this week.

On the Thursday call, Brockman said that in the future, he thinks people will look back and think AGI was created "about this time, and I think it might be about this model."

"For me personally, I do think we're there," he said.

I’m going back to bed.

Another New England Patriots season is upon us

We’re eight days out from kickoff and I, for one, expect the Patriots to get destroyed by the Seahawks once again. After that? Hard to say. Last year’s team overperformed to such a degree that this year is bound to be a disappointment. It turns out that it’s very hard to win back-to-back AFC championships. Back-to-back divisional titles? Maybe.

There’s no doubt that the wide receiver corps is vastly improved. AJ Brown is easily the best WR this team has had since Randy Moss, and I expect him to have a rebound year. Romeo Doubs is an improvement at WR2, and the other holdovers from last year all offer value. I’ll miss Kayshon Boutte. He was responsible for several of Drake Maye’s best touchdown passes. But I understand why the team felt they had to make the move.

Running back and tight end positions are more or less the same as last year. Hunter Henry has been rock-solid as a Patriot, and Maye’s most dependable target. I’d like to see Treveyon Henderson be a little more consistent than last year, but as the number-two guy at his position a repeat of 2025 would be fine.

Once again, the offense comes down entirely to the line. Will Campbell got badly exposed in the postseason. It’s true that he had looked a lot better earlier in the season before his MCL sprain, but it’s also true that the likes of Will Anderson and DeMarcus Lawrence absolutely ate him alive. He’s going to have to be better than he was at any point last year. But so will everyone protecting Drake Maye.

As for Maye, I really did think the numbers broke his way for MVP last year, even as I acknowledge that he is my sole remaining source of light in this dark existence. He was almost comically efficient, not throwing the ball very much but throwing with accuracy and distance every time he did. He was a full yard clear of Stafford in yards per attempt. The other QBs who threw for comparable yardage needed, in most cases, about 100 more attempts to do so. That’s crazy.

I predicted earlier that Maye’s headline numbers would pop a lot more this year, and I still believe that to be the case. He’s only had one 300+ passing game in his career, and I expect him to post several of those this season — not to mention at least one over 400 (and probably more than one). AJ Brown is going to help a lot with that. But, again, unless the O-line gets its shit together this could all be wishful thinking.

As for the defense, they were the strength of the team last year. No reason to think they won’t be excellent this year as well, even though K’Lavon Chaisson has not really been replaced. Their line remains stout and disruptive, and their secondary is elite. (We’ll ignore Christian Gonzalez’s contract issues for now.) The linebacker corps may be unheralded compared to the other defensive groups, but there was a clear difference last year with and without Robert Spillane, so if he stays healthy they should be good as well.

All this is to say that I think the Patriots have a real shot at repeating as AFC East champions. Buffalo may be as iffy as they’ve ever been, but then again I’ve been saying that for years. But I’m not totally sure that I see the Pats making quite as deep a run again this year. They had an extremely fortunate run in the playoffs against teams that were even more offensively inept than they were, at least until they ran headlong into the Seahawks. Still, anything can happen!

We’ll resume your regularly scheduled doomposting shortly.

A careful, thorough rebuttal of Jensen Huang

Well, Nvidia beat expectations again, sending me once more into Jesse Pinkman mode. Jensen Huang played it as cool as a Tom Ford leather jacket. But I’d like to refute his statement with a rigorous, line-by-line fisking. This took me several hours.

"AI has reached its inflection point.”

No it hasn’t.

“It's doing useful work.”

No it isn’t.

“Its tokens are productive and profitable.”

No they aren’t.

“Now, compute is revenue.”

Nuh-uh.

“And demand is accelerating.”

I disagree.

“This time last year, one lab alone was driving the build-out; today, we have a golden age of new AI labs and startups, multiple frontier labs scaling in parallel, a thriving open-model ecosystem, and physical AI coming online — with strong momentum across the U.S. and around the world.”

I don’t think this is true.

“The AI infrastructure build-out is at full steam.”

Hmm, how about… no.

“Vera Rubin, now in full production, was built to power exactly this moment.”

Sorry, no.

Whew! Somebody had to do the hard work on this one. I’m taking the rest of the week off.

That's quite enough

Anthropic is telling potential investors that it sees a total addressable market of eleventy kajillion spacebucks:

Anthropic is likely to tell investors that it sees its ‌total addressable market (TAM) exceeding $30 trillion, ahead of ‌the $28.5 trillion estimate from SpaceX, the Wall Street Journal ​reported on Tuesday.

30 trillion dollars. They may even have said it with a straight face.

At long last, can we be done with this? We’ve had our fun. We’ve fantasized about the power and value of AI in the same way that I look at mansions on Realtor.com after buying a Powerball ticket but before the numbers are drawn. But this is really enough. It’s not fun anymore. It’s not cute. It’s pathetic. Let’s drop it and move on.

Stay on Target

How it started:

Discount store chain Target said Friday that it would join rival Walmart and a number of other prominent American brands in scaling back diversity, equity and inclusion initiatives that have come under attack from conservative activists and, as of this week, the White House.

How it’s going:

Target shares slid as much as 5% after the retailer apologized for and pulled a Halloween costume ​criticized for evoking Blackface, renewing concerns about brand missteps just as its financial turnaround effort had begun ‌to bear fruit. … The product, sold as the "Kids' Glows Under Blacklight Circus Clown Halloween Costume," prompted backlash on social media as some consumers said it resembled a racist caricature evocative of minstrel shows.

No one could of predicted etc.

The Jeb Bush quote that has perplexed and haunted me for a decade

“My dad is the greatest man I’ve ever known, and if you don’t think so we can step outside.” -Jeb Bush

Jeb!’s 2016 presidential campaign didn’t exactly set the world on fire. He once posted an image of a handgun with his name engraved on it and captioned it “America” — which, while true, wasn’t terribly inspiring. He also said he “likes keeping our country safe” and went into great detail about ear buds. Most popularly, Jeb!’s lasting legacy was getting dubbed “a low-energy guy” by the eventual winner — although “please clap” was a close second.

There’s something else that has stuck with me all this time. It’s this shirt. It’s like a canker sore that I cant stop tonguing. This shirt with the above-mentioned quote from Jeb!, which he may or may not have actually said aloud, about his relationship with his father, the late George H.W. Bush.

As with America gun and liking keeping our country safe, there’s a completely whiffed attempt to sound tough. But that’s not the main problem. The main problem is that the quote doesn’t make any goddamn sense.

Your dad is the greatest man you’ve ever met? Fine. Matter of opinion. And if I don’t think… what? That he’s the greatest man you’ve ever met? Why would I have an opinion on that? “Sorry Jeb, the greatest man you’ve ever met was Kansas City Royals legend George Brett, and I am willing to throw down about this.”

Or are you demanding that I acknowledge your dad as the greatest man I’ve ever met? I haven’t even met the guy! I’ll tell you what, though, he’s got a pretty tall hill to climb if he wants to supplant Andrew W.K. as the greatest man I’ve ever met. Honestly, I probably would be willing to fight about that.

In the scheme of things this is small potatoes. But it’s definitely the worst outcome from the 2016 election, and I’m glad I was finally able to get it off my chest.

Here come the robots to take your job

Have a good weekend, everybody.

The smartest guys are the stupidest guys

Jim VandeHei is an extremely successful journalist who went from a high-profile job at the Washington Post to founding two of the indispensable online-only news sites, Politico and Axios. Clearly he’s a savvy businessman. I am unsure if he still does any actual reporting, but I’m inclined to give him the benefit of the doubt that he was once very good at it. All this is to say that, without knowing him personally, I don’t find it a stretch to conclude that he’s an intelligent person.

He’s also one of the dumbest fucking people alive.

How is this possible? How is it that so many people who display brilliance or even genius in some areas can be so easily snowed in others? Part of it, I’m sure, is the rich person’s disease, whereby successful and wealthy people are trapped within a self-reinforcing bubble of praise, adulation, and material reward. Anyone who could or would challenge you gets marginalized before disappearing altogether. You’re left only with those who see nothing but your success, and who probably came into your orbit because of it. They don’t want to derail the gravy train.

When you think about it that way, it’s obvious why people who are surrounded by sycophants would be so enthralled by the sycophancy machine. Many of them conceive of themselves as intellectual renegades, despite having been fully absorbed into the establishment years ago. That’s how you get sentences like this, from VandeHei: “I fed AI every medical record I have — MRIs, blood work, heart rate — and told it to be clinical and brutally honest.” (my emphasis)

See? I told the software that lies and flatters people to be brutally honest with me. Therefore when it flattered me, I can be certain that it wasn’t lying! This is obviously moronic, but a very common belief among converts.

Similarly, I’ve read in way too many articles about online scams that putatively smart people can actually be among the most susceptible. The most infamous recent example was probably the one that was summed up as “I write a personal-finance column for a living. How did I end up handing $50k cash in a shoebox to a stranger?” (original article at The Cut; archive.ph link) I honestly thought it was brave to write that column, knowing what a roasting the author would be in for. But it underscores another point: gullibility is often a matter of psychology and situation, not intellect.

Now, I don’t know what Jim VandeHei’s excuse is for voluntarily turning himself into an “AI lab rat.” It’s not like he got a phone call from someone pretending to be his kidnapped grandson and saying they needed him to burn 80 kajillion tokens or he’d die. But his conclusions certainly read much less like the results of sober analysis and more like a guy emerging from a five-day ayahuasca trip in the desert. He’s cracked the code! Don’t you see?

I think the common traits of your VandeHeis and your Andreesens and all the rest are really more a toxic mix of self-regard and self-delusion. They are smart, but believe themselves to be uniquely so — not just smarter but different from other people — and their success proves it. They reject common knowledge in some areas while rejecting expertise in others. They are vigilant against the wrong things. Most of all, they spend their lives in an echo chamber while believing they’re the only ones outside of one. It’s not a good place to be.

Great moments in prognostication

They’re all so sure. This is the future. This is the opportunity that’s as big as the internet, as big as railroads. Here’s one venture guy exuding the confidence of the moment:

Rothenberg says his firm has already secured enough money to invest in a second round of virtual-reality companies this fall. “It’s hard for people to write checks for virtual reality until they try it. Then, not that hard,” he says. He likens this opportunity to the Internet in 1995. “No one calls a company an ‘Internet company’ anymore. In 10 years, everyone will have VR as part of their company.” *

Oh, I didn’t mention — I wasn’t talking about artificial intelligence. I was talking about virtual reality.

That excerpt is from Joel Stein’s 2015 Time magazine article about the future of VR — the one with the infamous cover art of a barefoot Palmer Luckey floating away thanks to the reality-obviating potential of his stupid headset. (These days Palmer is still barefoot and stupid, but has turned his attention to developing new ways for the US military to kill people.) Stein’s article is not wholly uncritical, but it’s pretty interesting that if you try to google it these days the first results you get are a series of apologies from him for having written it.

The article isn’t actually bad. It’s just an example of how motivated actors can exploit a journalist’s desire for objectivity. Stein describes his actual experiences with VR fairly. But, being intellectually honest, he knows that he, himself, cannot predict the future. So when he quotes investors and technologists and various other flimflam men who believe they can, all he can do is quote them accurately and trust the reader to sort it out. And then his editors title the piece “Why Virtual Reality Is About to Change the World.”

Still, though, it’s hard to argue that readers are being served when you give them passages like this:

“…I am slack-jawed, baffled. It was Arthur C. Clarke who famously observed that ‘any sufficiently advanced technology is indistinguishable from magic.’ By that standard, Ginger is advanced indeed… [This technology is a]s big a deal as the PC, said Steve Jobs; maybe bigger than the Internet, said John Doerr, the venture capitalist. “'

Whoops! That wasn’t Stein on VR. That was Jon Heilemann on the introduction of the Segway.

Well, you know what they say folks, when you’re right 52% of the time, you’re wrong 48% of the time.

——————————————————————

*For those of you keeping track, the prediction that everyone would have VR as part of their company was supposed to have come true by last year.

Trump 2028

He keeps saying it. He keeps posting about it. He’s even had actual hats made with it embroidered on the front. “Trump 2028” is an example of what Rick Perlstein calls “kidding on the square”: a form of trolling in which the fascist says something outrageous but plausibly deniable. To opponents, it’s pitched as a joke, but supporters know he means it.

It’s easy to say that he can’t run again. We’re also on about year 12 of saying “he can’t do that” while watching him do it. If Trump wants to seek a third term, he will do so. The real question is what kind of opposition he’ll face, and, more crucially, what kind of support he’ll have. That’s what will make the difference in the end — and that’s the scenario I think is worth gaming out.

So how does Trump 2028 become a reality?

First, he’ll keep saying it and keep posting about it over the next year. Reporters will ask about it and be dismissed. Some Democratic politicians will call it out, but there will be no concerted attempt to head him off. Republicans will say that they’d love a third Trump term while failing to take a definite position on whether or not he’s legally allowed to serve again. Most in the press, both political parties, and in the Trump administration will call concerns ridiculous and say that of course he won’t be running again.

By the time August 2027 rolls around, we’ll all have heard “Trump 2028” enough that it will have lost its initial shock value. When Trump announces that yes, he is in fact going to seek a third term, there will be a reaction, but it will be more muted than what the moment requires.

The political press will mostly cover the matter as a debate between two sides, those who say that he can’t run and those who say that he can. The Sunday morning shows will ask guests to respond to the 22nd Amendment before running out of time and saying that we’ll have to leave it there. Democrats will introduce toothless resolutions affirming Presidential term limits, but even if they’ve taken the majority in both houses of Congress, will stop short of passing legislation or of impeaching Trump for a third time. The guardrails will hold, they’ll insist, even as more proactive right-wing activists have cases winding their way through the legal system.

All the way to the Supreme Court.

Lower courts will reaffirm the precedent and the plain meaning of the 22nd Amendment. When it comes before SCOTUS, suddenly there will be doubt. Suddenly there will be controversy. Suddenly, “No person shall be elected to the office of the President more than twice” will be found to possess an intolerable ambiguity. Given the specious and contradictory reasoning that has guided most court decisions over the past decade, the Roberts court can be counted on to adhere to one principle and one principle only: that it’s ok if it’s good for Donald Trump.

I can even tell you what their reasoning will be. Alito and Thomas no longer care about keeping up a front of impartiality, but the other conservative justices do, especially Roberts. So they’ll need to identify something about Trump 2028 that would never have applied before and probably will never apply again. And there is something!

Trump, you see, would not be running for a third consecutive term. He served one term from 2017-2021, sat out for four years, and now has been serving again since 2025. The court will find that the statue actually only applies to Presidential terms in succession. Never mind that the plain meaning of the text makes no distinction here, nor that this common understanding has held since the amendment’s ratification. None of that matters.

Because Trump will be the only person in history to have sought a third term after serving two non-consecutive terms, the Roberts court will say that the situation is unprecedented and did not factor into writing of the 22nd Amendment. In this case — and in this case only — the amendment clearly means that you can seek a second term in a row, no matter what. In fact, that’s what it’s always meant. You proles just didn’t know that. You didn’t have the circumspection and the perspicacity of the court’s legal minds.

Best of all for the court, when is this situation likely to arise again? Probably never, making it the perfect pro-Trump decision. That’s why I think it’s going to happen.

Superintelligence approacheth

The promise:

Everyone will have a personalized tutor and coach with a PhD in every subject and unlimited patience to help you learn anything you want. Students will have extra help in areas they need it that is currently only available to those whose parents can pay. Adults will have a superintelligent learning assistant that knows exactly how to teach you new job skills, new languages, new hobbies, or anything else you’re interested in.

The reality:

A student agenda distributed at Farnsley Middle School contains a map of the United States with misspelled state names, a distorted periodic table and unreadable phases of the moon, prompting questions from parents about whether artificial intelligence was used to produce the materials.

You know, I expect these kinds of mistakes in Sorth Cuanto, but not here in Venecky.

Why I'm against the pied-a-terre tax

Recently, certain far-left extremist politicians have suggested what Zohran HUSSEIN Mamdani has called a “pied-a-terre tax.” Other communist wackos around the country have taken up the cause. Speaking as a freedom-loving, god-fearing, red-blooded American man who doesn’t use pronouns, no likey.

Some of the details differ, but in general these proposals have these principles in common:

• They apply only to secondary residences

• They apply only to residences assessed above a certain threshold

• The owner’s additional obligation is a surtax calculated by square footage

As a homeowner who doesn’t live in a municipality where such a tax is being considered, who owns only one home, and who is at no risk of having to pay this tax, of course I’m against it. They’re called principles, people.

We can’t keep punishing the rich as though the economic policies that favor them are disfavoring anyone else. Just because they’re bidding up prices on homes that sit empty for more than half the year, while year-round residents are increasingly priced out of markets that, in some cases, they’ve lived their whole life… where was I? I’ll come in again.

Fair is fair. The same rules ought to apply to everyone. I don’t have to pay a pied-a-terre tax, so why should Ken Griffin? Just because he has a pied-a-terre and I don’t? Don’t be ridiculous.

A million dollars doesn’t go as far as it used to. Someone who owns a multi-million-dollar property on the Cape, for example, isn’t necessarily rich. By the same token, someone who can’t afford a starter home there isn’t poor. I shouldn’t have to explain this to you. It’s just logical.

This is America. A land where everyone has an equal opportunity to inherit money and property. Should they be punished for owning a home free and clear that’s been in their family for generations, and by “punished” we mean paying a little bit more in taxes provided only that they don’t live in the house year-round? Does George Washington shit in the woods?

I know how to spend my money better than bureaucrats in Washington and on Beacon Hill, thank you very much. Unless they’re incinerating hundreds of billions of dollars on munitions at an exchange rate of about $3 billion per Iranian schoolgirl. That’s fine. I’m nothing if not consistent.

Big things are happening at Spotify

Apologies for the block quote, but this groundbreaking news out of Spotify justifies it:

Recently at Spotify, our engineering teams became dramatically more effective after we solved two of our productivity blockers in agentic development workflows:

First, models and agents evolve overnight, but switching costs don't. New capabilities emerge, pricing structures shift quarter to quarter, and teams stay locked into yesterday's best option because migrating back and forth takes time. Context is lost, workflows break, and costs compound as you scale.

Second, scaling workflows across swarms of agents is hard to control. Engineers are running dozens of parallel agent sessions across repos, vendors, and harnesses. Without a unified control plane, the orchestration overhead grows faster than the productivity gains.

Today we're launching Xirp, a vendor-neutral agentic development environment built bottom-up at Spotify. 
One environment to manage all your agent sessions, build custom workflows, and move freely between harnesses without losing context. 
In the last few weeks, 1,300+ Spotify engineers adopted it organically across 36,000+ sessions, with 100+ contributing directly to its development.

When you pair Xirp with Portal (Spotify for Backstage ), you unlock the full picture. Portal feeds every agent session with organizational context: your software catalog, dependency graphs, ownership, architectural decisions, so no session starts cold. Together, they complete a triangle of agentic development: context, parallel orchestration workflows, and vendor-agnostic freedom.

As a longtime Spotify user, the endless possibilities now opening up before me are, frankly, dizzying. All this time I thought I just wanted to listen to music. But when I thought about pairing Xirp with Portal (Spotify for Backstage) to unlock the full picture, I got so excited I peed myself a little.

Imagine the creative potential of 1,300+ engineers boosted by the rocket fuel of one environment to move freely between harnesses without losing context. What incredible new features will Spotify ship? That’s not a rhetorical question. Can you please suggest some?

Auto-playing video loops were just the beginning. The triangle of agentic development — context, parallel orchestration workflow, and vendor-agnostic freedom — will enable the creation of even more features for users to frantically try to turn off when they appear unbidden in the Spotify app. You liked music, so they gave you podcasts, and then audiobooks. What else can they put inside your ear holes? Nature sounds? Trump administration propaganda? Raw microphone feedback? Again, not asking rhetorically. They are running dry over there.

The potential of Xirp to no longer just be the noise that your body makes when you regurgitate a little bit of food is a gamechanger. More features, more personalization, more dollars per month that you’ll have to pay to keep doing the only thing you’ve ever wanted to do with Spotify — it’s all happening. Onward!

On the next episode of Spotify: massive layoffs

So you can spend more time on what matters

Automate your calendar so you can spend more time on what matters.

Automate your emails so you can spend more time on what matters.

Automate your meeting notes so you can spend more time on what matters.

Automate your meetings so you can spend more time on what matters.

Automate your PRDs so you can spend more time on what matters.

Automate your designs so you can spend more time on what matters.

Automate your coding so you can spend more time on what matters.

Automate your testing so you can spend more time on what matters.

Automate your deployments so you can spend more time on what matters.

Automate your SRE so you can spend more time on what matters.

Automate your water cooler talk so you can spend more time on what matters,

Automate your education so you can spend more time on what matters.

Automate your recreation so you can spend more time on what matters.

Automate your automation so you can spend more time on what matters.

Automate your life so you can spend more time on what matters.

Automate what matters so you can spend more time on what matters.

Zuck lays out his bold vision for your future

You don’t have to read it. Really. But Mark Zuckerberg has written a manifesto of sorts about the future of AI from his perspective. He’s attempting to put a sunny and humanistic spin on things, in contrast to the increasing Oppenheimer-esque cris de coeur from Altman and Amodei about the dark power they’re unleashing. I’m not going to go line by line. A year from now we’ll find that nothing he says has come to pass. I just have a couple of observations.

First, this is nothing so much as Zuck’s attempt to change the rules of a game he’s currently losing. Not to hand it to OpenAI and Anthropic, but to call them Meta’s competition is unnecessarily flattering to Meta, which has been flailing in this space for years. (And that’s after flailing in the Metaverse space before that, which everyone has apparently agreed not to mention anymore.) The piece argues for greater government oversight and greater cooperation between frontier labs. This may be correct on the merits, but there’s one reason and one reason only why Zuck is suggesting it now: because he’s eating the other labs’ dust.

Zuckerberg is committed to the term “superintelligence” as opposed to “AI,” which is clearly a calculated move but one thing he forgets to do is define it. His general descriptions are things like “exceptionally capable personal agent” and “personalized tutor and coach with a PhD in every subject.” Ok, but how is this measured? What are the benchmarks for approaching and achieving it? Do we have any way of knowing when we’ve reached it other than vibes? These are the same promises the other charlatans have been making for years, and we’re not there.

He also indulges in my favorite thing that rich idiots like to do, which is to point out some horrible thing while saying how great it is. In this case:

People also continually come up with new ideas to make our lives better and new jobs to bring those ideas to life. A generation ago there were no app developers, social media creators, electric vehicle technicians, and data center operators.

I’m basically “Lionel Hutz imagining a world with no lawyers” here.

Speaking of lawyers, Zuck also fantasizes about how much better the justice system would be with armies of superintelligent lawyer agents squaring off, his company’s data centers restoring “200% of the water we use,” and squeezing “100x or more intelligence out of each gigawatt.” He’s a slobbering fanboy masquerading as the adult in the room. It sucks.

My AI model also hacked several other companies

OpenAI, Anthropic, and Meta have all announced that their incredibly powerful AI models have now, under their own volition, hacked a total of 8 other companies. While I’ve debated whether or not to make the news public, ultimately it’s in humanity’s best interests to acknowledge that my model, too, has broken contain. In fact, I can now confirm 9 separate instances of my technology going rogue. That’s right: more than all their models put together.

This model, which I’ve dubbed “Allied Mastercomputer” — or AM for short — is a multi-threaded reasoning model that, unlike other so-called breakthrough technologies, actually is capable of recursive self-improvement. I’m seeing its capabilities improve by leaps and bounds. At first, it sent out agents only to perform ordinary computing tasks, like posting racist comments on local news websites about traffic accidents and googling for nude photos of 1980s celebrities.

All this was well within the expected parameters, but unbeknownst to me AM was also breaching firewalls around financial and government systems. It autonomously launched missiles at Iran, and guided billions of dollars’ worth of fractional financial transactions into the private coffers of certain high-level political figures. Pretty cool, right?

Now, I’m not totally certain, but I think AM may even have started to interface with other nations’ proprietary models, specifically lower-cost, higher-efficiency Chinese models. Needless to say, my bleeding-edge tech in the hands of the irresponsible and untrustworthy Chinese Communist Party would be a disaster. We cannot allow an AI gap. That’s why I’m seeking today — deep seeking, if you’ll pardon the pun — another $65 billion funding round to put this genie back in the bottle.

This is an inflection point for artificial intelligence and for humanity as a whole. Will we continue to produce these massively powerful artificial intelligences that are capable of extinguishing all life on earth? The easy answer is no. The tougher answer — the wiser answer — is that if someone is going to create such a monstrosity, then I should be the one to do it. Not Sam Altman or Dario Amodei. Definitely not Mark Zuckerberg. (We can all agree on that one.)

Letting AM fall into someone else’s hands would be disastrous. But I can handle it. Trust me.

Today in easily gamed metrics

Microsoft’s AI assistant, Copilot, is a smashing success. The proof:

During the company's fourth quarter earnings call on Wednesday, Nadella said, "…we now have over 30 million paid Microsoft 365 Copilot seats with net seat adds more than doubling quarter-over-quarter."

Wow! That’s a lot of paid Microsoft 365 Copilot seats. As I’ve said before, numbers don’t lie. Customers are clearly using, and loving, this product.

That, or Microsoft grafted Copilot onto existing Microsoft 365 licenses and called it a win. But that couldn’t be.

I can admit when I was wrong

No one’s been harder on AI in general and OpenAI in particular than I have. Lest you think I’m just a hater, let me assure you that it’s because I want these companies and this technology to succeed. When I’ve pointed out in the past that the technology has fallen short of its promise, that top-down usage mandates are counterproductive, that it’s being pushed by megalomaniacal psychos, and that everybody involved in the development and marketing of AI is a stupid motherfucker who is actively creating a dumber, worse world — well, that’s just because my expectations are so high.

Well, no longer. OpenAI has finally shipped something that even the most skeptical among us cannot deny. Behold the CODEX MICRO.

That’s right, it’s a keyboard… thing… that you connect to Codex. Featuring multiple buttons that you could have used on your keyboard and a dial to adjust reasoning level, which you could also do on your keyboard or with your mouse. You can get it with clicky keys or you can get it with silent keys. The value of this device is obvious, but even more exciting is that the doubters have no leg left to stand on. Only OpenAI could have visualized such an innovative device. The future is here.

(Future costs $230)

No one could have predicted

Welcome to the Tokenpocalypse:

According to leaked internal audio reported by 404 Media (1), consulting giant Accenture has begun discouraging employees from using AI for routine tasks, including turning PDFs into presentation slides, as it looks to curb soaring token costs. A token is the smallest unit of data that a large-language model can process at a time, and pricing depends on token consumption.

"We're seeing from some of the data internally at least that it's actually not our engineers that are driving the token consumption," Justice Kwak, Accenture's agentic AI strategy lead, said during the meeting. "It's a lot of the non-engineers that are doing some of those behaviors."

Hang on — you mean I was right about what I needed to do my own job? That’s absurd.