AI growth and tokenmaxxing

It is difficult to convey the abundance of warning signs suggesting that the AI boom of the mid-twenties is in fact an AI bubble. Note, I'm not saying that AI won't eventually reshape our world (I believe it will). I'm not even saying that large language models won't someday live up to their promise (though I am far more skeptical on that point). What I'm saying is that there is increasingly little chance that the industry will go from a money sink to a money gusher in the next three or four years, which means that the chances of our avoiding a larger and more painful version of the dot-com bubble are growing vanishingly slim.

Pretty much every part of the booster narrative looks, at best, sketchy under close scrutiny. Take perhaps the main pillar of the bull case: token and revenue growth. The following, the beginning of a massive recent tweet from a prominent investor, lays out the standard view.

 

(I love how he tries to frame his argument as a conservative case by suggesting a slight tapering off of his impossible-to-sustain growth curve.)

The idea that you can sustain this kind of exponential upward path is silly, but the problems with his numbers actually start before he begins to extrapolate. Where exactly has this growth been coming from?

Given the unprecedented circularity of current AI financing and business deals, it is always wise to consider the possibility that the numbers you're seeing are the result of Peggy Bundy accounting, but let's drill down further and see how these tokens are actually being burned through.

 

 Joe Wilkins writing for Futurism:

Many employers, in turn, have begun mandating AI use on the job, some even going so far as to fire those who don’t hop on board in order to justify their big-time spending on tech industry contracts. While the forced adoption of AI has major implications for the financial viability of the tech overall, it’s also giving office workers a perverse incentive to increase their AI use for non-productive tasks.

Case in point, the Financial Times reports, Amazon’s office staffers are increasingly using the company’s in-house AI agent MeshClaw to run personal tasks in a bid to get their quotas up.

In an attempt to get more than 80 percent of its developers to use AI every week, Amazon has introduced employee-specific AI usage targets in addition to a broader “token consumption” leaderboard that tracks how much each employee uses AI (in machine learning parlance, tokens refer to basic units of data used by AI models to understand text.)

But according to staffers interviewed by the FT, employees are gaming the system by increasingly using the mandated AI systems to automate personal tasks, a tactic known as “tokenmaxxing.”

...

For its part, Amazon told the FT that “thousands of Amazonians to automate repetitive tasks each day,” adding that the retailer is “committed to the safe, secure and responsible development and deployment of generative AI for our customers.”

A quick jaunt through Team Blind, a message board for verified employees of companies like Google and Apple, shows that the practice is widespread — or at least, widely acknowledged.

“I burn tokens to s**t my [project manager],” one Amazon employee wrote in a post from May 8. “Whenever my PM says stupid s**t, I launch 10 sub agents to s**t him. Great use of GPUs.”

Asked by a Microsoft worker what that means, the Amazon staffer replied “just paste the Slack conversation history and tell the agent to analyze the guy using 10 sub agents.”

Though less malicious, there are plenty of other threads from workers across the tech industry asking how others are maximizing token usage — receiving lots of free advice in the replies.

 This isn't just Amazon.

 Deirdre Bosa and Jasmine Wu writing for CNBC.

Meta and Shopify say they have created internal leaderboards that track how many tokens employees use. Nvidia CEO Jensen Huang has said he’d be “deeply alarmed” if an engineer earning $500,000 a year wasn’t using at least $250,000 worth of compute — measuring what an engineer spends on AI instead of what they produce with it.

Once companies start measuring AI adoption by volume, employees optimize for the metric instead of the outcome.

“If your goal is to just burn a lot of money, there are easy ways to do that,” said Ali Ghodsi, CEO of Databricks, which processes AI workloads for thousands of enterprises. “Resubmit the query to ten places. Put up a loop that just does it again and again. It’s going to cost a lot of money and not lead to anything.”

 

Only slightly more absurd than some real-life cases

A classic from XKCD

A great Charles Bronson story.

This has nothing to do with any of the topics we focus on here at West Coast Stat Views. I have no good excuse for posting it other than, as previously mentioned, it's a great Charles Bronson story, and I've always been a big fan of Bronson, particularly before the Cannon Films years.

As long as he was actually given something to do, pretty much all of his film and television work before the 70s was wonderful, wry, economical, and with real sensitivity. His performances on the Twilight Zone and Have Gun Will Travel were series highlights. He was often the best thing in those big, manly ensemble pieces, the Magnificent Seven, the Great Escape, and the Dirty Dozen (his closing comment was perhaps the best line in the last one). He more than held his own with Henry Fonda and Jason Robards in Once Upon a Time in the West.

He also turned in first-rate performances in lesser but still worth-a-look movies like Master of the World, the Mechanic (complete with 70s nihilism and a surprisingly obvious homoerotic subtext), and Red Sun (forget the critics, if the idea of a buddy action comedy with Bronson and Toshirô Mifune sounds like fun, you'll enjoy this movie).

Among others who shared my high opinion of the actor. Ingmar Bergman was on record as admiring his work and calling him "scandalously underestimated."

Charles Bronson's childhood was something out of Dickens which explains this wonderful story from Kurt Russell.

 

A tale told in tweets (Make sure to use the Muppet announcer voice when you read the word "space")

 

Are data centers in space a dream or AI’s next big thing? SpaceX, Blue Origin and tech firms see orbital server farms addressing Earth’s power and land issues but costs could be out of this world.

[image or embed]

— The Wall Street Journal (@wsj.com) May 12, 2026 at 9:17 AM

PICHAI: “.. There’s no doubt to me that a decade or so away, we’ll be viewing it as a more normal way to build data centers.” @wsj.com $GOOGL www.wsj.com/tech/spacex-...

[image or embed]

— Carl Quintanilla (@carlquintanilla.bsky.social) May 12, 2026 at 9:59 AM

I wonder if there's some reason why Google would want to encourage all of this silly data-centers-in-space talk...

[image or embed]

— markpalko.bsky.social (@markpalko.bsky.social) May 13, 2026 at 12:47 AM

finance.yahoo.com/news/why-spa...

[image or embed]

— markpalko.bsky.social (@markpalko.bsky.social) May 13, 2026 at 12:51 AM

Truly incredible stuff here. These people are brain dead www.wsj.com/tech/ai/data...

[image or embed]

— Rude Law Dog (@esghound.com) May 12, 2026 at 9:27 AM

BAGNAROK is Ed Niedermeyer's term for Musk's scheme to use the SpaceX IPO to create a massive short squeeze and force everyone with a NASDAQ index fund to buy his insanely inflated stock.

Keeping the data-centers-in-space hype following may be a necessary condition for BAGNAROK (tm) just as credulous reporting on humanoids is a necessary condition for keeping Tesla valuation from imploding. (And Google makes a lot of money if this goes through.)

[image or embed]

— markpalko.bsky.social (@markpalko.bsky.social) May 13, 2026 at 1:13 AM

Google/Alphabet has even more money money on the line when it comes to keeping the AI bubble going and this definitely helps keep that river of hype flowing.

— markpalko.bsky.social (@markpalko.bsky.social) May 13, 2026 at 2:11 AM

Data centers in SPAAAAAAAACE! youtu.be/EmI77ZBeJrQ?...

[image or embed]

— markpalko.bsky.social (@markpalko.bsky.social) May 13, 2026 at 2:11 AM

It's amazing the mileage I get out of some posts

Weakliem and Gelman are talking about Trump's 2024 nomination, especially the role of the elites. I have some issues with the arguments, not with the analyses (both these guys know what they're doing), but with the questions being asked. 

I might follow up with a full post on the subject, but for now, I thought it might be helpful to look back at how the establishment, both right-wing (Fox) and mainstream (NYT et al.) were approaching the possibility of another Trump candidacy almost two years before the convention.  

 

Wednesday, August 24, 2022

I shouldn't have to say this but a 49-25 poll is not good news for the 25 (and it gets worse)

First off, the decision of the New York Times to even conduct a presidential poll more than two years before the election is irresponsible and bad for for Democracy. It distracts from important conversations and, since the data are largely worthless,  its main function is to introduce noise into the conventional wisdom. 

 But while the data are not worth wasting any time analyzing, the analysis in the NYT piece by Michael C. Bender is worth talking about, and I don't mean that in a good way. This represents a disturbing throwback to the wishful analytics of the second half of 2015, showing that many data journalists and the publications that employ them have learned nothing in the past seven years.

Back in the early (and not so early) days of the last Republican primary, 538, the Upshot, and pretty much everyone else in the business were competing to see who could come up with the best argument for why being consistently ahead in the polls was actually bad news for Trump. These arguments, as we pointed out at the time, were laughably bad.

Just as being ahead in the polls was not bad for Trump in 2015, the results of this poll (to the extent that they have any meaning) are not bad for Trump in 2022. When elections approach, parties tend to converge on whoever has the clear plurality, and 49% is a big plurality, particularly when a large part of it consists of people who are personally loyal to Trump rather than to the GOP. On top of that, 53% of self-identified Republicans had a "very favorable" opinion of the former president and 27% were "somewhat favorable."

80% favorable is a good number.

Politically, this is a time of tumult, and all predictions at this point are little more than educated guesses, but given the losses and scandals Trump had seen by the time this poll was taken, his support was remarkably solid, which is the opposite of how Bender spun it.

And it gets worse

Here's the headline and the beginning of Bender's piece. [emphasis added.]

Half of G.O.P. Voters Ready to Leave Trump Behind, Poll Finds

Far from consolidating his support, the former president appears weakened in his party, especially with younger and college-educated Republicans. Gov. Ron DeSantis of Florida is the most popular alternative.

By focusing on political payback inside his party instead of tending to wounds opened by his alarming attempts to cling to power after his 2020 defeat, Mr. Trump appears to have only deepened fault lines among Republicans during his yearlong revenge tour. A clear majority of primary voters under 35 years old, 64 percent, as well as 65 percent of those with at least a college degree — a leading indicator of political preferences inside the donor class — told pollsters they would vote against Mr. Trump in a presidential primary.

Notice the phrase "GOP voters." That 49% refers to the respondents who said they thought they would vote in the Republican primary. Among that group, those who identified as Republicans went for Trump over DeSantis 56% to 21%.

If we're talking about who is likely to be nominated (which is, as mentioned before, an incredibly stupid and irresponsible question to be asking more than a year before the election), people who say they are going to vote in the primary are a reasonable group to focus on, but they cannot be used interchangeably with Republicans, which is exactly what Bender does.

While we're on the subject, this was a survey of 849 registered voters, so when we limit ourselves to those who said they were going to vote in the Republican primary then start slicing and dicing that, we are building big conclusions on a foundation of very small numbers.

And it gets worse. [Emphasis added]

While about one-fourth of Republicans said they didn’t know enough to have an opinion about Mr. DeSantis, he was well-liked by those who did. Among those who voted for Mr. Trump in 2020, 44 percent said they had a very favorable opinion of Mr. DeSantis — similar to the 46 percent who said the same about Mr. Trump.

Should Mr. DeSantis and Mr. Trump face off in a primary, the poll suggested that support from Fox News could prove crucial: Mr. Trump held a 62 percent to 26 percent advantage over Mr. DeSantis among Fox News viewers, while the gap between the two Floridians was 16 points closer among Republicans who mainly receive their news from another source.

Here's a fun bit of context. Fox has been maxing out its support of DeSantis for years now.

Steve Contorno writing for the Tampa Bay Times

(from August of 2021):

The details of this staged news event were captured in four months of emails between Fox and DeSantis’ office, obtained by the Tampa Bay Times through a records request. The correspondences, which totaled 1,250 pages, lay bare how DeSantis has wielded the country’s largest conservative megaphone and show a striking effort by Fox to inflate the Republican’s profile.

From the week of the 2020 election through February [2021], the network asked DeSantis to appear on its airwaves 113 times, or nearly once a day. Sometimes, the requests came in bunches — four, five, even six emails in a matter of hours from producers who punctuated their overtures with flattery. (“The governor spoke wonderfully at CPAC,” one producer wrote in March.)

There are few surprises when DeSantis goes live with Fox. “Exclusive” events like Jan. 22 are carefully crafted with guidance from DeSantis’ team. Topics, talking points and even graphics are shared in advance.

Once, a Fox producer offered to let DeSantis pick the subject matter if he agreed to come on.

If I were DeSantis's campaign manager, this poll would scare the shit out of me. Fox has pushed him to a degree unprecedented for a politician at that stage of his career. He has also gotten tremendous (and appallingly credulous) coverage from the mainstream press, but he just doesn't register. I know political scientists and data journalists don't like to talk about things like personality, let alone charisma, but for whatever reason, DeSantis has not made much of an impression.

It's possible cataclysmic events (of which we're seeing a definite uptick) will hand the Florida governor the nomination or maybe even the presidency, but if this poll had any meaning, it would be bad new for him and good news for Trump.

And it gets worse.

This wasn't just an article based on worthless data sliced ridiculously thin wishfully analyzed to get conclusions completely at odds with the actual numbers; this was an influential and widely cited article based on worthless data sliced ridiculously thin wishfully analyzed to get conclusions completely at odds with the actual numbers. It instantly became a fan favorite among political journalists.

The article was published on July 12th and immediately became part of the conventional wisdom. A little less than a month later, the FBI raided Mar-a-Lago, and the "Republicans are moving on from Trump" voices suddenly grew quieter, as even the highest ranking party members responded with unhinged accusations and threats of retribution. Though the pundits desperately wanted to believe otherwise, they  had to acknowledge that the GOP still belongs to Donald Trump.

 

We laughed at her at the time, but in the age of AI bubble circular financing, Peg Bundy feels like a true visionary.

From Ed Zitron:

Let me spell it out in a way that boosters can understand, in the style of Gillam Fitness: Anthropic not have money to pay big cloud bills, because Anthropic company cost lots of money, more money than Anthropic make! So Anthropic only PAY cloud bills if OTHERS give it money! Amazon GIVE MONEY to Anthropic to GIVE BACK TO AMAZON, which mean no profit! And Amazon not give Anthropic enough money to pay it, so Anthropic have to ask OTHERS for money! That BAD! It mean BUSINESS not STABLE, and CLIENT not STABLE. 

This bad when client MOST OF AI MONEY!

This ALSO mean that Anthropic RELIANT on OTHERS to pay AMAZON, which make AMAZON dependent on VENTURE CAPITAL for FUTURE REVENUE! Amazon SAY it have BIG BUSINESS, but BIG BUSINESS dependent on ANTHROPIC, which mean BIG BUSINESS dependent on VENTURE CAPITAL!

This SAME for GOOGLE! Both say they have BIG CLIENT, but BIG CLIENT MONEY not supported by REVENUE, so BIG CLIENT actually mean “HOW MUCH VENTURE CAPITAL MONEY ANTHROPIC HAVE.” 

This bad business! 

Sidenote: Me know you say “ANTHROPIC STOCK WORTH BIG MONEY,” but me need you remember how much capex Amazon and Google spend! Even if Anthropic stake worth $200 Billion, Amazon and Google still spend MANY more dollar than that on capex! And stake so BIG that neither able to SELL ALL. Only make gain on PAPER, which not REAL MONEY!

As always, most important life lessons can be learned from late night cable.

ChatGPT still has trouble trouble with rise-over-run

As mentioned before, I regularly use this tool to proofreading and to assist with coding and I find it very helpful, but I'm constantly being reminded there's no real understanding behind its answers.

In this case, to illustrate a point in a conversation on Teams, I asked the chatbot to generate a graph showing the slope in a discontinuity analysis going from 0.6 to 0.3. It produced an appropriate image, but it wasn't quite right.

 

 

Remembering the world before the hyperloop -- a continuing series

From CNBC 2020/09/02

Opinion - Our New Future

Op-ed: The hyperloop will revolutionize transportation in the post-coronavirus world

Published Wed, Sep 2 20209:28 AM EDTUpdated Wed, Sep 2 20209:53 AM EDT

Even though hyperloop capsules can reach speeds of 760 miles per hour, on a practical level, this transportation sector appears to be stalled.

 

...

 

The challenges facing the transportation sector are significant and multi-faceted. But past the pitfalls of this antiquated system of steel railroads and cumbersome commercial flights lies a faster future. When we arrive, we’ll enjoy new travel experiences defined by safety, speed, environmentalism and improved comfort. To get there, you’ll need a fifth mode of transportation: Something low density, yet high volume. Something that is twice the speed of a plane, yet safer than any current public transportation; you’ll need something that is sustainable, reliable, and immune to weather variations. In other words, you’ll need a hyperloop.

 

Coined by Elon Musk in 2013, hyperloop is a bold engineering initiative to send elevated passenger pods through tubes using magnetic fields, allowing people to travel even faster than modern airplanes from city-to-city.  [Regular readers might want to skip this parenthetical—God knows you've heard it before—but the "hyperloop" that Elon Musk proposed in 2013 was a completely different technology than what companies like Virgin were trying to build in 2020. As profoundly silly and prohibitively expensive as those projects unquestionably were, they were nonetheless models of practicality and sound engineering compared to Musk's original idea for a near-supersonic train that ran on an air cushion, in his own words, like an air hockey table. It was a piece of engineering so bad, even by the debased standards of 21st-century, drug-fueled techbro babbling, that everyone in the field pretended it had never happened. -- MP]  Although the concept sounds futuristic, we are actually very close to realizing it. From Musk to Richard Branson, the world’s greatest minds are building out testing sites across the United States to conduct high velocity experiments. And both state and federal governments are recognizing the value this technology has to their communities.

Just last month, the U.S. Department of Transportation issued a direction to establish regulations for hyperloop technology, with Secretary Elaine Chao saying, “Inventors, investors, and stakeholders are ready to build out these technologies.”  

 

The Transportation Department has since categorized hyperloop under the Federal Railroad Administration, making it eligible for government infrastructure funds. Virgin Hyperloop just received sign off from the Indian government to build out one of its first projects in the country’s western provinces. As an early investor in Virgin Hyperloop, the value this technology has for the transportation sector, company shareholders, local communities, and major metropolitan regions are obvious. After all initial fixed cost investments, hyperloop projects are inexpensive to operate and have a high degree of operating leverage.

 

 ...

 

Many skeptics will point to partisan gridlock as one major hurdle the nascent industry faces, using the government showdowns over Amtrak’s operating budget as a case study. But once legislators recognize all of hyperloop’s benefits, these issues will be nonexistent. Republicans like Elaine Chao already love hyperloop because it is creating new economic opportunity zones, while Democrats are embracing its environmental component. Texas Democratic Congressman Joaquin Castro famously once said, “Texas should be building hyperloop.”  

 

And in case you hadn't guessed...

 

Disclosure: Mr. Asher is an investor in Virgin Hyperloop.

The Organization Mad -- Parody as historical snapshot.

I'd argue that looking at what people chose to mock often gives us a better take on contemporary attitudes than you get from standard sources. There's a tendency to look upon previous generations as being a little bit simple-minded. All too often, we assume that they unquestioningly believed what they were told and were oblivious to what seems obvious in retrospect. For this reason, subtle satire is often missed by current readers. Fortunately, excessive subtlety is not, and has never been, a problem with Mad magazine.

The second half of the '50s was a transition for Mad magazine. Its early years under Harvey Kurtzman had been distinctive and often brilliant but were largely a reflection of the editor's personality. When Al Feldstein took over, the magazine struggled for a while to find its identity before settling on a basic formula in the early '60s that would remain more or less constant for 40 or so years.

This transition period was arguably the least impressive; it certainly has inspired the least love from fans. But if you're looking for insights into the mood of the times, this might be the richest vein.

The title of this 1956 book (available on Internet Archive) was a play on the just-published The Organization Man, a highly influential critique of a perceived postwar loss of rugged individualism. This was a fertile field for satirists of the period such as Mort Sahl, Bob Newhart, and Stan Freberg.

The book is surprisingly text-heavy, including some pieces with almost no pictures, such as this one.

 

 

The writers and publications parodied were Westbrook Pegler, Louella Parsons, the Daily News, the Daily Worker, Hearst's Journal American, Time, and the New York Times. Of these, the most interesting from a historical point of view is Pegler.

Pegler may be forgotten now, but for those trying to understand the rise of the far right, he's worth taking a look at.

Francis James Westbrook Pegler (August 2, 1894 – June 24, 1969) was a Pulitzer Prize-winning American journalist described as "one of the godfathers of right-wing populism". He was a newspaper columnist popular in the 1930s and 1940s for his opposition to the New Deal, labor unions, and anti-lynching legislation. [Worth noting that Pegler was from  Minneapolis, in case you thought these positions were strictly a Southern thing. -- MP]

As an ardent proponent of states' rights, Pegler criticized a variety of targets whom he saw as extending the reach of the federal government, including Herbert Hoover, FDR ("moosejaw"), Harry Truman ("a thin-lipped hater"), and John F. Kennedy. He also criticized the Supreme Court, the tax system, labor unions, and any federal intervention on the issue of civil rights. In 1962, he lost his contract with King Features Syndicate, owned by the Hearst Corporation, after he started criticizing Hearst executives. His late writing appeared sporadically in publications that included the John Birch Society's American Opinion.

 ...

In the 1950s and 1960s, as Pegler's conservative views became more extreme and his writing increasingly shrill, he earned the tag of "the stuck whistle of journalism."  Despite having earlier called for the desegregation of baseball, Pegler denounced the civil rights movement and in the early 1960s wrote for the John Birch Society. He aligned himself with the white supremacist White Citizens Council.  He was ultimately expelled from the John Birch Society because of his extreme views. However, the Society did put his picture on the cover of its magazine, American Opinion, when he died. 

Mad's previous editor, Harvey Kurtzman, had gotten his start as a cartoonist for the Daily Worker, so this might be Feldstein's dig at a departed rival.  

I've long been fascinated by Hearst, so I'm the target audience for this one. 

Meta, Microsoft, and other tech giants are losing billions on AI. Can Apple catch up?

Steven Levy of Wired has been at this for a long time. In many ways, he might be the ultimate insider tech journalist. That means that even when his takes are bad, and this one certainly is, they're still interesting because they give us some insight into the way people in the tech industry are thinking.

Apr 24, 2026 11:00 AM 

Apple's Next CEO Needs to Launch a Killer AI Product

 

Sometime in the next year or two, Apple’s new CEO, John Ternus, will step onto a stage and tell the world that his company has a revolutionary product. This product, he’ll say, will put the full and awesome power of AI into everyone’s hands. It probably won’t represent a breakthrough in AI research, and it might not let people automate work or perform tasks any better than a lot of technically minded people are doing today. It may or may not involve a new device, though if it doesn’t, one should be in development. But if it all works out, that keynote will mark the moment when Apple did to AI what it has done for desktop computers, the internet, mobile technology, wearables, and music distribution. That is, it’ll offer a solution to a troublesome technology that’s so delightful and right that it seems obvious in retrospect.

This isn’t optional for Ternus. While AI is clearly the future and millions of people use it, even more are suspicious of it. Powerful new AI agent technologies such as Claude Code and OpenClaw are still too risky or technical for most people to adopt. If Apple doesn’t decode this for the masses, someone else will. Current CEO Tim Cook, who announced this week that he’ll vacate his role in September and become the company board’s executive chairman, has done a superlative job guiding the company after Steve Jobs, but he left this important box unchecked. Apple Intelligence, rolled out with much fanfare in 2024, was underwhelming and uncompleted.

The notion that companies must rush to embrace every technological next big thing or be consigned to the dustbin of history is an article of faith among techno-optimists. It hasn't been true in the past—there are countless exceptions—but the belief remains as strong as ever.

To get a full sense of the disconnect from reality here, consider what Levy is able to come up with in the way of actual AI products that Apple might offer its customers.

 That’s fine, but I look back to the mid-2000s when everybody was waiting for Apple to come out with a phone. When Jobs finally delivered in January 2007, the product defined the mobile era. It’s a big ask for Ternus to do something similar for the AI age—but it’s an opportunity that must be seized. AI threatens to disrupt the entire iPhone ecosystem. By the end of this decade, it’s unlikely that people will swipe on their phones to tap on Uber or Lyft. They will just tell their always-on AI agent to get them home. Or that agent will have already figured out where they need to go, and the car will be waiting without the friction of a request. “There’s an app for that,” may be replaced by “Let the agent do that.”

I am hard-pressed to think of something less in need of agentic automation than booking a rideshare. It's a process of three to five clicks, taking less than a minute in most cases. What's worse, the only way that an agent might streamline the process is by removing steps that most people would very much want to maintain control over, like reviewing the driver or selecting the type of car.

(Admittedly, there is a long Apple history of removing functionality in the name of innovation.)

Of course, there are real and important uses for large language model-based technology, but in the case of Apple, this seems to be someone trying and failing to come up with a problem for this particular solution. That's the story of far too much tech journalism in the 21st century.

Sixteen years ago at the blog: I'm not wasting time reading old paperbacks; I'm conducting serious research into Cold War political attitudes

  

Friday, May 7, 2010

Things I've learned -- 60s pulp edition

One of the pleasures of popular art is the glimpse it provides into the attitudes of the times. Sometimes this pleasure comes from the naivete of earlier times. More often, though, the experience is just the opposite -- you discover that the attitudes of previous generations were complex, surprisingly modern and often at odds with the conventional view of the era.

I recently read a couple of pulp novels from the Sixties that fell into the second category. One was The Thief Who Couldn't Sleep by Lawrence Block (discussed here). The other was The Ambushers, the sixth book in the Matt Helm series by Donald Hamilton.

Matt Helm was a counter espionage agent specializing in wet work, closer to a John LeCarre hero than to James Bond. The books were well reviewed (Anthony Boucher, writing, I believe, in the New York Times, said "Donald Hamilton has brought to the spy novel the authentic hard realism of Dashiell Hammett; and his stories are as compelling, and probably as close to the sordid truth of espionage, as any now being told."), they were extremely popular and they were sturdy enough to survive a god-awful series of in-name-only adaptations starring Dean Martin.

If you picked up a copy of the Ambushers in 1963 and turned to the first page, you'd find the narrator slipping into a Latin American country carrying a rifle with a high powered scope. The rebel leader he has been sent to kill may have had it coming, but it's not obvious that target was any worse than the leader the U.S. supported (later in the story, Helm is glad to hear that the latter has also been assassinated though his superiors most certainly are not).

The Russians in the Ambushers are sometimes enemies and sometimes allies, depending on the circumstances. The American agents generally hold the high moral ground but it's a distinction that can be rather thin and the protagonist has learned not to dwell on it too much. Everyone's hands are dirty.

As in many stories of the period, the threat nuclear war here comes not from either of the superpowers but from a third party. In this case, a Nazi war criminal who, not surprisingly, would like to see Russia and America destroy each other.

You will often hear the attitudes implicit in this story associated with the late Sixties and early Seventies, usually attributed to the escalation of the war and and the rise of a politicized youth culture, but this was 1963. It was early in the war and even the oldest of the boomers were still in high school.

If you surveyed the pop culture of the time you would find other evidence that a radical shift in the way we looked at the cold war took place in the late Fifties and early Sixties. Nuclear war was no longer likely to be depicted as a Pearl Harbor-style attack but rather as either a horrible mistake (Failsafe -- novel 1962, film 1964) or the work of a madman (Red Alert -- 1958/Strangelove* -- 1963) or the terrorist scheme of a third party (too numerous to mention). Anti-communist agents could be as morally compromised as the enemy (The Spy Who Came in from the Cold -- 1963). A sympathetic Russian was so acceptable that by 1964, you could have a loyal and openly communist Russian agent as a co-star in a popular spy show (albeit a Russian played by the Scottish David McCallum).

I've wondered if this shift was a reaction the Cuban Missile Crisis. Talking to those who were around at the time, I get the impression that for all the paranoia and anxiety and civil defense drills, the concept of nuclear war was never truly real to most people until 1962.

Good researchers in sociology or political could probably provide a rigorous answer to the question of what exactly drove the shift. I'd be interested in seeing what they came up with and if they need another topic after that, I have a whole shelf Gold Medal paperbacks for them to check out.

(You can watch Dr. Strangelove online here)

 

What we're talking about when we talk about the AI bubble

 

One. The AI stock bubble. Analogous to, but far larger than, the dot-com bubble. The main thesis here is that, regardless of the future economic impact of AI in general or large language models in particular, the current state of affairs around companies like Anthropic, OpenAI, xAI, and Meta is unsustainable, that there is virtually no chance of the industry becoming profitable enough quickly enough to justify the level of spending, especially associated with data centers, before bills come due and the current generation of chips becomes obsolete. The essential follow for this story has long been Ed Zitron.

Two. The AI hype bubble. Closely related to the AI stock bubble, with a strong symbiotic relationship. You could argue that one cannot exist without the other, but they are still very much distinct. Though it is sometimes couched in pseudo-technical terms, the hype bubble is largely magical, with claims ranging from the silly to the literally apocalyptic.

Three. Benefits and costs. If we limit ourselves strictly to those things that we know large language models are currently capable of, which ones are likely to be economically or socially useful, and which are likely to cause (or already have caused) substantial harm? Examples of the former include coding tools; the ability to extract information from text-based data at scale; and better interfaces for interacting with various devices. The latter includes (and this is a very partial list) creating severe mental health crises; polluting research papers, legal documents, and historical records with false information; degrading the quality of computer code and security systems; drowning original, high-quality writing and art in a sea of AI slop; crowding out potentially more valuable research; taking a huge toll on the environment; and allowing disinformation to be generated at an unprecedented scale.

It is worth noting that there is almost no overlap between the doomsters in the hype bubble and the critics raising the concerns mentioned here.

And sort of...

Four. Where do we go from here? This isn't really part of the bubble conversation, but it is adjacent. The current AI discourse is a profoundly dysfunctional. It is an ugly mass of hyperbole, misinformation, and badly framed conversations. Journalists covering the topic, more often than not, have no understanding of what they're talking about. Statements from the companies developing these models are, at best, based on questionable metrics and assumptions, and, at worst, are borderline lies. Much of the research (arguably most of what makes its way into The New York Times or The Atlantic) is sensationalistic, anecdotal, and terribly thought out.

But we can't and shouldn't simply dismiss large language models and their potential for positive impact. They represent the biggest breakthrough we've ever seen in natural language processing. The changes they have wrought to the way we code are already substantial and will only grow bigger. Our ability to analyze and classify text at scale will have huge implications for countless fields of study.

All of this means that it is imperative that we improve our understanding of this technology and have a serious, productive conversation about how best to use it while limiting the damage it can do. Unfortunately, that's not the conversation we are having.

Generational fairness

This is Joseph.

One of the recent questions that we have had to deal as a society is the way that housing price appreciation has concentrated wealth in older cohorts. This is not a small point as programs like social security are getting more expensive as the population ages and part of the rationale has always been anti-poverty. 

But the worst type of proposal is often been seen in Canada. Where you reform the mechanisms to hurt the generation that has not benefited from housing appreciation. 

Patrick Boyle addresses this generational problem in this video. 

I think that the same issue is more generally an issue across a lot of countries trying to balance generational issues. I admit, as Gen X, I am just assuming now that the systems I was hoping to be used will be trashed by the time I get there. Which is unfortunate. 

And it isn't a Canada-only problem:

At some point there probably needs to be a hard conversation about some of these decisions as the pressure mounts. 

I apologize for calling AI a bubble. It is actually a bouncy house, and right now everybody's listening to the blowers.

Last Monday, Allison Morrow wrote the following

Buckle on up: Alphabet, Amazon, Microsoft and Meta all report earnings on Wednesday — that’s about 19% of the S&P 500 by market cap — are all reporting after the closing bell, just a couple of hours after what is expected to be Powell’s swan song of a press conference announcing the Fed is holding rates steady yet again. Just 24 hours later, Apple also reports earnings. 

 

Of course, we don’t have a crystal ball to know what’s in those reports or how investors will react to them. But here is some important context: Wall Street’s AI fever appears to be back (if it ever went away, and we’ll get to that in a minute), and that makes these particular tech earnings much less about money coming in than money going out. 

 

Investors will be laser-focused on capital expenditures, aka how many dump trucks full of money the companies are committing to their AI buildouts. The “Magnificient 7” stocks that have been propping up the broader market — Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia and Tesla — are expected to raise their combined capex by 30% from last year to at least $680 billion in 2026. 

 

“If they say, 'we're going to continue spending at the pace we’ve been spending, or a faster pace,' that sort of vindicates the the crazy move we've just had in the SOX index,” Steve Sosnick, chief strategist at Interactive Brokers, told me. 

 

ICYMI: There’s a gauge called the Philadelphia Semiconductor Index, or SOX, that tracks the 30 largest US-traded semiconductor companies (like Nvidia, Advanced Micro Devices, Qualcomm, etc). That index has been on a tear, shooting up 45% in just four weeks. Why? Eh, why not. 

 

Strong earnings and positive forward guidance from a few key players helped, but “no new fundamental or technological development justified re-rating the group nearly 50% higher in the span of a month,” Mike O’Rourke, chief market strategist at JonesTrading, said in a note Sunday.

 

“We talk about bubble valuations and the market's pricing mechanism being broken — behavior like this clearly reinforces that thinking.”

Marketplace's analysis of the day's trading made the same "less about money coming in than money going out" point.

The AI landscape is only able to maintain its shape through a constant infusion of cash through endless funding rounds, mountains of debt, and, most importantly, spigots of money from some of the biggest and most profitable companies in the world, such as Microsoft and Alphabet/Google.

If investors truly believed that things were about to turn around and this bottomless money pit was about to become a gusher (sometimes when you get on a metaphor kick, you just can't stop yourself), they shouldn't be that excited one way or the other about whether or not Microsoft et al. keep the air pumps running. Instead, the mentality seems to be: how long do we have until the music stops?

As with so many things, it sounds better when Jeremy Irons says it.

Anthropic's Mythos: One moderate step forward for white-hat hacking; one giant leap forward for journalists' credulity

[I kid, of course. That particular shark was jumped ages ago.] 

You should probably approach any story of large language models displaying initiative, or trying to mislead or blackmail users, or generally doing anything of the sort with the same mindset you approach accounts of paranormal activity. In both cases, virtually all the reporting will be sensationalistic, anecdotal, and likely to collapse under scrutiny.

Example du jour, Anthropic is getting an enormous amount of sky-is-falling coverage over what appears to be the development of a good but hardly revolutionary white-hat hacking tool. 

Here's Gary Marcus's assessment:

To a certain degree, I feel that we were played. The demo was definitely proof of concept that we need to get our regulatory and technical house in order, but not the immediate threat the media and public was lead to believe. 

Not only has the reporting been credulous and incurious, it has largely ignored the ever-present elephants in the room when discussing OpenAI, Anthropic, etc.

Cal Newport follows up:

Since Marcus published his essay, I’ve come across several more similar findings:

• The AI security expert Stanislav Fort ran ​an experiment​ to see if existing, cheap open-weight models could find the same vulnerability in FreeBSD (an open-source operating system) that Anthropic touted as evidence of Mythos’s scary abilities to uncover bugs that had been hiding for decades. The result: all eight existing models they tested discovered the same issue.
• Meanwhile, the renowned security researcher Bruce Schneier ​weighed in​, similarly concluding: “You don’t need Mythos to find the vulnerabilities they found.”

And of course, it doesn’t help that a week before Anthropic released this supposedly super-powered vulnerability detector, they accidentally leaked the Claude Code source, and security researchers immediately found ​serious vulnerabilities​. (I guess Anthropic forgot to use Mythos to clean up their own software…)

Journalists covering this story need to constantly remind themselves that hundreds of billions of dollars, possibly even trillions, are at play here. What's more, the constant flow of funding that keeps this game going appears to be drying up, making this the highest-stakes game of musical chairs ever played. One of the key motivators that has kept the music going this long has been the carefully promoted belief that the end of the world is possibly days away and the only thing that can save us is if the good wizard discovers the incantation before the bad wizard does (at the risk of putting too fine a point on it, the bad wizard here is China).

Software developer Carl Brown of the Internet of Bugs has a good take. In particular, pay close attention to the part about Responsible Disclosure. 

Brown got on my radar through this excellent discussion with Ed Zitron, Over an hour but well worth the time.