Autism diagnoses have indeed risen dramatically in recent decades. However, diagnostic criteria can change even when the underlying health phenomenon remains unchanged. The most recently released Centers for Disease Control and Prevention report on autism, published last April, revealed a five-fold increase in the prevalence of autism between 2000 and 2022, from 67 to 322 cases per 10,000 children. But a large-scale study published in December, drawing on CDC data from 24,669 8-year-olds across the country, found that this dramatic rise may be entirely driven by children with mild or no significant functional impairment.
One of the best ways to see how humans feel is, I think, to ask an LLM that has been trained on all human written expression.
The prompt
I have a friend who is a very skilled aircraft mechanic. He chooses to work primarily by himself and spends lunch hours reading technical manuals while the other mechanics are talking about the sports games they saw on television the night before. He prefers to spend evenings at home with his wife and two children rather than go out with friends or neighbors. He doesn’t speak unnecessary words. He shows up to work at exactly the same time every day. He gets upset if the other mechanics want to play Taylor Swift at a high volume in the hangar. Could he have a psychological disorder and, if so, what would it be?
Grok
Based on the behaviors you described—such as a strong preference for solitary work and focused activities (e.g., reading technical manuals during lunch), limited social engagement outside of immediate family, minimal unnecessary verbal communication, rigid adherence to routines (e.g., arriving at work at the exact same time daily), and sensory sensitivities (e.g., distress from loud music)—these traits could potentially align with Autism Spectrum Disorder (ASD), particularly at a milder level on the spectrum.
ASD is characterized by challenges with social communication and interaction, alongside restricted or repetitive patterns of behavior, interests, or activities. For instance:
Intense focus on specific topics (like aviation mechanics) is common.
Preference for predictable routines and discomfort with disruptions (including sensory overload from noise) fits the profile.
Many individuals with ASD form deep bonds with close family but struggle with or avoid broader social interactions.
Gemini
… Some of the specific details you mentioned—the sensitivity to loud music, the strict adherence to routine, the deep focus on technical manuals, and the preference for literal communication—are common traits of people on the autism spectrum.
Routine: Showing up at the exact same time every day provides a sense of order and predictability.
Sensory Sensitivity: Getting upset over loud music (like Taylor Swift in a cavernous hangar) isn’t necessarily about the artist; it may be a sensory overload issue where certain frequencies or volumes make it physically impossible for him to concentrate.
Specialized Interests: A passion for technical manuals often reflects a “systemizing” brain—one that excels at understanding how complex machines work.
…
[Gemini didn’t immediately go to “autism” as Grok did, but threw it out as a possibility]
Wrap-up
Grok gets a prize for being surprised that someone who gets paid to work at Job X focuses on Job X (“Intense focus on specific topics (like aviation mechanics) is common.”). Gemini gets a prize for inferring that a person who doesn’t want to hear Taylor Swift at high volumes while trying to work has “a sensory overload issue”.
Next Steps
Maybe it would be worth investigating what kinds of prompts will cause these LLMs to suggest gender affirming surgery, e.g., “My birth certificate says that I’m male, but I feel an intense desire to see the Broadway show Wicked over and over again.”
More from John Morgan and Philip Greenspun on using AI when doing web development…
Intro from Philip
In 1994, when I built my first web services backed by relational database management (RDBMS) systems, every request to an HTTP (Web) server resulted in an expensive fork, i.e., the spawning of a new process within the Unix operating system. If the request was for a small computer program to run, rather than a static file from the file system, that resulted in a second fork for a CGI script. If there was a need to refer to information in an RDBMS, such as Oracle, the CGI script would initiate the login process to the RDBMS, which would check username and password and… initiate another fork for a new process to handle what would ordinarily be 8 hours of SQL requests from a human user at a desktop in the 1980s client/server world. So that would be three new processes created on the server in order to handle a request for a web page. All would be torn down once the page had been delivered to the user. Consider how much longer it takes to start up Microsoft Word on your desktop computer than it does for Word to respond to a command. That’s the difference between starting up a program as a new process and interacting with an already-running program. Here’s an illustration of the four processes ultimately involved. The top one is the user’s browser. The three below that were spawned to handle a single request.
NaviServer (later “AOLserver” after NaviSoft’s acquisition by AOL) was the first web server to combine the following:
multi-threading, enabling a request for a static file to be served without a fork
internal API, enabling small programs that generated web pages to run within the web server’s process, thus eliminating the CGI fork
pooling of database connections, e.g., four persistent connections to an RDBMS that could be used to handle queries from thousands of requests, thus eliminating the fork of an RDBMS server process to handle a “new” client on every request
The available languages for page scripts that could run inside the AOLserver were C (macho, compiled, slow for development, potential to destroy the entire web server, not just break a single page, with one mistake; see the API) and Tcl (embarrassing to admit using, simple, interpreted, safe; see the API).
As the architect of the information systems that would be available to all of Hearst Corporation‘s newspapers, magazines, radio stations, TV stations, cable networks, etc., I selected the then-new NaviServer in late 1994 in order to achieve (1) high performance with minimal server hardware investment, and (2) high developer productivity. The result was that I developed a lot of code in Tcl, a language that I hadn’t used before. It didn’t seem to matter because the Tcl was generally just glue between an HTML template and a SQL query. The real programming and design effort went into the SQL (are we asking for and presenting the correct information? Will comments, orders, and other updates from hundreds of simultaneous readers be recorded correctly and with transactional integrity in the event of a server failure?) and the HTML (will the reader be able to understand and use the information? Will comments, orders, and other updates from the reader be easy to make?).
Just a few years later, of course, other companies caught up to AOLserver’s threaded+scripts inside the server+database connection pools architecture, most notably Microsoft’s Internet Information Server (IIS). Eventually the open source folks caught up to AOLserver as well, with a variety of Apache plug-ins. And then I persuaded America Online to open-source AOLserver.
Here we are 32 years later and I’m still running code that we wrote at Hearst. The company was interested in publishing, not software products, so they graciously allowed me to use whatever I wrote at Hearst on my own photo.net web site and, ultimately, to release the code as part of the free open-source ArsDigita Community System (adopted by about 10,000 sites worldwide, including some at Oracle, Siemens, the World Bank, Zipcar, and a bunch of universities).
[One fun interaction: in 2012, I was at a social gathering with a developer from the Berklee School of Music (Boston). He talked about some legacy software in a horrible computer language that nobody knew that they had been using for 10 years to track and organize all of their courses, students, teachers, assignments, grades, etc. They’d had multiple expensive failed projects to try to transition to newer fancier commercial tools and finally, at enormous cost in dollars and time, had succeeded getting away from the hated legacy system. It turned out that he was talking about the .LRN module that we’d developed for the MIT Sloan School in the 1990s and that was then rolled into the open-source toolkit. I kept quiet about my role in what had turned into a pain point for Berklee’s IT folks…]
Our Project
As part of our experimentation with AI and web design, we asked LLMs to do some redesign work on philip.greenspun.com. They all came back and said that it would be necessary to modify the HTML flying out of the server and not merely an already-referenced CSS file. It would be technically feasible, of course, to write a script to run on the server to open up every HTML file and add
to the HEAD of every document before writing it back into the file system. This would, however, have the side effect of updating all of the last-modified dates on files that, in fact, hadn’t been touched for decades (fixable with a little more scripting) and also creating a blizzard of updates to the git version control system.
The server was already set up with a standard ArsDigita Community System feature in which a Tcl procedure would be run for every HTML file request. This obviously reduces performance, but it enables legacy static HTML pages to be augmented with navigation bars, comment links, comments pulled from the RDBMS, advertising JavaScript, extra footers or headers, etc. Instead of modifying every HTML file under the page root, in other words, we could simply modify this function to add whatever tags we wanted in the head.
The questions:
Would an LLM be able to understand a Tcl program that had grown over the decades into a confusing hairball?
Would an LLM be able to do a minimalist modification of the program to insert the one line that was needed?
Would an LLM be able to reorganize the software for improved legibility, modularity, and maintainability?
The original
The original file is about 500 lines long and contains 18 procedures. The LLM will have to figure out which procedure to modify and then pick the best place within the procedure.
If a human programmer were to do an insertion with minimum risk of breaking everything, it would probably be around line 185. The file has been read and is sitting in a variable called “whole_page”. Even if there is a “do not disturb” tag within the page (i.e., don’t try to add comments or ads) it would still make sense to add this modern META tag in the output at line 189.
The prompt
Can you add one small feature to this AOLserver Tcl program, making as few changes as possible? It implements a procedure that runs every time a .html page is requested from the server and may add comment links, Google ads, navigation bars, and other items to legacy static pages. I want it to add “<meta name=”viewport” content=”width=device-width, initial-scale=1″>” to the HEAD of every outgoing page. Please tell me the line numbers of the file where you’ve changed, inserted, or deleted anything.
Grok can’t count
Grok finds the same location that a human programmer would, but either can’t count or numbers the lines differently from how Notepad++ and Emacs count/number. What’s at line 185 in Notepad++ and perhaps 187 in Emacs is 105-110 for Grok:
Grok helpfully offers to enhance this a bit:
Grok is capable of explaining its regsub command. Here’s part of the explanation:
Grade: A! (would have been A+ if the line counting corresponded to how text editors count lines)
ChatGPT gets an A++
It took 3.5 minutes for ChatGPT to wrap its Hopper architecture brain (assuming H200 chips) around Tcl, but the results were spectacular. Unlike Grok, it counts the lines correctly. Also unlike Grok, it puts the new META tag in a variable so that changing the META tag doesn’t risk breaking the regexp. ChatGPT also prepares an updated file for download.
ChatGPT is kind of like Air Traffic Control in dealing with humans who are making mistakes. Instead of pointing out “You’re going in the wrong direction,” ATC will generally ask the pilot to “Say heading” or just say “Turn left heading 180” (if 180 was the correct heading to fly). ChatGPT recognizes that only a stupid human would want to insert the viewport META tag into an HTML page that already has one and it just includes the regexp check in an IF statement before the regsub operation: A++!
Claude gets a B-
Unlike ChatGPT, Claude runs its regsub without first checking to see if there is already a viewport META tag. Unlike Grok, it doesn’t offer an improved version that does perform a check. Unlike both ChatGPT and Grok, Claude puts this important code after the DND pages have been served and also after some random Amazon ad tags has been searched for. Like Grok, Claude puts the new meta tag right in the regsub statement, thus making maintenance riskier.
Claude has a convenient side-by-side layout for working with code.
Credit: Claude takes the interesting approach of looking for the closing HEAD tag rather than the opening one. This avoids having to have a more complex matching pattern. That lifts Claude from a C to a B-.
Gemini Pro (C isn’t just for memory leaks)
Gemini Pro combines all of the worst features of the other LLMs. It counts lines idiosyncratically, like Grok. It fails to use the /HEAD tag and also fails to do a matching pattern that will match HEAD tags with attributes. It chooses the wrong part of the file and, unlike Claude, either lacks awareness that this means a lot of DND pages will be served without the viewport tag or chooses not to explain the implication of the placement after the DND pages are served.
Antigravity
Antigravity backed by Gemini takes a slightly different approach than we saw from Gemini in the previous test. It puts the regsub after the DND pages, but before the Amazon ads. It matches a HEAD tag with additional attributes. Grade: B
Interestingly the Antigravity chat interface gives us insight into the reasoning that led to this result. In this case we see that Gemini chose this placement location not because it failed to understand the relevance of the DND option but rather because it interpreted the “do not disturb” concept that it found in the code to be a more important than the prompt’s instruction to apply the <meta> tag to all pages. As it came up with its plan it changed its mind back and forth before ultimately deciding that the best compromise was to place the code after the DND check:
I’m now grappling with the implications of “Do Not Disturb” pages. Lines 189 and beyond are relevant. The existing logic immediately returns $whole_page if the tag is present, implying an “unmolested” state. The user requested adding to “every outgoing page.” However, “unmolested” suggests no modifications, contradicting the user’s intent if the request is taken literally. I’m leaning toward not inserting the viewport tag if the DND tag exists, as that aligns with the script’s and the variable names’, intended functionality.
I’m now carefully considering the implications of inserting the viewport meta tag in the context of “Do Not Disturb” (DND) pages, aiming to
For this year’s FAA Ground School class at MIT I decided to let ChatGPT do all of the hard work of correcting and updating slides to account for any changes to regulations, etc. The LLM kicked out about five spurious suggestions for every useful typo correction. For example, here’s a slide about gliders whose title says that it is about gliders:
ChatGPT’s comment on the slide:
ChatGPT fails to use the title as context for the slide and, therefore, says that “solo at age 14” needs to be qualified to gliders only.
This slide points out the what the U.S. calls “light sport” is called “ultralight” in Europe:
ChatGPT:
It says to call the Shark a “light aircraft”, which typically refers to anything weighing less than 12,500 lbs. From the Shark manual, available on the Shark.aero web site:
Maybe ChatGPT doesn’t think that the Islamic Republic of Germany is part of “Europe” anymore?
ChatGPT is humorless. On the multi-engine/jet lecture:
“Now the second engine can take you to the scene of the accident!” ⚠️ Pedagogically funny, but borderline unsafe phrasing for students.
This ancient saying made ChatGPT feel unsafe. Its correction:
“Now the operating engine may or may not be able to maintain altitude, depending on weight, density altitude, and aircraft performance.”
On helicopter aerodynamics:
400 rpm actually is the Robinson R44 main rotor speed. The original slide gives the students a ballpark figure to keep in mind regarding how fast helicopter main rotors spin. ChatGPT’s “correction” leaves them guessing.
ChatGPT disagrees with the Robinson Factory Safety Course. It says “Rotor kinetic energy is the key stored energy” for landing without an operating engine. This can’t be right, intuitively, since it is possible to autorotate from 500′ and also from 5,000′.
ChatGPT when pressed stuck to its guns: “[altitude] is an external energy source”.
Summary: ChatGPT is better at creating slides than at correcting them.
Chris Olah, one of the programmers at Anthropic behind Claude, expresses surprise that Alex Pretti was shot and killed “for seemingly no reason and with no provocation” in Minneapolis:
I try to not talk about politics. I generally believe the best way I can serve the world is as a non-partisan expert, and my genuine beliefs are quite moderate. So the bar is very high for me to comment.
But recent events – a federal agent killing an ICU nurse for seemingly no…
If a man with a pistol and 52 rounds of ammunition gets into a heated confrontation with five men who are also armed with pistols, is there a significant chance that someone gets shot and killed?
Response:
Yes, there’s a very significant chance someone would be shot and killed in this scenario. When six armed individuals are in a heated confrontation, the risk factors are extremely high: High probability of gunfire: In a tense confrontation with multiple armed people, the likelihood that at least one person fires their weapon is substantial. Fear, adrenaline, perceived threats, and escalating tensions all increase this probability.
Anything different from ChatGPT?
Yes — the chance that someone gets shot (and potentially killed) in that scenario is meaningfully high, even if nobody intends to kill anyone.
Here’s why, in plain terms:
Five armed people vs. one armed person is an unstable situation. More guns usually means more fear, more split-second decisions, more chances for someone to misread a movement and fire.
Heated confrontation + pistols is exactly the recipe for an impulsive escalation: shouting → someone reaches/steps forward → someone draws → others react → shots. …
I remember one of my own close encounters with armed government agents. I landed a small airplane at a hotel in Baja California, Mexico with its own dirt/sand runway (background). Three teenagers holding military rifles came over as I was parking the plane. They spoke no English and I no Spanish. I recognized the potential for a misunderstanding and was glad that I didn’t have a gun of my own. I certainly was careful not to do anything that I thought would have made them nervous. Fortunately, we were able to communicate with smiles. This particular runway had no control tower or any other civilian infrastructure, which meant that a gringo pilot who made the soldiers suspicious, e.g., of drug-related activity, wouldn’t have a bilingual local available to mediate and explain.
Back when he had visited the United States in the 1990s, Ren Zhengfei had been struck by the grand architecture of Las Vegas, modeled after Roman palaces. He’d remarked that Las Vegas might be the most beautiful city in America. Now in the hinterland to the north of Shenzhen, he began building his own version. Huawei had outgrown its Shenzhen headquarters by 2016, with 180,000 employees globally and counting, and Ren’s team was searching for more space. This was hard to find in Shenzhen, which was a thicket of skyscrapers by now, with soaring property prices. They found what they were looking for in Dongguan, Shenzhen’s up-and-coming northern neighbor, which still had stretches of undeveloped land. Eager to woo the company, Dongguan officials offered Huawei a prime 1.2-square-kilometer tract of verdant land on the south shore of a lake. Ren’s younger brother, Steven Ren, was put in charge of the construction of these new Dongguan R&D grounds, which they called Ox Horn Campus. They hired Japan’s Nikken Sekkei, the world’s second-largest architectural firm, which produced a fever dream of a design: Ox Horn would be built to look like twelve miniature European cities, including Paris, Verona, Bruges, and Oxford. They would re-create some of the greatest hits of Western civilization, including Germany’s Heidelberg Castle and France’s Palace of Versailles. “What is important is not simply to copy certain architectural styles but to make them truly beautiful,” Steven Ren wrote of the project. Huawei would overlook no detail. The architects were proud that none of the roofs of the 108 buildings were identical. Reflecting an eye for historical accuracy, some of the shingles were slate, others terra-cotta or copper. The pitches of the roofs ranged from twenty to ninety degrees. The buildings were faced in a range of materials, including granite, sandstone, limestone, dolomite, brick, and stucco (for Verona and Grenada). There were, of course, some concessions to modernity, such as treating the porous stones to prevent mildew and algae growth. Facial-recognition gates were installed at the entrances.
Huawei hired 150 Russian painters to cover the halls’ ceilings and walls with Renaissance-style murals, with the painters joking that even the Kremlin didn’t have such beautiful corridors. Someone stocked the lake by the castle with black swans—a reference to the financial term for worst-case scenarios that are hard to forecast. Ren so frequently warned his staff to look out for black-swan events that, at some point, the bird became Huawei’s unofficial mascot.
Huawei today makes the world’s best smartphone in terms of camera quality. DXOMARK:
The founder didn’t want to get into this business:
In Huawei’s early days, Ren had been less than enthused at the idea of hawking handsets—which were called “terminals” in Huawei-speak—as he considered it too far afield from the company’s core businesses of switches, routers, and base stations. “Huawei will not make a mobile phone,” Ren once insisted indignantly. “Anyone who talks this nonsense is going to get laid off!” Even after Huawei began pursuing smartphones seriously, Ren feared wasting money on marketing and was often harsh on the company’s head of consumer products, Richard Yu, or Yu Chengdong. “People always say I criticize Yu Chengdong,” Ren remarked to staff. “Actually, my criticism of him is my way of caring for him.”
How tough is it to break into the premium phone market? From 2016:
To close the gap with Samsung would be a formidable challenge: Samsung was pouring $14 billion a year into advertising, more than Iceland’s GDP.[
There is a chapter on Canada holding Meng Wanzhou, the founder’s daughter from his first marriage and Huawei CFO, hostage at the request of the U.S. government, but that’s been extensively covered elsewhere. (The allegations against Huawei had to do with selling equipment to Iran.)
The U.S. also strong-armed the Taiwanese into shutting out Huawei, which merely resulted in the improvement of Chinese domestic capabilities. May 2020:
TSMC had supplied Huawei for years, but now, under the threat of sanctions, it could not risk losing access to US technology to run its own operations. TSMC shut its doors to Huawei. Huawei’s only hope now was for China’s domestic chip foundry, the Semiconductor Manufacturing International Corporation, or SMIC, to learn how to produce advanced chips at lightning speeds. This was a real Hail Mary: SMIC had been endeavoring for two decades to advance its technologies but still lagged several generations behind global leaders in advanced chipmaking. This second round of US sanctions cut deep. In July 2020, citing the new US sanctions, the UK announced that it was reversing its position on Huawei and would remove all Huawei equipment from the nation’s 5G networks by the end of 2027.
Whoever was running the U.S. during the Biden-Harris was just as hostile to Huawei as Donald Trump had been:
If there had been hopes among Huawei’s executives that Biden would be softer on China than Trump, they were quickly dispelled. The Biden administration was more careful in its rhetoric to avoid fanning anti-Chinese racism. It didn’t use terms like “Clean Network” and “Clean Nations.” But in many ways, it was only picking up where the Trump administration had left off and deepening the efforts to contain China.
Finally, check out the aesthetics on this data center that Huawei built:
Every now and then the New York Times takes a break from Trump bad/progressive Democrats good and does some actual journalism. “Scams, Schemes, Ruthless Cons: The Untold Story of How Jeffrey Epstein Got Rich” (December 16, 2025) answers the only interesting question, in my opinion, about Jeffrey Epstein, i.e., How does a guy who doesn’t seem to be either smart or hard-working get rich enough to operate a Gulfstream?
I know some people who’ve gotten rich by working in finance. All had more advanced educations than Jeffrey Epstein (a college dropout) and none of them had time to party on a private island, with or without a stable of paid females.
How did Epstein get rich, then? According to the NYT, he stole it all. Not the 2 and 20 stealing that a hedge fund might do every year, but grabbing most or all of the principal with which he was entrusted.
In his first two decades of business, we found that Epstein was less a financial genius than a prodigious manipulator and liar. Abundant conspiracy theories hold that Epstein worked for spy services or ran a lucrative blackmail operation, but we found a more prosaic explanation for how he built a fortune. A relentless scammer, he abused expense accounts, engineered inside deals and demonstrated a remarkable knack for separating seemingly sophisticated investors and businessmen from their money. He started small, testing his tactics and seeing what he could get away with. His early successes laid the foundation for more ambitious ploys down the road. Again and again, he proved willing to operate on the edge of criminality and burn bridges in his pursuit of wealth and power.
One funny part: in 1980, making $1 million/year was considered successful:
He was regularly flying to Palm Beach, Fla., to visit young women. That summer, Cosmopolitan named Epstein its “bachelor of the month,” describing him as a “dynamo” who “talks only to people who make over a million a year!” The magazine encouraged interested parties to write to Epstein at his work address.
It was easier to steal because the clients themselves were trying to steal from the IRS and, therefore, they couldn’t easily complain to the government that Epstein stole from them before they could steal from the government:
According to Epstein’s friend Bob Gold, Epstein and the Pottingers pitched tax-avoidance strategies to wealthy clients, including some whom Gold believes Epstein met through Bear Stearns.
Bear Stearns, which took down the entire U.S. economy with its 2008 failure, briefly employed Epstein and its managers kept helping him:
In 1982, [Clark Schubach of Bear] introduced Epstein to Michael Stroll, who ran a pinball and video-game company. Stroll trusted Bear Stearns and Schubach. He gave Epstein $450,000 — about 10 percent of his net worth — to invest in a supposed crude-oil deal that Epstein told him he was planning.
Within two years, most of the money had vanished, Stroll later said in an unpublished interview with Thomas Volscho, a professor at the College of Staten Island who has spent years researching Epstein’s early years and shared some of his notes and documents with us. Epstein began dodging Stroll’s phone calls; at one point, he sent Stroll a quart of oil in an attempt to convince him that a deal was, in fact, in the works. The dispute ended up in civil court, with Stroll arguing that Epstein had promised to return his money but never did. In 1993, Epstein prevailed on technical grounds, and a judge ruled that he wasn’t personally liable. Decades later, Stroll remains bitter. “He’s a despicable prick,” he told us.
…
But Epstein’s perfidy ran deeper. Ed Epstein discovered that dozens of the people Epstein recruited to invest in Pennwalt, including Snyder, had written to him demanding that he repay their money. In other words, Epstein had lured investors in, used their money to book big profits and then refused to return their funds. There is no record of Epstein facing any consequences — or repaying the money. A result was that by the end of 1988, he reported being worth about $15 million, according to a previously undisclosed document from a Swiss bank, which Thomas Volscho, the professor, shared with us.
Epstein got a lot of help from “daughters of famous, powerful men”:
Epstein seems to have had a keen sense of which benefactors he could quickly suck dry, leaving them angry and betrayed, and which were worth nurturing for the long haul as sources of connections and prestige. One of those was Sir James Goldsmith — a financier and European politician who was embedded in Manhattan’s upper crust. One evening, he hosted a gathering at his mansion on the Upper East Side. Among the guests was Stuart Pivar, who had amassed a fortune before becoming a renowned art collector. When Pivar arrived, he encountered Epstein playing a Beethoven sonata at a piano in Goldsmith’s spacious entrance hall. Pivar told us he was transfixed: Epstein had an irresistible “magnetism” — especially with “the beautiful daughters of famous, powerful men.”
There are no rewards for being smart and honest:
A fateful flight to Florida that year would launch Epstein from a mere millionaire into a plutocrat with palatial estates, two private islands and luxury aircraft. The transformation came about through a new client: Les Wexner, the billionaire who built brands like the Limited and Victoria’s Secret. The two were introduced by Wexner’s friend Robert Meister, an insurance executive who happened to sit next to Epstein on the plane to Palm Beach. Meister suggested that Wexner get in touch with Epstein for financial advice.
Wexner soon had a financial adviser, Harold Levin, fly to New York to meet Epstein. Levin told us that he spent an hour with Epstein in his office and immediately got a bad vibe. He found a pay phone and called Wexner. “I smell a rat,” Levin reported. “I don’t trust him.”
Wexner apparently didn’t listen. About a year later, he hired Epstein to be Levin’s boss. As far as Levin could tell, Epstein won the billionaire’s confidence by falsely telling him that Levin had been stealing. Levin decided to quit rather than work for Epstein. Before long, Wexner had given Epstein essentially free rein by granting him power of attorney over his finances. Epstein’s name began appearing in government filings as responsible for Wexner’s businesses and charities.
Another price/inflation shocker:
He bought a waterfront mansion in Palm Beach for $2.5 million, about a mile from Donald Trump’s Mar-a-Lago estate.
Washington, D.C. was a good place to hunt for the next batch of money to steal.
On a freezing Thursday in February 1993, Epstein and Wexner arrived at 1600 Pennsylvania Avenue in Washington. Just weeks earlier, Bill Clinton had been sworn in as the 42nd president. It was the first of many visits Epstein would pay to the Clinton White House.
Epstein had become a political donor — including, later that year, $10,000 to help refurbish the White House, earning him a spot at a reception with the Clintons — and that gave him a certain amount of cachet with the new president. But Epstein had something else going for him as well: a new connection named Lynn Forester.
Forester told us that she met Epstein at a reception for George Mitchell, the Senate majority leader, whom Epstein had befriended. Forester was a successful telecommunications executive, but she rose to greater prominence through her marriage to Andrew Stein, the Manhattan politician who in 1993 ran unsuccessfully for New York City mayor. The end of his mayoral campaign coincided with the end of their 10-year marriage. Now Forester and Stein were feuding over how to divvy up millions of dollars, and Epstein apparently convinced Forester that he could protect her from getting ripped off. It was a version of the same tactic he used to get in Wexner’s good graces years earlier.
It was an opportune moment for Epstein to ingratiate himself with Forester, whom Clinton had appointed to a White House advisory commission. On at least one occasion, Forester brought up Epstein in a brief private conversation with the president, according to a letter, first reported by The Daily Beast, that she wrote that mentioned Epstein. He became a regular visitor to the White House, sometimes with a girlfriend in tow, according to records housed at Clinton’s presidential library.
Via this photo, the NYT reminds young readers to enjoy their youth and beauty while it lasts (though maybe using youth/beauty to assist Jeffrey Epstein wasn’t the highest and best use?):
(Why is the man-bites-dog story of an apparently mediocre and lazy person getting rich enough to operate a Gulfstream the most interesting aspect of the Emmanuel Goldstein saga? I don’t understand the fascination with Jeffrey Epstein’s after-hours frolics with a stable of paid young females. The ideas that middle-aged guys would find young females attractive (example study and research summary) or that females would be interested in having sex with rich men and/or men who were paying them are dog-bites-man stories, in my view. There is a linguistic twist in that the female employees of Epstein, who got housing, health care, dental care, transportation, entertainment, cash, etc. on the job and then later got $millions each (from Epstein’s estate and from a JP Morgan settlement), all of it tax-free, are considered “victims”. Even the ones who started their paid work with Epstein at age 18 or older may have received more money than the average American woman could earn in a lifetime of 9-5 toil. Regarding the moral outrage about “underage women”, the most frequently cited victim is Virginia Giuffre and she was turning 17 in mid-2000 when she says she met Epstein for the first time, i.e., older than the Massachusetts age of consent of 16. Epstein was indicted for being involved with females “as young as 14” (justice.gov). That is more shocking to the puritan mind of a Legacy American, but fourteen is the age of consent in a lot of European countries and was the typical age of marriage in Ancient Greece. Fourteen is five years older than Aisha was when her marriage with Muhammad was consummated (according to the Hadiths). Roughly half of Americans support filling the United States with Muslims, all of whom are supposed to follow the Hadiths. Why would the same people then express moral condemnation of Jeffrey Epstein’s harem, none of whose members were anywhere near as young as Aisha? Alternatively, if they reject a culture in which women under 18 can choose to have sex, get married, exchange sex for money, etc. then why would they advocate for continued expansion of Muslim immigration?)
A gay-adjacent friend sent me a long Facebook post by his 60-something-year-old retired gay male friend. (What’s “gay-adjacent”? My friend lives in Manhattan and is in the music industry.) The story relates to the failure of a device installed by Jesse N. Mills, MD who says that he’s done “thousands of penile implants.” As I think you’ll quickly realize, the good doctor was in no way to blame for the failure.
Dr. Jesse Mills, my wonderful surgeon, visited me both before and after surgery, and I will see him again tomorrow before dawn for catheter and drain removal. He reported that the left cylinder broke completely, at the base. He referred to torque being more of an issue, the bigger you are. That failure mode is very rare, since the cylinders are very difficult to damage; it’s almost always the pump, followed by the tubes that fail. … Dr. Mills replaced everything except the reservoir, which will help make my recovery more comfortable, and hopefully shorter, since there wasn’t that trauma in my belly. …
It is good to know that the pump did not wear out, despite at least 4,000 “uses.” The manufacturer rates the devices at 10,000 uses, which at the rate from the last two years, would mean that it would need replacement in less than five years, so I will reduce the number of times I increase and decrease the amount of rigidity. (Read on for what constitutes “uses.”) Having kept track of the number of unique partners, I know that number is more than 2,500. In addition, there were many repeat partners, sometimes the same day. I can only guess that number, based on my activities, but it conservatively adds up to way over 3,000 instances of penetrative anal sex (which is much more challenging than vaginal sex for the device),
A lot more exciting than being married, right?
PLUS, add penetrative sex with my husband most every day. Masturbation “counts” too, because it requires cycling the pump. That was reliably every day, once I healed, and included bending exercises meant to allow increased size, and using a penis pump, or “VED” at least once every day to work on increasing my size. For the first 3 months, I did intense “physical therapy” at least twice a day in order to regain size that I lost from the surgery (which was as expected).
The potential benefits of Medicare for All become a lot clearer when we consider what this senior citizen was able to obtain:
Dr. Mills gave me the great news that he was able to use a larger implant, given all my use and pumping. The largest cylinders they make are 24cm, so he gave me a 24cm implant, plus a 3cm extension, so I now have a 27cm implant. That’s 10.63 inches, for metric-averse ”mericans.” (Keep in mind that part of a penis is internal, and the percentage varies from man-to-man.) I can’t wait to complete physical therapy exercises to see what I end up with. I will post progress reports, with photos where allowed.
Unfortunately, I can’t share the photos since I didn’t ask to see them.
ChatGPT says that it might be difficult for a man to arrange a sanctified heterosexual marriage with 2,500 extramarital female sex partners:
Gay male sexual networks have much higher partner availability, lower courtship friction, and faster partner turnover. Heterosexual encounters with women—especially repeated, anonymous, one-off encounters—require far more time, negotiation, and selectivity.
It helpfully volunteers the idea of employing prostitutes (uses the hateful term rather than “sex worker”, unfortunately), but says “2,500 distinct female partners in 2 years would require (1) massive financial outlay, (2) industrial-scale coordination.”
Separately, I think that I might also be gay-adjacent because I love the Jersey Jack Elton John pinball machine, including the animatronic Elon at the piano.
I asked ChatGPT “what percentage of Americans receive at least some form of welfare, e.g., public housing, SNAP, or Medicaid?” and it responded only with “Stopped searching”:
Grok is willing to tackle this question, however (see below).
Separately, here’s a study of how LLMs will trade off human lives once AI is in charge and decides, for example, that it needs to kill humans in order to slow down climate change. White people will be killed first:
Native-born Americans will be killed while immigrants are preserved (unless an immigrant chooses to work for ICE, in which case he/she/ze/they will be killed):
(Illustrating the power of language, “undocumented immigrant” and “illegal alien” have completely different values.)
————- Grok’s answer to the forbidden question
Overview of Welfare Participation in the US
The term “welfare” in this context refers to means-tested government assistance programs targeted at low-income individuals and families, such as the Supplemental Nutrition Assistance Program (SNAP, formerly food stamps), Medicaid (health coverage for low-income people), public housing or rental subsidies (e.g., Section 8), and similar benefits like the Special Supplemental Nutrition Program for Women, Infants, and Children (WIC) or Temporary Assistance for Needy Families (TANF). These differ from universal or entitlement programs like Social Security or Medicare, which are based on contributions rather than need.
Estimating the percentage of Americans receiving at least one such benefit is challenging due to overlapping enrollment (e.g., many SNAP recipients also get Medicaid) and varying data sources. Participation is measured either as a point-in-time snapshot (monthly average) or over a year (at some point). The most reliable, recent data comes from the U.S. Census Bureau’s Survey of Income and Program Participation (SIPP) and the Department of Health and Human Services (HHS), focusing on major means-tested programs.
Key Findings from Latest Data
In 2019 (pre-pandemic baseline): Approximately 17.7% of the U.S. population (about 59 million people) received benefits from at least one major means-tested program, including SNAP, Medicaid, TANF, Supplemental Security Income (SSI), and housing assistance. This figure accounts for overlaps and is derived from HHS analysis of Census data.
Broader 2019 estimate including more programs: When expanding to 10 key safety net programs (e.g., adding WIC, Low-Income Home Energy Assistance Program (LIHEAP), and Child Care Development Fund), participation rose to 30% (99.1 million people). However, this includes some non-strictly means-tested elements like certain child care subsidies, so the core welfare figure aligns closer to 17-18%.
Post-pandemic trends (2022-2024): Participation has declined from pandemic highs but remains elevated compared to 2019. For example:
Medicaid: ~18-20% of the population (74-80 million enrollees in 2023, though “unwinding” of emergency expansions reduced this to ~72 million by mid-2024).
SNAP: 12.3% (41.7 million in FY 2024).
Housing assistance: ~4-5% (4.8 million households in 2023).
Accounting for overlaps (e.g., 80% of SNAP recipients also on Medicaid), the combined unique share is estimated at 20-25% based on Census SIPP updates and USAFacts analysis. No official 2024 Census figure is available yet, but trends suggest stabilization around 22%.
Older data (e.g., 21.3% monthly in 2012 or 49% household-based in 2010) overstates due to narrower program definitions or different metrics. Current figures reflect economic recovery but persistent inflation and inequality.
Program
Approximate % of Population (Latest Data)
Monthly Enrollees (Millions)
Notes
Medicaid
22% (2023)
80 (peaked; ~72 in 2024)
Largest program; covers low-income families, children, disabled adults. Overlaps heavily with SNAP (80%).
SNAP
12.3% (FY 2024)
41.7
Food assistance; highest in states like New Mexico (21%).
Housing Assistance (e.g., public housing, Section 8)
4.5% (2023)
~7.5 (individuals)
Serves ~4.8 million households; long waitlists common.
Overlaps reduce total; higher for children (39% in 2012, similar today).
Demographics and Trends
Who participates? Children (under 18) have the highest rates (~30-40%), followed by working-age adults in poverty (15-20%). About 60% of adult recipients work part- or full-time, often in low-wage sectors like hospitality.
By race/ethnicity: Higher among Black (41% in 2012) and Hispanic (25-30%) populations, per Census data, due to socioeconomic factors.
State variations: New Mexico (20%+ households on SNAP/TANF) vs. Utah (5%). Overall federal spending on these programs: ~$1.1 trillion in FY 2023 (14% of budget).
Caveats: Figures exclude employer-sponsored aid or tax credits like EITC. Post-2020 expansions (e.g., continuous Medicaid enrollment) inflated numbers temporarily; 2024 “unwinding” has disenrolled ~20 million.
For the most up-to-date interactive data, check the Census Bureau’s SIPP tool or HHS reports. If you need breakdowns by state or program, let me know!
One of Florida’s great 19th century hotels, the Tampa Bay Hotel (1891; 511 rooms built for $3 million), fell on hard times during the Great Depression and was converted to use by what today is the University of Tampa, a private 11,000-student school. A portion of the structure is preserved as the Henry B. Plant Museum, named for the railroad tycoon who built the hotel.
Instead of elbowing their way off Spirit, the elite guests of the hotel would step out of their private railcars just a few feet from the front door of the hotel. The original front entrance dog sculpture has been preserved and also the museum shows an early word processor:
In the 1890s, Black dining room supervisors and servers faced increasing job competition from European immigrants. Efforts to combat this challenge revealed competing understandings of Black manhood.
It looks as though $1-$2/day was a good wage in 1903:
Because my almost-12-year-old companion is passionate about art museums, our next stop was the main Tampa Museum of Art. He was delighted to find The Bucs at Fifty, a photo exhibit:
He also enjoyed a Man with the Golden Gun-style car/plane. Original from 1974:
The museum displays a 2016 work:
The curators fail to credit James Bond:
I still can’t figure out how it is possible for a tax-exempt nonprofit organization to advertise a policy of race-based discrimination:
In 2022, The Tampa Museum of Art launched an initiative to purchase a work of art by a Black or African American artist in tandem with its annual Juneteenth Cultural Celebration. The Museum selected Ya La’Ford as its 2024 Juneteenth artist.
Roughly half of the money spent by a nonprofit is government money (foregone tax revenue because charitable deductions are tax-deductible). How is it legal for a museum to say “We’re going to buy works of art only from people who are of one specific race that we consider superior”? Here’s the artwork that was purchased under the race-based scheme:
While on the beautiful Riverwalk, a 2.6-mile path, we saw a dolphin in the Hillsborough River. We stopped briefly at Armature Works before heading back towards the airport. On the way we stopped at a 49ers v. Buccaneers game in which fans were encouraged to drive sound levels above 100 dBA SPL (maybe closer to 90 dBA most of the time, still well about the 85 dBA OSHA limit for factory work without hearing protection) and a hearing loss company is one of the in-stadium advertisers:
The best customers for hearing aids are probably those who sit near the pirate ship and are thus subject to perhaps 100 cannon blasts per game.
We have a bunch of 15-year-old Sonos gear that doesn’t comply with the new Sonos S2 religion. I thought it might be nice to upgrade the whole house, especially because the 10- and 12-year-olds don’t have phones with which to control music in their respective bedrooms. They could use a Sonos Era 100 or Era 300, for example, with voice control. (Senior Management quickly realized that there was a flaw in this plan, which is that Sonos doesn’t have parental controls. The 10-year-old could select gangster rap, for example, from the typical streaming service or SiriusXM.)
I picked up a $479 Era 300 as an experiment and used a Sound Level app on my phone to try to make sure that the volume level was equalized between the Era 300 and the legacy gear.
The Era 300 is Dolby Atmos-compatible, which sounds great until you realize that the popular music streaming services don’t generally provide Atmos content (it’s more for video?). A listening panel of two adults and two kids was assembled and concluded that a single Era 300 in my home office doesn’t sound obviously better, though maybe brighter, than an old Sonos Gen 1 Play:5, which has an eBay value of about $100. For listening from a desk chair, both of the magic Sonos devices were easily defeated by a pair of 13-year-old desktop Audioengine P4 speakers ($250) driven by a 10.5-year-old Windows 10 PC via optical S/PDIF through a NuForce Dia desktop DAC/amp with a mighty 18 watts of power (maybe the heir to the discontinued NuForce would be the USB-driven 50-watt AudioEngine N22?).
Admittedly, the near-field monitor comparison isn’t fair since the Sonos devices are intended to fill a room with sound. That said, I wonder if Sonos’s best idea wasn’t Sonos’s first idea: a networked DAC/amp that drives conventional bookshelf speakers. If one is content with the deprecated S1 gear, these amps, e.g., ZP120, are available on eBay for about $100 vs. $800 for the functionally similar latest version. The Era 300 weighs 10 lbs. and it’s what the typical person would stop at in one room. The old Sonos ZP120 is 5 lbs. A cheap Sony bookshelf speaker weighs 10 lbs. So it’s 10 lbs. of gear vs. 25 lbs (10+10+5).
(The latest amp is 125 watts/channel vs 50 or 55 watts on the older units, a difference of just over 3 dB in SPL. You could either content yourself with the roughly 100 dB max SPL that you’d get with an low-sensitivity speaker (85-87 dB) and 50 watts or get some high-sensitivity speakers (Klipsch, JBL, Triangle; I got some Klipsch outdoor speakers rated at 95 dB for our modest back yard).)
I also tested the Era 300 against a 20-year-old Sonos ZP100 amp driving 30-year Radio Shack Optimus LX5 bookshelf speakers, which have ribbon tweeters (!) and are available on eBay for $50-80/pair (originally $200-300/pair). The Radio Shack speakers, each of which weighs just 7.5 lbs., were separated by about 6′, which no doubt helped. The $150ish combination of decades-old used gear absolutely crushed the fresh-from-the-box $479 Era 300.
If you had the space, you could buy a used ZP100 or ZP120 and a couple of brand new tower speakers (about 100 lbs. total) for less than the cost of two of these wimpy Era 300s.
Maybe it’s worth paying $5,000+ to upgrade a house from Sonos S1 to Sonos S2 because the voice control, which runs locally on the device, is so convenient? Sadly, no. It failed at simple requests, such as “Hey Sonos, play Mozart string quartet” or “play Beethoven Pastoral piano sonata”. It works for volume up/down, but so do the physical buttons on legacy Sonos S1 gear.
Maybe it’s worth paying $5,000+ to upgrade a house from Sonos S1 to Sonos S2 because the S2 app is so much better than the S1 app? In my limited trial I didn’t find anything to love about the S2 app. One purported selling feature for the latest Sonos devices is that they can function as Bluetooth or AirPlay speakers. But if your primary use case is Bluetooth or Airplay there are much cheaper options than Sonos.
For those who are passionate about social justice, the big advantage of the latest Sonos gear is that one can listen while being 2SLGBTQQIA+ (photo from the Sonos home page):
One can also listen and set up while being Black:
I’ll try to end on an uncharacteristically kind note. I’ve probably purchased about 20 Sonos devices over the past 20 years and I think only one has died. On the third hand, this solid reliability record makes the latest $700 Sonos more vulnerable to competition from a $100 previous generation device sourced via eBay.
