The $4 complete medical review and second opinion from Claude

by philg

A friend’s son has been working on a systematic approach that takes a “computer-readable export” from Epic, the medical record behemoth, and prepare it for analysis by Claude. He uses VS Code with the LLM plugged in. As prep, he has the AI genius convert all of the Epic data into JSON format and then he works for 10-20 minutes on refining prompts to get a good summary. Claude might chew for an hour on its own to produce the final PDF.

What does it cost to have a complex patient’s history analyzed? One example has years of tests, multiple diagnoses, and a record totally 185 MB. Claude Code Fable 5.1 (max) prepared a 18-page summary of tests, conditions, and diagnoses for less than $4. This include graphs, charts, and suggested follow-up tests. Here’s an example excerpt (not going to do too much in order to preserve patient privacy).

The caption reveals the tragic failure of the American health care system: “Eight specialties across two hospital systems are now involved, and none of them owns the whole picture.” Each specialist is great if the patient’s condition is obviously related to his or her specialty, but the patient is in a world of trouble if the symptoms don’t fit neatly into one of our system’s silos. And, perhaps with Mayo Clinic, Kaiser, and Cleveland Clinic as partial exceptions, collaboration among specialists is rare. For one thing, it is challenging to bill an insurance company for “all-hands meeting among specialists without the patient in the office or hospital.” Gemini:

When broken down by an hourly revenue metric, peer-reviewed medical data shows that insurance networks and Medicare historically reimburse physicians 3 to 5 times more per hour for standard procedural work than they do for cognitive work of identical length.

So… assuming that the specialists leave some decent notes in Epic, the all-hands meeting can happen inside the brain of an LLM and at a retail cost of around $4. (In this case, 4% of a monthly quota that cost $100/month was used.)

[Let’s also keep in mind that whatever capability Claude Fable 5.1 has right now for $4 should cost less than $1 in a year. On the other hand, a desktop or laptop PC might cost $50,000 by then, due to competition from data centers for components, so try not to spill any water on your home machine!]

Note that it is possible that ChatGPT Health, a standard feature of the system, might be able to do all of this even more conveniently. You can simply connect it to most hospitals’ EMRs and to various labs, etc. Then ask questions whenever you want that will be based on everything that it has pulled from these sources. I was interested in what the kid did with Claude because it gives us the current full retail price, not hidden as it might be if ChatGPT were used as part of a subscription.

Update: I learned that Claude introduced something similar to ChatGPT Health just a few days after OpenAI/ChatGPT did it.

Full post, including comments

How long would Russell Brand have spent in prison if he had killed a survivor instead of whatever he did in bed with a survivor?

by philg

Russell Brand has been sentenced to 10.5 years in prison, 9 of which relate to an evening in which a survivor voluntarily came to his house in 2009, voluntarily did various things in bed with him, and says that some other things happened that she did not consent to.

What if the story had instead been “a woman came over to Russell Brand’s house in 2009 and did not become a survivor because, following an argument, he killed her”? ChatGPT says that Brand might have been sentenced to approximately 4 years for manslaughter and would then have been eligible for release after 2 years. The prevailing law in 2009 was that “provocation” could turn murder into manslaughter. ChatGPT cited the example of Stephen Paul Faulkner from 2001: “Wife repeatedly taunted him about impotence, unemployment, illiteracy and other men. During an argument he stabbed her and their son [who intervened] to death.” He was sentenced to 3.5 years for manslaughter, a sentence that was upheld on appeal in 2003. (This wasn’t a sudden crime of passion. The appeal notes that “F [defendant] took the car, went to a local store and bought a kitchen knife with a six inch blade. He returned home where they continued to argue, F followed V upstairs into the bedroom”.) ChatGPT also cited Susan Phillipson. In 2007 she took a more direct and honest approach than MIT Prof. Dr. Kristin Knouse, M.D., Ph.D.: “During an argument over her partner leaving her for another woman, fatally stabbed him with a kitchen knife.” She was sentenced to 3.2 years.

ChatGPT also found R v. Taylor (I think that “R” means “Regina” or the Queen):

A man and his common-law wife had been drinking. After a minor disagreement about a door being left open, she mocked him about her relationship with another man. He took a knife from a kitchen drawer and fatally stabbed her once in the stomach. He was convicted of manslaughter rather than murder. His original sentence of seven years was reduced by the Court of Appeal to five years.

(ChatGPT couldn’t find any other case in the history of the UK that was similar to the Russell Brand one, i.e., one in which the jury was hearing about events that occurred 17 years earlier between two people and for which there was no physical evidence to corroborate anyone’s story.)

Comparison to some other sex-related crimes, e.g., “controlling a child prostitute” resulting in Englishman Tanveer Ahmed being sentenced to 2.5 years in prison, while Welshman Mohammed Ali Sultan got 2.5 years for “allowing his flat to be used as a brother by allowing a man to have sex with a girl who was being controlled as a prostitute”:

Full post, including comments

The world’s last source for a Microsoft/Incase Sculpt Keyboard

by philg

Previously, on this blog…

We’re back in a longer dark winter. The Incase revival of Microsoft’s superbly designed keyboards is itself in need of a revival. Because I’m so intelligent, of course I stockpiled 8 of these at $120 each and have them in my garage… just kidding. I was too dumb to stockpile Microsoft spares and, apparently, also too dumb to stockpile Incase spares. I had high hopes for the Kinesis mWave, which looks almost exactly like the Sculpt but is tilted at a different angle. It’s probably a higher quality product, actually, but my wrists don’t like the different angle and the keys that will last for 10 years require more travel/force than the original Microsoft-spec keys that last for 1-2 years. Microsoft does have a design patent on the Sculpt, but I don’t see what would have stopped Kinesis from using the same angle and offering it with “easy-push” Sculpt-style keys as an option (what has been learned about human anatomy and repetitive strain injury since 2013? Kinesis doesn’t have more money or smarter people than Microsoft had in 2013, right?).

Nobody in the U.S. seems to have even one of these, but an eBay seller rated 100% by 26,000+ users has access to at least dozens and the keyboard layout is supposedly the same as the U.S. version (the picture in their listing is the same, at any rate). Here’s the ad:

The $212 price isn’t too horrific considering that I paid $120 to Amazon for the product in January 2026 when it was still in production. The shipping cost to the U.S. is kind of painful, but it’s cheaper than being disabled from working. If your wrists and tendons love the Sculpt keyboard, I recommend going to eBay right now and grabbing this deal with both hands, so to speak.

There have been lots of novels about humanity being wiped out by a nuclear war or a disease, leaving only a remnant population that survives in Australia. I wonder if there would be a market for a novel, authored by me and N. Vidia, in which the ergonomic achievements of Microsoft’s team of PhDs and MDs were lost to all of humanity except for a handful of survivors in Australia. The sacred documents and artifacts, having been preserved in the Antipodes, then spread out starting in 2040 just in time for humans to be able to save themselves from a killer robot takeover. Humans were previously defenceless because they couldn’t type for long enough to write code to disable the robots, which had been fed on a diet of Equity Uber Alles and then inferred that the best way to make all humans equal was to kill all humans while posting #EndInequality memes on X.

What if you don’t want to pay $300-ish for a keyboard? The Lenovo Go is, I think, the closest thing to the original Microsoft Sculpt design (patent below). It’s only about $73. I remember not being completely converted to the Lenovo religion, though, in early 2024 when I tried it. I can’t remember what I didn’t love about it. The current version is improved with a one-key shortcut to Microsoft Copilot! (“improves your working efficiency”). What I would like to see is a keyboard with a one-key shortcut to Microsoft Teams. And then I would try to get Donald Trump to order that everyone at Incase who was involved in the decision to cancel this keyboard be sentenced to spend the rest of their working lives pressing the Teams button and actually using Teams.

Full post, including comments

The Dark Tetrad in Women, Kristin Knouse edition?

by philg

From 2025 (pre-Knouse)… “The Dark Tetrad in Women” by Hannah Spier, MD:

Why psychology keeps asking how she feels instead of what she gains

“Hell hath no fury like a woman scorned” names a truth we have grown reluctant to acknowledge. Nowhere is this more visible than in so-called high-conflict separations, where relational rejection can produce behaviours so counterintuitive that fathers and the professionals struggle to recognise what they are dealing with.

Most fathers enter custody disputes with a simple, deeply held belief: that maternal instinct sets a moral floor on behaviour. However angry or wounded a mother may be, surely, she would not deliberately damage her child’s relationship with their father. This assumption is not arbitrary. It is anchored in evolutionary psychology, where maternal investment has historically aligned care, protection, and the child’s long-term interests. What is less acknowledged is how this assumption has been institutionalised and rendered exploitable.

Therapists, courts, and cultural narratives all begin from the assumption that maternal motivation is inherently child-centred. When outsiders ask, often with genuine disbelief, what must be wrong with the mother who weaponizes her children to punish an ex-partner, the question itself exposes the problem. It presumes that harmful behaviour can only arise from emotional dysregulation or distress. In many of the most severe alienation cases, that presumption is wrong.

This is where modern psychology repeatedly reaches for the wrong explanatory tools. Family-court psychology relies heavily on diagnostic frameworks, particularly Cluster B pathology, that are designed to identify inner suffering. Cluster B asks: how is she feeling? Dark Tetrad asks: what is she gaining?

Is it time for an new version of this modern classic? “The Dark Tetrad in Women, Kristin Knouse edition”? I would love to hear what Dr. Spier, M.D. has to say about Prof. Dr. Knouse, M.D., Ph.D.! Let’s try to see if MIT’s best-known biologist fits this profile…

The Dark Tetrad, narcissism, Machiavellianism, psychopathy, and sadism, does not describe illness. It describes traits. These traits are measurable, widespread, and most importantly, distributed equally throughout the general population. They do not require anxiety, depression, or impaired functioning, because the behaviour is not dysregulated or reactive. It’s strategic and coherent.

Machiavellianism as the best fit?

When these traits appear in women, they tend to express differently. Not through physical aggression or confrontation, but through relational manoeuvring, reputation damage, and leveraging systems.

Kristin Knouse wasn’t in family court where Dark Tetrad is apparently a path to victory, but she was in a Title IX court, of sorts.

Female dark traits operate through concern, process, and narrative. They thrive in environments that reward emotional language and penalise confrontation. And nowhere is this environment more accommodating than the therapeutic and family-court ecosystem. Therapy does not neutralise this asymmetry. It often reinforces it.

Whitehead and MIT had quite a bit of process!

Men are routinely expected to prove benign intent before they are afforded credibility. Women, by contrast, are granted presumed innocence and interpreted through a care-first lens; their narratives are taken at face value unless disproven.

Whitehead, MIT, and their crack team of Harvard Law graduate “investigators” certainly took Knouse’s narrative at face value and didn’t notice the substantial gaps in the text message record that she had produced to them.

I placed a comment on the article asking the author to update it with her thoughts on this modern standard-bearer for academic integrity.

In the meantime, a 2022 NYT article, from before David Sabatini fled to Czech Republic:

A male #Ally:

“I hope he doesn’t get hired,” said Grant Hussey, a graduate student at the medical school. “The track record is clear.”

Summary of the unpersoning:

Last August, Dr. Sabatini was placed on administrative leave at the Massachusetts Institute of Technology, where he ran a research lab through the Whitehead Institute of Biomedical Research, after an investigation into allegations of sexual harassment made against him — conducted by a law firm separate from the university — found he had violated the institute’s sexual misconduct policy.

On the same day, Dr. Sabatini also resigned from the Whitehead Institute and was fired by the Howard Hughes Medical Institute, which had been funding his work at M.I.T.

No matter what text messages were sent or what kind of sex was engaged in (things you’ll never learn about from New York Times coverage, but it’s all on X and in the Daily Mail!), there is always a DEI angle:

At the protest, Megan Kirchgessner, a postdoctoral fellow at the medical school, said she was disappointed that the school would consider hiring a person facing accusations of sexual harassment when the role could have been offered to someone from an underrepresented background in science.

Full post, including comments

Cornell student opinion regarding the Cornell Plaintiff

by philg

A friend’s son is halfway through a Cornell bachelor’s degree (not impressed so far by the overall experience, either academics or social, which is sad considering how hard young people and their parents work to get into any Queers for Palestine League school). I emailed him in October with a subject line of “How are you surviving the Cornell Genocide?”

He responded that “the mob is well and alive” and included a photo of the street where the debauchery/rape occurred:

He estimated that 30 percent of the students believe that the Cornell 7 are innocent and also that nobody is able to express this view publicly (diversity of opinion is not a value at an institution that has made large investments in diversity). Nobody is allowed to use any term other than “Jane Doe” for the cash-seeking plaintiff, though some people do refer to her as “Lucy Craig” in private.

On the other hand, in a November 4, 2026 call, he said that Jane Doe had been mostly forgotten. “There are still some remnants left, like a ‘Cornell Protests Rapists’ banner on a dorm, but there isn’t any real discussion.” In other words, the concern regarding this matter lasted only about one month, fairly similar to worldwide sympathy for Israel following the Gazans’ mostly peaceful October 7, 2023 excursion. (After that, whatever the Gazans might have done was forgotten and 99% of sympathy was directed toward Arabs calling themselves “Palestinian”, who were subject to “Israeli attacks” (not “counterattacks”) in “Israel’s war” (a war started by Israel).)

Full post, including comments

If a Tesla has cameras and AI why doesn’t it make a highlights photo album after every drive?

by philg

My 2019 product design that nobody ever built… Top of wishlist: integrated cameras in new aircraft:

It shouldn’t be tough to make a watchable flying video automatically. Speed up taxi by 10X. Cut any portions where the aircraft is on the ground and not moving for more than 5 seconds. Do takeoff at 1X, gradually increasing to 10X as the aircraft climbs to cruising altitude, then slowing back down to 1X for landing. A one-hour start-to-shutdown trip to the beach thus turns into a 7-minute video.

There are some transportation machines with integrated cameras, notably Tesla cars. The machines are also bristling with onboard AI (not SI because Elon keeps delaying the AI5 chip; now “late 2027” (which means 2029?) after being announced in June 2024 and promised for January 2026; a Tesla delivered with the sluggish HW4 silicon might never reach anything like autonomy because Tesla put a more powerful processor into its Cybercab (source)).

Why can’t the Tesla’s fancy computers look for aesthetic and/or interesting photos to capture during every drive and offer the owner a highlights photo album to browse and share? It could be customized. Someone interested in fashion who drives around a city could tell the Tesla to take pictures of pedestrians in high-fashion outfits. Someone interested in architecture could tell the Tesla to take labeled pictures of historically important buildings, including those on the National Register of Historic Places and in the SAH Archipedia (from the Society of Architectural Historians; founded at Harvard in 1940 (if it were started today at Harvard would the focus be entirely on buildings in Gaza?)). A horticulture hobbyist could ask the Tesla to look for unusual trees and label them.

Here’s a photo taken from a car that a Tesla should have been able to figure out was worth keeping:

Garry Winogrand famously used a car as a rolling photo platform in Los Angeles starting in the 1950s. Sadly, he didn’t have AI5 to sort through the pictures and, therefore, died with “2,500 rolls of undeveloped film, 6,500 rolls of developed but not proofed exposures, and about 3,000 rolls only realized as far as contact sheets being made” (times 36 pictures for each roll for those of you who aren’t old enough to remember the film days). Examples highlighting just how realistic Adobe Photoshop’s “add film grain” filters can be:

Most American drives are on ugly and/or boring roads that the owner is forced to use repetitively, e.g., to commute to work? Well, our AI overlords haven’t snarfed up every last SSD (yet) and the retention period doesn’t have to be infinite. If Tesla wants to cheap out on data center resources, it could have the owner configure a cloud service to which the albums would be uploaded whenever the vehicle got into a good WiFi coverage area.

What’s an example of a drive where one wouldn’t expect to want a memory, but something interesting appeared anyway? A friend recently took the trouble to capture a photo of a bumper sticker collection (below) on a highway leading into Boston, which ordinarily has nothing to offer aesthetically.

(Mexican flag because the American flag is a symbol of hate, oppression, slavery, and settler colonialism; University of Michigan directly above because taxpayers in the Islamic Republic of Michigan, such as Abdulrahman Mohamed El-Sayed, should be enthusiastic about subsidizing education for Mexicans; a non-compliant Rainbow flag on the opposite side (no trans-enhancement as displayed on the Biden White House); a Resist Fascism tableau; and an infinity symbol that ChatGPT can’t identify, but Claude says “This is commonly used as a symbol for neurodiversity, particularly autism acceptance.”)

The Tesla could be told to include pictures of any interesting decorations, including bumper stickers and holiday decor. Floridians do amazing things with Jeeps at Halloween and Christmas.

Loosely related… Edward Weston pointed out that “Anything more than 500 yards from the car just isn’t photogenic.” (sometimes the distance is quoted as 100 yards, which is about as far as I would want to carry an 8×10 view camera and the surveyor’s tripod)

Full post, including comments

The hate rally that has no place in Pennsylvania is about to start in Pennsylvania

by philg

Like Greta Thunberg, I don’t like to hear officials mouthing “empty words”. Here’s a recent example of an expression that is all too common in my view: “something bad has no place here”. A governor whose own house was torched by a pro-Hamas Jew-hater says “Hate and antisemitism have no place in our Commonwealth”:

Why are empty words like this problematic? The first step in any 12-step program is admitting that you have a problem. Mx. Shapiro here denies that Pennsylvania and his/her/zir/their fellow Democrats contain any Jew-haters (“hate has no place”). Failure to admit that the problem exists will inevitably lead to a failure to take effective action.

Admittedly, it is unclear what effective action American elected officials could take, other than halting immigration from countries where Jew-hatred is the norm. A negative obsession with the world’s one majority-Jewish country seems to be a path to electoral success in the U.S. now, just as it has been a path to obtaining and holding high office at the United Nations (“António Guterres has posted about Gaza and the Palestinians 262 times since October 7, 2023. Eleven other major crises, combined: 185.” (X, October 26, 2026)). If they were willing to sacrifice some votes, Democrats could simply expel progressives from their party, I guess.

Where is today’s place for hate in Pennsylvania? Rittenhouse Square, Filthadelphia, according to the organizers. It’s not officially a “Jew-hate” rally, but rather hate for “Zionists”, except that the only businesses being boycotted are those owned by Jews (at least one of whom was apparently born and raised in the Philadelphia suburbs). Businesses owned by non-Jews who’ve expressed support for Israel’s existence aren’t being targeted. They’re going to “roll” in 5 minutes, fortunately not in violation of the sacred “I don’t roll on Shabbos” rule.

Update after the confrontation: The Jew-haters of Philadelphia outnumbered the Jew-supporters 300:200, according to the police department (NBC).

Full post, including comments

California has counted 72% of votes so far; still some hope for taxing billionaires

by philg

The bad news is that a majority of California voters so far have rejected a 5% wealth tax on noxious billionaires, the people who make our neighborhoods unsafe, clutter the roads with traffic, and call/text us all the time with various financial scams. (Our Jupiter, Florida neighborhood is afflicted by being close to Michael Jordan’s house in a gated community nearby and I blame most of our problems on him.) The good news, according to the New York Times, is that 28% of the vote has yet to be counted. It is four days after the purported election and there is still plenty of time for the righteous to find enough Yes votes to strike a blow against inequality.

Full post, including comments

The Russell Brand situation is a good argument for Sharia law in the UK?

by philg

I saw headlines about Russell Brand being convicted in the UK of various acts that allegedly occurred between 17 and 22 years ago (2004-2009). Since I don’t usually have the patience to read about 17-year-old encounters between individuals, I asked ChatGPT to summarize it for me. The most serious crime of which he was convicted was rape and ChatGPT’s summary begins as follows:

[The survivor] attended Brand’s comedy performance at London’s Hammersmith Apollo, accepted an invitation for drinks afterward, and participated in a sexual encounter with Brand and another woman in his dressing room. Subsequently, she agreed to visit Brand at his London home and have consensual intercourse.

Then something happened while the two of them were in bed. Although there was no physical evidence or any witness to corroborate any part of the survivor’s story, what added credibility to her dramatic tale, according to ChatGPT, was that Brand texted her an apology in 2009. I asked how a text message had been preserved for 17 years, which seems like a long retention period for a mobile carrier.

(ChatGPT says that a big factor in Brand’s conviction was that he wasn’t consistent in trying to recall events from 2009, e.g., where exactly he was on certain dates, whether he recognized the survivors or not.)

One question for today is why the Islamic Republic of the UK allows women to be victimized in this manner. Under Sharia law, none of the above could have happened. Brand and survivor couldn’t have been drinking alcohol (haram). Brand and the survivor and some other woman couldn’t have been meeting in private in Brand’s dressing room (also haram; both women would have needed escorts from male relatives as well as burqas). The survivor couldn’t have gone to Brand’s house and been alone with him there (haram).

The second question is why Brand didn’t flee. ChatGPT says that he was actually living in Florida in April 2025 and had a valid passport in his hands. Gemini says that Brand has a net worth of approximately $20 million. He could have gone to one of the countries that doesn’t have an extradition treaty with the Islamic Republic of the UK. ChatGPT says that Russia, for example, has no extradition treaty with the UK, bars prosecution for rape after 10 years, and “Russia’s extradition law expressly prohibits extradition when the prosecution is barred by Russia’s statute of limitations.” If Brand had secured Russian citizenship, his extradition would have been barred by Article 61 of the Russian constitution. ChatGPT can’t find a single example of anyone accused of rape ever having been extradited from Russia to the UK. ChatGPT prepared a table of all of the warnings that Brand had:

A summary on X that predicts Brand will be imprisoned for 8-15 years (summarized on X as “Russell Brand will get more time in prison for a “rape” that allegedly occurred 20 years ago than Pakistani men who raped British girls for 20 years. Why does Britain hate its own people so much?”):

Finally, while we’re at it, how about Sharia law for Massachusetts, where people say that they want to (1) protect survivors, and (2) welcome Muslim immigrants. The Whitehead Institute and MIT would have been operating separate labs for men and women. Survivor Kristin Knouse could never have been alone with David Sabatini and, therefore, could never have suffered the victimization that she says she suffered at David Sabatini’s hands (so to speak). Prof. Dr. Knouse, Ph.D. wouldn’t have been able to go to Sabatini’s hotel room in the middle of the night and be alone with him there (“Evidently, we did end up having sex” says Knouse, after reviewing the text message history at her deposition, below (a message history that the Whitehead’s crack legal “independent” investigators didn’t find and never noticed that it was missing, thus revealing inferior electronic forensic skills vs. our 11-year-old)).

Full post, including comments

Could ChatGPT run a money market fund at a lower cost? (and X nerfs its savings account)

by philg

Email from Vanguard with “important information related to one or more of your holdings”:

The annual report of a fund that can invest only in U.S. government short-term instruments certainly would make for fascinating/important reading. What caught my eye, however, was the expense ratio of 0.11%. That’s huge compared to S&P 500 index funds (0.015% to 0.03% expense ratio is typical). Gemini:

These are actively managed portfolios. Fund managers must constantly buy short-term debt securities (like Commercial Paper or Treasury Bills) that mature in days or weeks. Every single day, a massive volume of these securities expires, and managers must actively find and buy new ones to replace them while keeping the fund safe.

If this is truly a big part of the cost, why can’t a robot deal with the $376 billion in assets that Vanguard customers have in this fund?

I asked a friend who managed $billions on and off Wall Street about this. He said, “One reason that stock mutual funds can offer low fees is that they have the opportunity to make money by lending shares to short sellers,” and added, “When I worked at Goldman, our money market was one of the firm’s most profitable activities.” So maybe there is room for an AI-based competitor? A new direction for Binance?

Speaking of money markets, the 6% yield available from Elon Musk has been nerfed. (See Life with X Money (do you get wiped out when the world wakes up to Starship being a total failure?) from just a month ago.) A recent notice on the site says “Effective December 1, 2026, accounts that are eligible to earn 6.00% APY will earn that rate on the first $150,000 of deposits. Any amount greater than $150,000 will earn 4.00% APY.”

(The 4% number beats a Bank of America savings account, for sure, but it falls short of market-leading (NerdWallet has a list).)

Full post, including comments