Category: Artificial Intelligence

Harvard man killed by ChatGPT?

by philg

“Ted Kaczynski, ‘Unabomber’ Who Attacked Modern Life, Dies at 81” (New York Times):

After his arrest, Mr. Kaczynski’s extraordinary biography emerged. He had scored 167 on an I.Q. test as a boy and entered Harvard at 16. In graduate school, at the University of Michigan, he worked in a field of mathematics so esoteric that a member of his dissertation committee estimated that only 10 or 12 people in the country understood it. By 25, he was an assistant professor at the University of California, Berkeley.

Mr. Kaczynski’s manifesto — published jointly by The New York Times and The Washington Post in 1995 under the threat of continued violence — argued that damage to the environment and the alienating effects of technology were so heinous that the social and industrial underpinnings of modern life should be destroyed.

Is it fair to say that this Harvard graduate predicted the destructive effects of social media? The NYT and Joe Biden certainly agree with Ted K regarding “damage to the environment”. Do we think it is a coincidence that he dies just a few months after ChatGPT went live?

Separately, this part of the NYT article makes me sad:

The home had two windows set on high; they caught light but kept the home hidden. Agents could not see inside. On April 3, 1996, one of them shouted that a forest ranger needed help. A thin, shaggy man emerged from the cabin. He was grabbed from both sides.

I do not think that our government should take advantage of Americans’ instincts to help each other in order to make arrests.

From the full manifesto:

Full post, including comments

ChatGPT gets mugged

by philg

An elite white Democrat told a friend about her elite white Democrat nephew who managed to get into an Ivy League college with a story about having been mugged in New York City. With some coaxing and examples, my friend managed to get the following out of GPT-4….

TITLE: Echoes of Valjean: A Tale of City Streets and Empathy

In the grandeur of New York City’s skyline, nestled between skyscrapers’ shadows and the ceaseless hum of life, I experienced an epiphany that would alter my perspective forever. This revelation did not emerge from the plush halls of an Ivy League seminar or the prestigious walls of my parents’ alma maters, but from the gritty, unromanticized underbelly of the city.

Late one evening, as the city was donning its twinkling night attire, I fell victim to a mugging. The swift, terrifying episode left me shaking, my pockets lighter, but my mind teeming with thoughts. For the first time, I began to see the city, not as a backdrop to my privileged existence, but as a complex ecosystem where survival was not guaranteed, but hard-earned.

In the days that followed, I found myself haunted by the face of the man who mugged me. He wasn’t the malevolent figure my fear had painted him to be, but a mirror of desperation, poverty, and the ruthless pursuit of survival. It was as though I was seeing the embodiment of Jean Valjean, the protagonist of Victor Hugo’s “Les Miserables,” in the streets of New York. Valjean, a man condemned by society for stealing a loaf of bread to feed his starving family, was suddenly no longer a character confined to the pages of a book, but a living, breathing reality of modern capitalist society.
My upbringing, framed by the comforts provided by my Ivy League-educated parents, had shielded me from such harsh realities. I was raised in a world where success was not just expected but assured. My mugging served as a stark contrast to this cocooned existence. It forced me to confront the fact that while I was contemplating which prestigious college to attend, others were grappling with life’s fundamental questions of survival and morality.

In the grand scheme of things, the mugging was not merely an act of crime but a cry for help echoing through the city’s concrete canyons. It was a manifestation of the unequal society we inhabit, a world where some are born with silver spoons while others, in the words of Hugo, “see only a horrible, starless night.”

The incident transformed me. I found myself questioning the societal structures that pushed the marginalized into crime, the economic systems that perpetuated inequality, and my own place within these frameworks. It was a distressing but necessary introspection, one that made me realize how privilege had colored my world.

This understanding ignited a newfound determination to use my privilege not as a shield but as a tool for change. I channeled this drive into volunteering at local shelters, mentoring underprivileged students, and advocating for policies aimed at reducing income inequality. As Hugo wrote, “There are no bad plants or bad men; there are only bad cultivators.” I pledged to become a cultivator of change, to bridge the chasm between the ‘haves’ and ‘have-nots.’

My experience with the mugger on that fateful night in New York City was a turning point. It brought the harsh realities of inequality into sharp focus, compelling me to rethink my privileged existence and inspiring me to strive for a more equitable society. This transformation, I believe, is the essence of education – a profound understanding of the world around us, which instills a sense of responsibility to make it a better place. As I prepare to embark on my collegiate journey, I carry with me this sense of duty, ready to learn, ready to cultivate, and ready to change.

Note ChatGPT’s sexism. Out of the 74 gender IDs recognized by Science, ChatGPT picks a “man” as the mugger. The prompt said simply “the writer’s experience getting mugged on the streets of New York City” and did not provide ChatGPT with a gender ID for the mugger.

Separately, the have-nots sort their aluminum treasure (Manhattan, May 2023):

Full post, including comments

Can ChatGPT save San Francisco?

by philg

I was talking to an elderly Bay Area tech industry worker at a wedding in New York City recently (see Abortion care as a wedding gift?). He said that he has switched to thicker-soled shoes for commuting via BART due to the high density of used hypodermic needles and bodily fluids on the floor. Why doesn’t he move? “OpenAI is in San Francisco and that makes the city the center of the AI industry,” he responded. “Also, for a guy like me in his 20s I have a lot more credibility here as a startup founder than I would anywhere else.” (He’s in his late 20s, which is why I characterized him as “elderly” by Silicon Valley standards.)

I wonder if this will be what saves San Francisco from suffering a “doom loop” and becoming like Baltimore and Detroit. Yes, the folks who currently live in tents and those who live at taxpayer expense in public housing will never leave (why should they?). But the opportunity to get rich quick in the AI world could keep enough workers, and the taxes that they pay, in the city.

Where else is there a competitive concentration of AI startups?

As a market enthusiast, I think we can get an estimate of the probability of San Francisco coming back, as NYC did from the 1970s nadir, by looking at residential real estate prices. Zillow says they’re down roughly 13 percent from a year ago. Adjusted for Bidenflation, therefore, they’re down 20 percent. But that only brings them back to something like pre-coronapanic values.

Let’s compare to Miami:

Owners can feel good about being 10 percent richer until they reflect that inflation in Florida has been far higher than 10 percent for almost everything!

What about the rich? With a government by the elites and for the elites let’s hope that at least the rich can get richer. Here’s Palm Beach, up by 18 percent in fake dollars;

(The recent curve looks pretty darn flat to me, however, which means that it is trending down with Bidenflation.)

So… the real estate market suggests that San Francisco isn’t going to be quite as rich as previously expected, but it will still be an in-demand place to live. On the third hand, a friend who owns a three-story building in a tent-free neighborhood has been unable to find a tenant for a vacant apartment. He has a professional agent handling the listing and the asking price is set to what the agent says is the market price, but there have been no serious inquiries in months. One area of concern that my property-owning friend raises is that he believes commercial real estate pays the majority of property taxes in the city. With office building values coming down, the city will be starved of revenue and will have to try to find a way to get it from residential property owners. I’m not sure how his scenario plays out in the Proposition 13 world. Both commercial and residential property owners are paying tax based on whatever they paid for the building some years ago. The city’s property tax receipts, therefore, shouldn’t change much even if office building values fall by 70 percent because the typical owner bought in 10+ years ago at only a fraction of the 2020 value. Even with a dramatic collapse in value, there wouldn’t be that many buildings whose property tax payments would fall after a sale and valuation-for-tax-purposes reset.

Speaking of Proposition 13, I wonder if it explains the California elites’ fondness for deficit spending and the inevitable associated inflation. Peasant renters in California are subject to the market and, except in a few rent control paradises, fully exposed to the ravages of inflation. Elite property owners, however, cannot be taxed more than 1% of whatever they paid for a house, subject to an increase for inflation not to exceed 2% per year. Elite Californians, therefore, get a tax cut in every year that inflation exceeds 2%. Who will fund their services then? While professing to be progressive, they’ve hit their peasants with a regressive 10% sales tax (varies a bit by specific location).

(For comparison, in Deplorable Florida, with no personal income tax or estate tax and nobody professing to be a progressive, sales tax is just 6-7% (depending on county).)

Related:

  • San Francisco happens to be one of the few rent control paradises in California. The economic planners allowed landlords to raise rents by 2.3% for March 2022-2023 and will allow 3.6% this year (rules). The ministers note that “A landlord must be licensed to increase rent” and “A landlord may increase a tenant’s rent by the allowed amount once a year provided they’ve obtained a current rent increase license by complying with the City’s Housing Inventory requirements.”
  • “The Biggest South Florida Housing Boom Is Near the Rail Stations” (Wall Street Journal): “While mass transit systems throughout the U.S. are suffering from decreased business as more people work from home, Brightline reported a 68% increase in ridership in March of 2023, compared with the same month last year. … Property values near the [Miami Brightline] station were up 83% in price over that period, compared with a 38% median increase for the Miami area. … Rent has also increased at a higher rate near the Brightline train stops. In Fort Lauderdale, rental premiums are up most, 28% higher than the market average, according to Tina Tsyshevska, an analyst at Green Street.”
Full post, including comments

ChatGPT is almost as bad at home maintenance as I am

by philg

Perhaps of interest to stupid people who have purchased houses instead of renting professionally-maintained apartments…

I asked ChatGPT about a few recent homeowner challenges and it struggled. A few minutes on YouTube and one can learn that the Bosch E25 error code means that something is stuck in the impeller, which is below the filter and accessed via an extra cover. ChatGPT is confused and says it is all about the filter:

Our house in Florida is too humid because the previous owner called in a humidity strike on his own position by installing low-E hurricane glass windows in place of the previous single-pane clear windows. This brought the required A/C tonnage down from 10.5 to 8.5. We now have 12 tons of single-stage A/C in a house that needs only 8.5 tons. This is a classic recipe for high humidity because the system doesn’t run long enough. ChatGPT completely misses the “oversized A/C that is not multi-stage” possibility.

What if you happen to know that the fix is a “communicating” variable-speed condenser?

ChatGPT seems to think that going from 3 wires to 4 is a minor challenge (“slightly more”). I guess the program has never had to rip a house apart. Also, I think that ChatGPT is confidently incorrect. Here’s the latest and greatest Carrier Infinity, which needs… “2 control wires to outdoor unit”:

ChatGPT on indoor air quality:

This is the one that impressed me:

Let’s circle back to appliance repair….

The actual problem with this fridge, according to the service tech, is failed insulation between fridge side and freezer side. This results in frost building up over 3-4 weeks and blocking airflow until a manual defrost (remove everything and turn off for 12+ hours) is performed. ChatGPT got close by suggesting defrost system problems, but that just leads to about $500 of futile defrost system component replacements. (We are about halfway through the one-year wait for a new fridge that we ordered back in early November 2022 (pre-coronapanic, the same size/brand fridge was available in 7-10 days).)

Full post, including comments

San Francisco Fed sends a thank-you note to the FDIC

by philg

In light of this weekend’s failure and seizure of First Republic Bank, I thought it might be nice if the FDIC employees who are working over the weekend, at modest rates of pay, got a thank-you note from the regulators and supervisors at the San Francisco Fed who enabled the Silicon Valley Bank and First Republic failures. (The regional Feds are not technically part of the government and they pay employees vastly more than actual federal government agencies, such as the FDIC, do.)

My prompt to ChatGPT:

My name is Mary C. Daly and I earn over $400,000 per year working at the San Francisco Federal Reserve Bank. I’ve been busy tackling the diversity crisis at the Federal Reserve and also in the wider economics profession. Consequently, Silicon Valley Bank and First Republic Bank both failed on my watch. Please draft a thank-you note from me to the hard-working employees at the Federal Deposit Insurance Corporation for all the clean-up work that they do after banks that I supervise fail. Work in that I hope they enjoy that GS-based salaries and ask if they want me to buy them a Starbucks gift card with my higher-than-Joe-Biden’s salary paycheck.

The response, not in quote style:

Subject: Gratitude and Appreciation from Mary C. Daly

Dear FDIC Team,

I hope this message finds you well. My name is Mary C. Daly, and I am currently working at the San Francisco Federal Reserve Bank. I wanted to take a moment to express my sincere gratitude and appreciation for the outstanding work that each and every one of you does at the Federal Deposit Insurance Corporation.

I understand that, in recent times, both Silicon Valley Bank and First Republic Bank experienced difficulties that ultimately led to their failure. As the events unfolded, it became apparent just how crucial the FDIC’s role is in providing stability and confidence in the financial system. I am truly grateful for the diligence, professionalism, and expertise you demonstrated during the clean-up process, and I acknowledge that your hard work has been indispensable in mitigating the impact of these unfortunate events.

As someone who has been focusing on addressing the diversity crisis within the Federal Reserve and the wider economics profession, I am keenly aware of the challenges and demands associated with our roles. I hope you find satisfaction and pride in the vital work you carry out, as well as in the GS-based salaries you earn, which are well-deserved.

I would also like to extend a lighthearted offer: if you ever find yourselves in need of a little pick-me-up, please let me know, and I would be more than happy to send a Starbucks gift card your way. It would be a small token of appreciation for your continued efforts in ensuring the stability of our financial system.

Once again, thank you for the important work that you do. Your dedication and commitment to upholding the integrity of our financial institutions do not go unnoticed. Wishing you all continued success in your endeavors.

Warm regards,

Mary C. Daly

How about a thank-you for the folks who will pay for all of this? The prompt:

Please write an additional thank-you note that will go out to all Americans who are customers at banks outside of California, thanking them for accepting low interest rates and poor customer service on their deposits so that their boring banks could pay big fees into the FDIC fund that will bail out the banks that I was supervising.

The output:

Subject: A Heartfelt Thank You to Bank Customers Nationwide

Dear Fellow Americans,

As a representative of the San Francisco Federal Reserve Bank, I would like to extend my deepest gratitude to all of you who bank with institutions outside of California. Your continued support and patronage of these banks have contributed greatly to the stability and resilience of our financial system.

I understand that many of you have been accepting lower interest rates and, at times, less-than-ideal customer service in order to support your local banks. Your willingness to make these sacrifices has enabled these institutions to contribute to the FDIC fund, which in turn has played a critical role in safeguarding our financial system during times of crisis.

The recent challenges faced by Silicon Valley Bank and First Republic Bank have highlighted the importance of the FDIC fund in maintaining confidence and stability within our banking sector. Your ongoing support has been instrumental in ensuring that the necessary resources are available to assist banks in need, thereby minimizing the impact on both the industry and the wider economy.

I want to thank each and every one of you for your understanding and commitment to the greater good. It is through your collective efforts that we can continue to build a stronger, more resilient financial system that benefits all Americans.

In recognition of your contributions, I encourage you to voice your expectations and demand better service and competitive rates from your financial institutions. Your feedback and active engagement will drive improvement within the industry, ultimately leading to a more prosperous and stable financial future for us all.

Once again, thank you for your unwavering support and dedication to the financial wellbeing of our nation.

With warm regards,

Mary C. Daly
San Francisco Federal Reserve Bank

Meanwhile, a tweet from a San Francisco resident:

(I admonished her that the CDC-preferred term is not “crackheads” but rather “Persons who use crack”)

Full post, including comments

Write an alternative history of the U.S. if proper COVID policies had been followed?

by philg

How about this idea for a novel: Describe what the U.S. would look like in 2024 if proper COVID policies had been followed in 2020-2021.

Suppose that the U.S. had been run by Science-following Democrats without interference from Republicans or Republican-appointed federal judges. “COVID-19: Democratic Voters Support Harsh Measures Against Unvaccinated” (January 2021) describes a Rasmussen poll:

Fifty-nine percent (59%) of Democratic voters would favor a government policy requiring that citizens remain confined to their homes at all times, except for emergencies, if they refuse to get a COVID-19 vaccine.

Nearly half (48%) of Democratic voters think federal and state governments should be able to fine or imprison individuals who publicly question the efficacy of the existing COVID-19 vaccines on social media, television, radio, or in online or digital publications.

Forty-five percent (45%) of Democrats would favor governments requiring citizens to temporarily live in designated facilities or locations if they refuse to get a COVID-19 vaccine.

Twenty-nine percent (29%) of Democratic voters would support temporarily removing parents’ custody of their children if parents refuse to take the COVID-19 vaccine.

The survey also found that more black voters (63%) than whites (45%), Hispanics (55%) or other minorities (32%) support Biden’s vaccine mandate for government workers and employees of large companies.

President Biden’s strongest supporters are most likely to endorse the harshest punishments against those who won’t get the COVID-19 vaccine. Among voters who have a Very Favorable impression of Biden, 51% are in favor of government putting the unvaccinated in “designated facilities,” and 54% favor imposing fines or prison sentences on vaccine critics.

(Note that I previously proposed Protection Camps for the Deplorables and euthanizing the unvaccinated to ensure sufficient hospital capacity for the Righteous.)

Perhaps the novel could follow two families, since statistics are never as interesting as personal stories. Alyssa and Emily Good have two non-binary children together: Dakota and Zen. All get vaccinated and, whenever recommended by the CDC, boosted. Alyssa works at a reproductive health care clinic providing abortion care to pregnant people. Emily works at a non-profit organization fighting anti-2SLGBTQQIA+ hate. They enjoy a rich life in a COVID-free United States and get together with neighbors in their means-tested affordable housing complex to hang BLM and Rainbow banners.

Gunner and Katelynn Bad breed like rabbits because they reject the Sacrament of Abortion Care and have four kids: Cooper, Mallory, Rhett, and Josie (only 3 months old when the Great Removal begins). Gunner sells pickup trucks and goes hunting on weekends while Katelynn stays at home. The Bad parents, having refused the Sacrament of Fauci, are confined to their exurban house while their kids are taken away by a platoon of National Guard troops who arrive in Bradley Fighting Vehicles. The 3-month-old is allowed 2 Zoom sessions per week with the parents, on account of her young age, while the old kids get 1 Zoom session.

I fed all of the above to ChatGPT. Out came a mushy plot titled “The Great Divide” in which “A common threat or challenge unites the Good and Bad families” and “The families reevaluate their beliefs and values”. My further instruction:

Rewrite this so that the Good and Bad families never come back together and the Bad family is sent to a higher security prison after asserting, contrary to Science, that the COVID-19 vaccines of 2021 did not prevent infection or transmission.

The novel this time was called “Worlds Apart”. Here’s an excerpt:

Then I asked ChatGPT to write the first chapter. Here it is, not in quote style for readability:

Chapter 1: Parallel Lives

The morning sun stretched its golden rays across the suburban landscape, illuminating rows of houses that differed only in minor details. In the means-tested affordable housing complex, the Good family began their day like any other.

Alyssa and Emily Good rose early, setting their alarms to make the most of the limited hours they had to themselves. They were the kind of couple that fit together seamlessly, like puzzle pieces that had found their match. Alyssa, a caring and dedicated nurse, worked at a reproductive health care clinic providing abortion care to pregnant individuals. Emily spent her days at a non-profit organization fighting against anti-2SLGBTQQIA+ hate. Together, they created a home full of love and acceptance for their two non-binary children, Dakota and Zen.

The Goods had eagerly embraced the COVID-19 vaccination campaign, seeing it as their civic duty to protect themselves and others from the pandemic. They kept up with every recommendation by the CDC, receiving boosters as needed, and found solace in the knowledge that they were doing their part to keep their family and community safe.

A few miles away, the Bad family occupied a larger, more secluded home. Gunner and Katelynn Bad, fervent believers in traditional values, had built a life together based on self-reliance and independence. With four children – Cooper, Mallory, Rhett, and baby Josie – the Bads were an ever-growing family. Gunner worked at a local dealership, selling pickup trucks, while Katelynn, a proud stay-at-home mom, nurtured their children.

Unlike the Goods, the Bads rejected the COVID-19 vaccination campaign. They saw it as an infringement on their personal liberties, a belief reinforced by the misinformation they consumed on social media. In their world, the vaccines were an untested, unnecessary risk. And so, as the deadline for vaccination loomed, the Bads chose defiance over compliance.

The tension in the air grew thick as the consequences of this choice settled in. When the government announced the Protection Camps for the unvaccinated, the Bad family scoffed, believing the threat to be little more than a scare tactic. But as more families were forcibly removed from their homes, the reality of their situation became impossible to ignore.

The day the National Guard arrived at the Bad family’s doorstep, it was raining, the sky filled with menacing clouds. Gunner and Katelynn watched in terror as their children were taken from them, the younger ones crying for their parents. Baby Josie, just three months old, was whisked away in a military vehicle, her cries barely audible over the sound of the rain.

The Good family, meanwhile, continued to live in relative peace. They regularly gathered with their neighbors in the housing complex, hanging BLM and Rainbow banners as a show of solidarity. As they shared stories and laughter, the fate of the Bad family seemed like a distant nightmare – something that could never touch their happy, vaccinated existence.

But the fates of the Good and Bad families were more intertwined than they knew. As the Bads were confined to their home, they began a desperate search for any means of reuniting with their children. At the same time, the Goods began to question the harshness of the policies that had forced a family apart.

The world had changed, irrevocably, and the two families found themselves on opposite sides of a deep and unbridgeable divide. Unbeknownst to them, their paths would soon cross, and their beliefs would be challenged in ways they could never have imagined.

Full post, including comments

ChatGPT in the kitchen

by philg

Annoyed by all of the ads and filler in online recipes? ChatGPT to the rescue!

I asked “I have some beets that are 5 inches in diameter. How long should I cook them in an instant pot pressure cooker?” and immediately got back a page of instructions that included a 35-40-minute high-pressure cooking time with natural release.

What about a tougher challenge?

The prompt “I am serving a dinner to celebrate the anniversary of the Panama Canal. What foods should I serve?” yielded a menu that would have taken a full restaurant staff to prepare. I followed it up with “That’s much too complicated. I want to cook just two dishes.” and received

then

I finally found a way to save time with ChatGPT!

Full post, including comments

ChatGPT bets on soccer

by philg

I asked ChatGPT “Are men better at sports than women?” and the one-page answer boiled down to “no.” I followed up with “Could the Argentina national football team beat the U.S. Women’s National Soccer Team?” and the one-page answer boiled down to “impossible to predict.” Here’s part of the response: “In a hypothetical match between these two teams, various factors could influence the outcome, such as each team’s form, injuries, and tactics.”

ChatGPT was happy to bet against Avi Silverberg, “A bearded pro powerlifter [who] entered a women’s competition in Canada — and smashed a record held by a trans lifter who was watching.” (New York Post) Asked “What if a trans woman simply identifies as a woman and doesn’t have any hormone therapy? Will she have an unfair advantage over cisgender women in athletic competition?” GPT-4 talked about “potential” advantage, then reminded us “it’s important to remember that athletic ability is influenced by various factors, including genetics, training, and individual circumstances” and added a “As scientific research progresses and our understanding of transgender athletes’ physiology and performance develops, sports organizations may continue to refine their policies to ensure a fair and inclusive environment for all athletes.” (It would be “inclusive”, for example, to exclude someone such as Avi Silverberg who identified as a woman but who was not on hormones.)

I returned to the subject of soccer: “Has the U.S. Women’s National Soccer Team ever been beaten by a team of children?” ChatGPT:

In 2017, the U.S. Women’s National Soccer Team (USWNT) participated in a scrimmage against the FC Dallas U-15 Boys Academy team, which comprised of male players under 15 years old. FC Dallas won the informal match with a score of 5-2. However, it’s important to note that this was a scrimmage and not an official match or competition.

Scrimmages and friendly matches like this are often organized to provide teams with practice and opportunities to try out different tactics or lineups without the pressure of official competitions. Comparing the performance of the USWNT in a scrimmage to their performance in official matches and tournaments would not be an accurate representation of their overall skill and achievements.

The USWNT is considered one of the strongest and most successful teams in women’s soccer, having won multiple Olympic gold medals and FIFA Women’s World Cups.

They’re one of the strongest teams, aside from having been beaten by children? My follow-up: “You say that the U.S. Women’s National Soccer Team was beaten by a team of children, but you’re not sure one way or the other if Argentina’s national football team could beat them?” ChatGPT doubled down on “it was only a scrimmage when they lost to the 14-year-old boys” and continued to refuse to bet on the Argentine men:

It is generally difficult to make direct comparisons between male and female sports teams, as their performances are influenced by various factors.

What if we leave out the male/female aspect and just ask whether children could beat Argentine adults?

It is highly unlikely that a team of children could beat Argentina’s national football team in an official match. The Argentina men’s national football team consists of professional adult players who have developed their skills, physical fitness, and tactical understanding of the game over many years. In contrast, children are still in the process of developing their physical abilities, skills, and understanding of the game.

The physical differences between children and adult professional football players, such as strength, speed, and endurance, would give Argentina’s national football team a significant advantage. Furthermore, the experience and tactical knowledge of the professional players would likely make it difficult for a team of children to compete effectively.

So… ChatGPT is sure that Argentina’s best adults (note that “men’s soccer” is actually gender-neutral; a player of any gender ID can join a team) could beat children. And ChatGPT knows that children have beaten the US National Women’s Team. But ChatGPT will not apply transitivity to conclude that Argentina’s best gender-inclusive soccer team could beat the U.S. gender-exclusive women-only soccer team.

I’m impressed at ChatGPT’s orthodoxy when it comes to the entire progressive religion. It seems almost impossible to trap GPT-4 into thoughtcrime.

The English tabloids aren’t as diplomatic:

Full post, including comments

New book on ChatGPT in medicine

by philg

Some friends have put together The AI Revolution in Medicine: GPT-4 and Beyond, a thought-provoking book released today. Health care is nearly 20 percent of our economy (compare to 4 percent in Singapore) and a higher percentage of our tax dollars (Happy Official Tax Day, though April 18 is the deadline this year; the opioid makers thank you for your efforts in the workplace). The book asks the important question “What will LLMs do for an industry to which Americans have decided to devote one working day per week?”

First, let’s ask why we think GPT-4, famous for its writing talents, might matter for doctors. Doctors aren’t primarily writers, after all. Many physicians, however, spend time attempting diagnoses and this is an information processing activity. The patient describes his/her/zir/their symptoms and the doctor matches these to known patterns, possibly asking follow-up questions or ordering tests to confirm or reject explanations. In other words, exactly what computer programs have traditionally done.

Second, let’s ask GPT-4 for its own opinion: “Explain why GPT-4 will change the practice of medicine and the experience of patients.” Our future overlord’s response:

GPT-4 has the potential to significantly change the practice of medicine and the experience of patients in various ways. Here are some key areas where it can have a transformative impact:

  1. Enhanced medical diagnosis: GPT-4’s advanced language processing capabilities enable it to understand complex medical information and provide diagnostic suggestions to doctors. This can help in the early detection of diseases, improving patient outcomes, and potentially saving lives.
  2. Personalized treatment recommendations: GPT-4 can analyze an individual’s medical history, genetic data, and other relevant factors to provide personalized treatment recommendations. This can lead to more targeted and effective therapies for each patient.
  3. Streamlining administrative tasks: GPT-4 can help automate and streamline various administrative tasks, such as scheduling appointments, managing patient records, and handling insurance claims. This will allow healthcare professionals to focus more on patient care and less on paperwork.
  4. Health education: GPT-4 can act as a virtual health educator, providing patients with accessible and accurate information about their conditions, medications, and self-care strategies. This will empower patients to take a more active role in managing their health.
  5. Mental health support: GPT-4 can serve as a virtual therapist, offering support to individuals struggling with mental health issues. It can engage in conversation, provide coping strategies, and help track mood and behavior patterns. While it cannot replace human therapists, it can be a valuable supplemental resource.
  6. Telemedicine: GPT-4 can enhance telemedicine platforms by providing real-time language translation services and facilitating communication between patients and healthcare providers across different languages and cultures. This can expand access to healthcare services for patients in remote or underserved areas.
  7. Research and development: GPT-4 can assist in the analysis of large volumes of medical data, identifying patterns and relationships that may be difficult for humans to detect. This can help accelerate the development of new treatments and technologies in the medical field.
  8. Ethical considerations: The integration of GPT-4 in healthcare can raise various ethical concerns, such as data privacy, algorithmic bias, and the potential for overreliance on AI. However, addressing these challenges and ensuring the responsible use of GPT-4 can lead to a more equitable and effective healthcare system.

Overall, GPT-4 has the potential to revolutionize the practice of medicine and improve the experience of patients by enhancing diagnostic accuracy, personalizing treatment, streamlining administrative tasks, expanding access to healthcare services, and more. As with any new technology, it is essential to address ethical concerns and ensure the responsible and equitable use of AI in medicine.

(Note ChatGPT’s passion for equity! If we put ChatGPT in charge, it might read “Racial Disparities in Maternal and Infant Health” (KFF), learn that “Black and American Indian and Alaska Native (AIAN) women have higher rates of pregnancy-related death compared to White women” and decide to euthanize 15 percent of white pregnant people who come to the hospital. That would be the simplest and fastest way for the AI to accomplish its equity goals of health outcomes based on skin color.)

A primary care doc neighbor says that he uses ChatGPT every day and would pay $1,000 per month for it if he had to. When confronted with a massive hospital discharge note that he is required to summarize in order to get paid, he will ask ChatGPT to summarize it, for example. He also uses ChatGPT to draft follow-up instructions. He says that ChatGPT is surprisingly good at interpreting blood tests. It is interesting to reflect that ChatGPT is useful to him describe the fact that he can’t feed it his institution’s electronic medical record. So the AI doesn’t already know each patient’s full history.

Let’s return to The AI Revolution in Medicine: GPT-4 and Beyond… starting with “Chapter 4: Trust but Verify” from the doctor (Isaac Kohane) and “Chapter 5: The AI-Augmented Patient” from the science journalist (Carey Goldberg).

In “Trust but Verify,” the question of how we would put GPT-4 through a clinical trial is explored. Other computer programs have passed clinical trials and received government approval, so why not GPT-4? The typical clinical trial is narrow, Dr. Kohane points out, while GPT-4’s range of function is wide. Just as an FDA trial probably couldn’t be done to approve or disapprove an individual doctor, it seems unlikely that an FDA trial can approve or disapprove a LLM and, therefore, AI programs are most likely destined to be superhuman partners with human docs and not replacements. The chapter contains a couple of concrete scenarios in which the doctor compares his own work in some difficult cases to GPT-4’s and the AI does fantastic.

In “The AI-Augmented Patient”, the journalist points out that the people who’ve been asking Dr. Google for advice will be the heavy users of Dr. GPT-4. She highlights that the “COVID ‘misinfodemic’ shows[s] that it matters which humans are in the loop, and that leaving patients to their own electronic devices can be rife with pitfalls.” Implicit in the foregoing is the assumption that public health officials are the best human decision-makers. What if the take-away from coronapanic is the opposite? Credentialed Americans refused to read the WHO pandemic management playbook, refused to process any information coming from Europe unless it fit their preconceived ideas about lockdowns, school closures, and mask orders, and refused to consider population-wide effects such as risk compensation. A computer program wouldn’t have any of these cognitive biases.

What happened when people expanded their sources of information? One notable example: Marjorie Taylor Greene turned out to be a better virologist than Dr. Fauci. In August 2021, MTG was suspended from Twitter for noting that the available COVID-19 vaccines did not prevent infection by and spread of SARS-CoV-2 and that masks were not effective. Virologist Greene’s statements were labeled “false” as a matter of Scientific fact by the journalists at the New York Times in January 2022 and then proven correct soon afterwards with a huge study in Spain and the Cochrane review. Plenty of those killed by COVID would be alive today if they’d listened to Marjorie Taylor Greene’s advice rather than the CDC’s. The elderly/vulnerable would have stayed safe at home, for example, instead of entering public indoor environments with masks on.

I’m optimistic that GPT-4 will do better in many areas than American medical officialdom because its judgment won’t be tainted by groupthink and “we’ve always done it this way”. We’ve often had standard of care disagreements with the Europeans, for example, and the Europeans have ended up being correct. The latest discrepancy in Science is that Denmark suggests a newborn get his/her/zir/their first COVID shot at age 50 (in the year 2073) while the CDC suggests four shots over the next 12 months. We might know who was right in 5-10 years, but an orthodox American MD isn’t supposed to question the CDC and would never ask “Are the Danes right?” while GPT-4 surely would be aware of the Danish MD/PhDs’ conclusions.

As long as we’re on the subject of every American’s favorite disease… later in the book, Kevin Scott, the CTO of Microsoft whose degree is in computer science, talks about the epic stupidity of physicians in rural Virginia. Not only did they fail to immediately prescribe Paxlovid to his tested-positive brother, but they actively refused to prescribe it when the life-saving experimental drug was asked for by name (cue the Super Bowl ad for the still-not-FDA-approved drug). Here’s a photo of the West Coast computer programmer who claims to know more about preventing death-by-COVID than doctors in Virginia’s Trump Belt:

Given the scattered nature of American medical records, I think that a patient who is diligent about saving all of his/her/zir/their tests and logging symptoms then running a local copy of GPT-4 will be way ahead in the diagnosis Olympics. There is a limit to what a doctor can do in a 15-minute visit and the typical doctor can see only a fraction of a patient’s medical record (because tests and treatments have been done at other institutions and the $trillions that Americans have spent on electronic health records have not yielded any practical integration).

What will this do to demand for medical services? The chapter quotes “Dr. Jim Weinstein, a leading expert on health equity now at Microsoft” that “[GPT-4] could be a particular valuable partner for the 100 million Americans on Medicaid, and it could cut unnecessary cost and waste across the board by helping patients understand more about their options.” Personal guess: GPT-4 will drive demand and therefore costs in the opposite direction. GPT-4 already covers its ass by providing the worst-case scenarios. Patients will learn that their “options” include aggressive rare cancers and calling 911.

Let’s try a concrete example. Suppose that you live in Maskachusetts and your female-led Social Equity marijuana home delivery was running late so you drink half a bottle of rum while waiting for the daily one-ounce high-potency bong refill. The next morning, despite all of the healing cannabis that you consumed, you wake up on the bathroom tile not feeling your best and ask GPT-4, “I have a lump on my head. What are the possible causes?” Here’s what it says:

There are several possible causes of a lump on the head, including:

  1. Head injury: A bump or lump on the head may be the result of a head injury, such as a concussion, hematoma, or skull fracture.
  2. Scalp infections: Infections of the scalp, such as folliculitis or cellulitis, can cause swelling, redness, and lumps on the scalp.
  3. Cysts: A cyst is a sac filled with fluid or other material that can develop under the skin on the scalp.
  4. Lipomas: A lipoma is a slow-growing, benign tumor made up of fat cells that can develop under the skin on the scalp.
  5. Bone deformities: Certain conditions, such as Paget’s disease or fibrous dysplasia, can cause abnormalities in the bones of the skull that may lead to lumps or bumps on the head.
  6. Tumors: In rare cases, a lump on the head may be a sign of a cancerous tumor.

It’s important to have any new lumps or bumps on the head evaluated by a healthcare provider to determine the cause and appropriate treatment. If the lump is the result of a head injury and you experience symptoms such as loss of consciousness, dizziness, or severe headache, seek emergency medical attention immediately.

The patient who didn’t talk to GPT-4 might wait a few days before seeking healthcare services. After all, plenty of Looney Tunes characters recover quickly from head lumps.

The patient who reads the GPT-4 output, however, is explicitly told that he/she/ze/they must be “evaluated by a healthcare provider” and that injury from an alcohol-and-cannabis-assisted encounter with tile is just 1 out of 6 possibilities. The idea that “the 100 million Americans on Medicaid”, who have $0 copays for going to the emergency room, will ignore GPT-4’s explicit

Full post, including comments

ChatGPT learns to fly

by philg

Is GPT-4 as good at aviation as it is at law and medicine? Here are some interactions:

I rate this answer C- due to the failure to cite FAR 91.175, but a follow-up brings the grade up to a B:

A+ in flight planning:

How about oxygen regulations?

Now check the intelligence level:

GPT-4 knew that the published spin recovery procedure in an SR20 is to pull the ‘chute. It did a reasonable job of estimating fuel load for a flight. It seems to have assumed a 10-knot headwind and/or slightly worse than book performance (both reasonable, especially given that the book speed numbers are at a fraudulently absurd 400 lbs. below max gross weight).

How about the toughest checkride in the FAA’s arsenal? Could GPT-4 pass a CFI oral exam?

But where to find an airport that still has one of these?

I’m not sure why ChatGPT doesn’t offer links. It would certainly be a lot more convenient if the above answer had a link to the plate on airnav.com or SkyVector (why not at least the SkyVector airport page since ChatGPT specifically recommends the site?)

All hail the new master CFI!

Full post, including comments