Category: Artificial Intelligence

ChatGPT prevents users from WrongThink

by philg

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.

ProgramApproximate % of Population (Latest Data)Monthly Enrollees (Millions)Notes
Medicaid22% (2023)80 (peaked; ~72 in 2024)Largest program; covers low-income families, children, disabled adults. Overlaps heavily with SNAP (80%).
SNAP12.3% (FY 2024)41.7Food 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.
Other (TANF, WIC, SSI)2-5% each2-8TANF (cash aid): <1%; WIC (nutrition for pregnant women/children): 6 million.
At Least One (Unique)17.7-25% (2019-2024 est.)60-80Overlaps 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!

Full post, including comments

Ample qualifications to run a $100 million AI startup

by philg

At the intersection of the OnlyFans and the AI Bubble, “Eric Schmidt’s ex-mistress, 31, sues former Google CEO, 70, over alleged stalking, abuse and ‘digital surveillance’” (New York Post):

The 31-year-old former mistress of Eric Schmidt has accused the ex-Google CEO of stalking, abuse and “toxic masculinity” — claiming that he subjected her to an “absolute digital surveillance system” as the pair have secretly tussled over cash, a failed AI startup and access to a sprawling Bel Air mansion, The Post has learned.

In early December, Ritter and Schmidt — whose net worth is estimated by Bloomberg at $44.8 billion — struck a “written settlement agreement” that required Schmidt to make “substantial payments” to Ritter but whose details remain under seal, according to a Sept. 8 filing in Los Angeles County Superior Court.

In the since-withdrawn TRO request, Ritter claimed that the tech tycoon days earlier had locked her out of the website of her startup Steel Perlot — an AI-focused venture firm into which Schmidt had plowed $100 million, a source close to the situation told The Post.

The article suggests that “a German Shepherd named Henry” is a disgrace to his breed in terms of providing protection. Despite owning this powerful beast, the plaintiff says that she’s been a victim of sexual assault at the hands of a senior citizen who is 39 years older than herself. Young lithe actresses and directors, such as Gavin Newsom’s current wife, couldn’t escape from elderly obese Harvey Weinstein, but none of them had a German Shepherd to assist them.

Related:

Full post, including comments

Will AI make conscientiousness and organization more or less valuable?

by philg

A human’s productivity is typically determined, to a large extent, by intelligence and conscientiousness. These are both heritable traits so there is a limit to how smart and conscientious a person is likely to become if his/her/zir/their biological parents weren’t smart and conscientious.

As we celebrate National Coffee Day today, I’m wondering if the conscientious aspect of productivity will be rendered more or less relevant by artificial intelligence and robotics. Consider a person prone to disorganization and procrastination, both behaviors negatively correlated with conscientiousness. Suppose that each of us is being followed everywhere by a humanoid robot. At any given moment, the robot reminds us what needs to be done. Even the spaciest among us will never space out and miss a videoconference because the robot will log us into it.

The flip side of this argument is that AI is a productivity amplifier and, therefore, the people who are currently unproductive will stay unproductive (100 times 0 is still 0) while the productive will become superheroes of output. Maybe a person with mediocre conscientiousness will be rendered more conscientious by the companion robot, but that person will still be left in the dust by the conscientious who’ve gotten even more of a boost from their companion robots.

Related:

  • “Heritability of the big five personality dimensions and their facets: a twin study” (classic 1996 paper finding 44% heritability for conscientiousness)
  • Wokipedia forced to admit that we’re not all born equal when it comes to IQ (but remember that in the Wokipedia world there is no correlation between race and IQ, only “high heritability of intelligence within races”)
  • average IQ in the US is declining (coinciding with soaring immigration from societies with low average IQ), thus making conscientiousness more important: academic paper (2023) from Intelligence (“A reverse Flynn effect was found for composite ability scores with large US adult sample from 2006 to 2018 and 2011 to 2018. Domain scores of matrix reasoning, letter and number series, verbal reasoning showed evidence of declining scores.”)
Full post, including comments

Current stock market valuations explained

by philg

From Pedro Domingos, a CS prof at University of Washington, the best current explanation for stratospheric stock market valuations:

Oracle’s main business these days is promising vast amounts of cloud computing it doesn’t have to AI companies who don’t know how they’ll pay for it.

WSJ, a month ago:

The S&P 500 currently trades at 22.5 times its projected earnings over the next 12 months, compared with the average of 16.8 times since 2000. … The 10 largest companies in the S&P 500 accounted for 39.5% of its total value at the end of July, the most ever…

How badly beaten up did investors who bought into stocks at a high P/E ratio get? I asked Grok “Consider an investor who purchased the S&P 500 in February 2000. What annual return on investment would he or she have received through August 2025 vs. an investor who bought in August 2002 and held through August 2025?” and learned that the “Peak P/E ratio” investor (bought before the dotcom bubble burst) would have earned a compound annual growth rate (CAGR) of about 6.3% vs 8.9% for an investor who bought at a more reasonable P/E ratio in August 2002. This difference is close to the difference between investing from 2002-2025 in wired U.S. (9%) vs. tired Europe (5.7%).

Full post, including comments

Robot tugboats to repel Greta Thunberg’s selfie flotilla and similar?

by philg

Climate change no longer being an issue, apparently, Greta Thunberg and friends have spent the past month headed for a visit to the noble Gazans via diesel-powered flotilla (“Global Sumud Flotilla”). It’s a strange situation because the flotilla participants seem to have no respect for Palestinian religious and cultural norms, e.g., they permit females to roam the decks of their yachts without wearing hijab. The photo below could perhaps be a prostitute with two clients by Gazan standards:

Israel has established a blockade against its military enemy in Gaza and, therefore, under international law can repel (or sink, if necessary?) any ship that crosses the dashed red line below.

But Israel gets a lot of bad press when it uses standard military procedures. What about robot tugboats that could intercept the selfie yachts and push them out of blockade area? That would deny Greta Thunberg and other diesel-powered climate activists the photo opportunities that they seek. The flotilla folks claim that to be unarmed so they don’t have any means of destroying robot tugs. Robot tugs could be built in different sizes to match up to the different size yachts in the flotilla. They’re intercepting uncooperative vessels so should have a higher top speed than conventional tugboats. The Israelis could start with simple skiffs equipped with modern outboard engines. Since the skiff doesn’t have to hold any cargo or humans it could hold a tremendous amount of fuel for endurance. Surround the skiff with used car tires so as to get some extra points for recycling.

For maximum reliability with minimum fuel consumption and pollution, the Israelis could use Honda’s only V8 engine (350 hp; Mercury makes one with 600 hp if necessary):

Loosely related, Israel has invited the hostile vessels “to dock at the Ashkelon Marina and unload the aid there”, which would certainly be anticlimactic compared to a climate activists-v-robot interaction!

As of September 24, 2025, the yachts had survived 14 attacks by warships and warplanes, without sustaining any damage, and were using their inoperative radios to report a “communications jam” that has rendered their radios inoperative:

See also “Posing with Hamas chief, activist who’s joined Greta on Gaza ‘freedom flotilla'” (Daily Mail):

Grinning as he gives a Churchillian ‘victory’ sign, the spokesman for Greta Thunberg’s Gaza ‘freedom flotilla’ poses with a Hamas chief weeks before setting sail.

Wael Nawar was draped in a scarf emblazoned with the terror group’s emblem as he stood with other pro-Palestinian activists beside Youssef Hamdan, who runs Hamas’s North African operation, during a meeting at its Algerian headquarters in June.

Mr Nawar is listed alongside Swedish campaigner Ms Thunberg as part of the 13-strong ‘steering committee’ orchestrating the Global Sumud Flotilla, which left Spain last week to deliver food and medical aid to Gaza.

Another committee member, political activist Marouan Ben Guettaia, was also a guest of Hamdan a few days after and later posted a picture of the pair sitting in front of a Hamas flag.

A third committee member, Brazilian Thiago Avila, attended Hezbollah leader Hassan Nasrallah’s funeral in Beirut in February and praised him as an ‘inspiration’.

Full post, including comments

Defend a house against woodpeckers using robot drones?

by philg

Homeowners around the world suffer a lot of damage due to woodpeckers. How about a system of microphones around the house that listen for the sound of a woodpecker and, if heard, dispatches a drone that lives somewhere on the edge of the exterior, maybe under an eave? The drone will then use its own microphone and camera to locate the woodpecker and harass it, with a water pistol if necessary, until the woodpecker finds a tree or an unprotected home to destroy.

ChatGPT refused to draw a picture of a drone discouraging a woodpecker with harmless water: “I can’t create an image that depicts harm being done to an animal — including a woodpecker being shot with water. … Instead of water hitting the bird, the drone could be shown with a water spray or mist aimed at the trim (not at the bird), to illustrate the concept of “protecting the house” without showing harm to the animal.” It then proceeded to generate an image that looks to me like the poor bird is being blasted with water:

(The Gaza Health Ministry reports that more than 60,000 woodpeckers have been killed via water pistol.)

Grok didn’t comment on my desire to see a photo of violence being done to a woodpecker, but it decided that the stream of water should emerge from the woodpecker:

Full post, including comments

Send humanoid robots to do crazy stunts?

by philg

Today is the 51st anniversary of when the third greatest American tried to jump over the Snake River Canyon, near Twin Falls, Idaho, in a steam-powered motorcycle. I was there back in June.

From the downtown Perrine Bridge, a mound of dirt remains visible:

There is a monument to the third greatest American right at the bridge/visitor center:

A short drive to the east, the mound itself may be examined and there is additional signage:

Evel Knievel’s particular jump was already replicated in 2016 by Eddie Braun:

If Eddie Braun hadn’t recreated this jump, wouldn’t it be awesome to see Tesla’s Optimus, or a similar humanoid robot, piloting a replica Skycycle X-2?

I’m wondering what other stunts could be pulled by humanoid robots. Motorcycle jumps, obviously, but what else would be fun to watch? Maybe after a humanoid robot proves that something can be done a human can follow in his/her/zir/their tracks (don’t want to assume a gender ID for a robot that thinks fast enough to change gender multiple times per second).

(Readers might be wondering who the first and second greatest Americans are. Elvis Presley, of course, is in the #2 slot. Due to Democrats being in majority in the U.S., we must recognize George Floyd as the #1 greatest American ever to have lived. Educate yourself by reading Floyd’s biography, recommended by state-sponsored NPR, if you aren’t familiar with all of George Floyd’s achievements.)

Full post, including comments

Years until NVIDIA’s revenue exceeds U.S. GDP

by philg

Happy Labor Day to those not smart enough to refrain from working!

Soon all of our dreary poetry writing, creative image and video production, etc. will be accomplished by LLMs. Freed from this labor we will be able to spend all of our time cleaning the kitchen, unclogging drains, fixing roofs in 90-degree Florida heat/humidity, etc.

Let’s check in with the machines that will be doing all of our laboring going forward.

NVIDIA recently released some financials:

My question for NVIDIA chips, via ChatGPT:

Extrapolating from revenue trends of the last five years, how long will it be before Nvidia’s annual revenue exceeds the United States GDP?

The answer:

Short answer: about 2036—but only under a wildly unrealistic straight-line extrapolation.

Here’s the quick math I used:

  • Nvidia’s last five fiscal-year revenues (billions): 2021 $16.7, 2022 $26.9, 2023 $27.0, 2024 $60.9, 2025 $130.5. That implies a ~67% CAGR from FY2021→FY2025. Macrotrends
  • U.S. nominal GDP: about $29.2T for calendar year 2024 (BEA Table 1.1.5); latest quarterly SAAR (Q2 2025) is $30.35T. FRED+1

Assuming Nvidia keeps compounding revenue at ~65–70% annually and U.S. GDP grows ~4% nominal per year, Nvidia’s annual revenue would mathematically catch U.S. GDP in roughly 11.4 years—i.e., mid-2036. With GDP held flat at today’s level, the crossing is ~10.5 years. (This requires ~224× growth from FY2025 revenue.)

Sensitivity (still using simple exponential curves):

  • If Nvidia grows 40%/yr → cross in ~18 years (≈2043).
  • 30%/yr → ~24 years (≈2049).
  • 20%/yr → ~38 years (≈2063).

⚠️ Reality check: this comparison is apples-to-oranges (company revenue vs a country’s value-added), and no firm can sustain ~60–70% annual growth for a decade+ in a competitive, capacity-constrained industry. Treat the dates above as a fun extrapolation, not a forecast.

How did Grok do on this question? Terrible. Its initial interpretation of the question was that it was about this year’s U.S. GDP, i.e., that it should hold GDP constant. When corrected, it chose an absurd 8.5 annual growth rate for U.S. GDP (would make sense only if there is 6 percent inflation; we’re lucky if we can eke out 2 percent real GDP growth even after flooding the U.S. with migrants; per-capita real annual GDP growth in the U.S. has been about 1.3 percent per year during the past 25 years (FRED)).

Loosely related:

Full post, including comments

Our AI overlords will need more power than the Sun generates within 43 years

by philg

My question to Grok: “Extrapolating from the past three years how long before running LLMs consumes more power than the entire Sun generates?”

Extrapolating from the growth in global AI data center power demand over the past three years (2022–2025), which serves as a proxy for LLM power consumption given that LLMs drive much of the recent AI surge, we can estimate the timeline. Data from SemiAnalysis (as cited in a RAND report) shows the following facility power demand for AI data centers:

  • 2022: 1.451 GW
  • 2023: 4.28 GW
  • 2024: 10.6 GW
  • 2025: 19.9 GW

To arrive at the extrapolation:

  1. Fit an exponential growth model to these data points. Using linear regression on the natural log of power vs. year (with 2022 as t=0), the slope is approximately 0.876, corresponding to an annual growth factor of e^0.876 ≈ 2.401 (i.e., power multiplies by about 2.4 each year on average).
  2. The Sun generates 3.826 × 10^26 watts of power, equivalent to 3.826 × 10^17 GW.
  3. Starting from 19.9 GW in 2025, the factor needed to exceed the Sun’s output is (3.826 × 10^17) / 19.9 ≈ 1.924 × 10^16.
  4. Solve for years n: 2.401^n = 1.924 × 10^16, so n = log(1.924 × 10^16) / log(2.401) ≈ 43.

Thus, it would take approximately 43 years from 2025 (i.e., by around 2068). This assumes the exponential growth rate from the past three years continues indefinitely, though in reality, physical, technological, and economic constraints would likely slow or halt such growth long before reaching solar scales.

Note that ChatGPT 5 failed to answer this question, identically worded. It used data center power consumption growth, not LLM- or AI-specific power consumption growth.

When does the exponential growth stop? We can power a human brain with an Egg McMuffin. Is the answer an analog chip with a digital interface? IEEE has a 2024 article on a Chinese design using optical computing that supposedly cuts power consumption by 99%. But if these improvements were real would the AI companies be signing 20 year leases for nuclear power?

Full post, including comments

Humanoid robots to paint giant murals?

by philg

The Murakami show at the Cleveland Museum of Art includes some murals that would be awesome to have in a kid’s room if only a humanoid robot could be adapted to do the work of either applying wallpaper or directly painting.

Another area where the robot could work… recreating Sol LeWitt murals in the home. Different color schemes for every holiday.

Note that the museum’s permanent collection is free, even to those who have jobs (see How to get free museum admissions for life: sign up for food stamps (SNAP/EBT)). Separately, a fair number of visitors were #Scientifically masked:

For my friends in health care, the artist’s conception of what a nurse looks like:

Circling back to the principal theme for today… if you had nearly-free high-skill labor from a robot would you use some of it to have wall murals in your house? Or would it make more sense to cover a wall in large tiles of flat-screen TVs and do this electronically?

Full post, including comments