AI has hallucinations, it is mathematically inevitable it will make false statements

[Maol]

https://techxplore.com/news/2025-09-ope ... atgpt.html

OpenAI's latest research paper https://openai.com/index/why-language-m ... llucinate/

diagnoses exactly why ChatGPT and other large language models can make things up—known in the world of artificial intelligence as "hallucination." It also reveals why the problem may be unfixable, at least as far as consumers are concerned.

The paper provides the most rigorous mathematical explanation yet for why these models confidently state falsehoods. It demonstrates that these aren't just an unfortunate side effect of the way that AIs are currently trained, but are mathematically inevitable.

[Maol]

ChatGPT appears to improvise when put through ancient Greek math puzzle

https://phys.org/news/2025-09-chatgpt-a ... uzzle.html

The Artificial Intelligence chatbot, ChatGPT, appeared to improvise ideas and make mistakes like a student in a study that rebooted a 2,400-year-old mathematical challenge.

[Maol]

Nothing like some "But Wait, There's More" when it's getting worse instead of better.

https://www.newscientist.com/article/24 ... e-to-stay/

AI hallucinations are getting worse – and they're here to stay

An AI leaderboard suggests the newest reasoning models used in chatbots are producing less accurate results because of higher hallucination rates. Experts say the problem is bigger than that

By Jeremy Hsu
9 May 2025

AI chatbots from tech companies such as OpenAI and Google have been getting so-called reasoning upgrades over the past months – ideally to make them better at giving us answers we can trust, but recent testing suggests they are sometimes doing worse than previous models. The errors made by chatbots, known as “hallucinations”, have been a problem from the start, and it is becoming clear we may never get rid of them.

Hallucination is a blanket term for certain kinds of mistakes made by the large language models (LLMs) that power systems like OpenAI’s ChatGPT or Google’s Gemini. It is best known as a description of the way they sometimes present false information as true. But it can also refer to an AI-generated answer that is factually accurate, but not actually relevant to the question it was asked, or fails to follow instructions in some other way.

Read more
How to avoid being fooled by AI-generated misinformation
https://www.newscientist.com/article/24 ... formation/

[Aardwolf]

Calling them “intelligent” is a con for funding. They should only ever be used for entertainment.

[Maol]

Read this and weep for mankind.

https://archive.is/wAcD3

A,I. Chatbots Can Go Into a Delusional Spiral. Here’s How It Happens. - The New York Times

Over 21 days of talking with ChatGPT, an otherwise perfectly sane man became convinced that he was a real-life superhero. We analyzed the conversation.

For three weeks in May, the fate of the world rested on the shoulders of a corporate recruiter on the outskirts of Toronto. Allan Brooks, 47, had discovered a novel mathematical formula, one that could take down the internet and power inventions like a force-field vest and a levitation beam.

Or so he believed.

Mr. Brooks, who had no history of mental illness, embraced this fantastical scenario during conversations with ChatGPT that spanned 300 hours over 21 days. He is one of a growing number of people who are having persuasive, delusional conversations with generative A.I. chatbots that have led to institutionalization, divorce and death.

[Maol]

Just a gentle reminder.

https://archive.is/wAcD3

[Brigit]

Thank you, "delusional spiral" is so much more accurate than "hallucinations," for the reasons given!

The channel Startup Hakk is informative.

So here is a really good analysis of the MIT "paper" that asserted a future scenario of agentenic programs.

• AI 2027 Was Just Science Fiction — Even Its Creators Admit It
  https://www.youtube.com/watch?v=flz4A9qm6NA
  CH: STARTUP HAKK
  
  Remember AI 2027? That viral scenario that predicted superintelligent AI would take over by 2027 and potentially end humanity?
  The one that got referenced by Vice President Vance? The one that had tech executives sweating and policy makers scrambling?
  
  Well, here's the thing - even the guy who WROTE it just admitted it's not happening. Daniel Kokotajlo, one of the main authors, has pushed his own predictions back. And we're not even through 2025 yet, and the scenario is ALREADY missing its milestones.
  
  Gary Marcus, one of AI's most respected critics, called it what it actually is: "a work of fiction, not a work of science." And the math? The scenario requires about 8 different improbable things to ALL happen perfectly. The odds of that? Basically zero.
  
  So how did science fiction almost become the basis for American AI policy? Let's break it down.

[Brigit]

And there is an incredible study in "Predictive Programming" on this subject
that's wide open, if someone loves fiction and the movies.

The most recent movie plot based on the "Self-Improving, Escaped" ai program I saw was Mission Impossible.

But I also only recently realized that the Avengers movies back in 2015 had an escaped ai called "Ultron" along with another ai character named "Jarvis." These were foils of one another, one good and one bad. At the time I took no notice.

[Brigit]

What concerns me most is the fact that platforms we all use, and that Thunderbolts Project (along with the Electric Universe) uses, have changed their terms of service.

Terms of service for Youtube grants permission for ai tools to be applied to their videos. The channel Anton Petrov had horrible alterations made to some of his videos.

The new terms also allow the titles to be changed by ai.

Also, wordpress has introduced ai tools into blogs and websites. Allowing any changes to your published materials by an ai is a catastrophic course of action!

My foremost concern in all of this is the archival problems this introduces into the entire internet, and I already find that much of what I have used is lost or is moved to the webarchive.

[Brigit]

What people need to keep in mind is that every time they use a summary by an ai, they are neither reading the original website, nor visiting the website. This is something that small creators may not be able to survive.

For example, if people use ai to summarize material here, the web traffic is drastically reduced. Not only that, the description or summary generated by a large language model is not reliable.

I am at a loss for understanding the enormous confidence people are placing in large language models, or machine learning, for that matter. Do Global Circulation Models predict the weather, or even trends? Did computer model accurately model the spread of pathogens in animals, or in humans? Do astronomical models tell us why galaxies are shaped the way they are, why their rotational curves are as they are, or how planets form? Based on past catastrophic failures in these areas, why do computer models now have such a huge war chest with internet-search users?

Why do people who are usually rational enough to reject computer model results in systems like the Earth's atmosphere, the spread of pathogens, and the creation of stars, suddenly embrace a computer model for the human language?

These are things that keep me up at night.

[Brigit]

I am very concerned about the destruction and alteration of original works on the world wide web.

I don't know how many of you like the BeeGees, or have even heard of them, for that matter! But they were a band made up of three brothers who were famous for their harmonies***. Although their name is most famously linked with Disco hits, and the subsequent backlash against disco music in the 70s, they performed and recorded music from the 50s all the way up to +-2012. I think that there may be occasional releases by Barry Gibb, who has written songs for and worked with many other artists, even up to the present.

I bring this up because their voices have provably and demonstrably been altered by pitch correction. Pitch Correction of Barry, Robin** and Maurice Gibb? ?? ???

Here is the analysis by the CH: Wings of Pegasus.

• "WE HAVE BARRY GIBB'S ORIGINAL LIVE VOCAL FROM 1989!!!"
  https://www.youtube.com/watch?v=RHWlzc5H8a8
  Wings of Pegasus

He has been doing a great deal of work on his channel showing the effects of autotuning, pitch corrected lyrics, and lip sync in live performances. While these tricks are not exactly new to the industry, he shows that autotune programs or ai alterations are being applied to the music of bands from the past, like the BeeGees. This was done on the platform, Youtube. But the original inspiration of Wings of Pegasus is his gift for helping people all over the world to appreciate the ranges, expressive power and beauty of the voices of singers, past and present. The most powerful performances are rarely pitch-perfect. There is something else to it.

This is not an improvement to the world wide web, and what I am trying to illustrate is that the introduction of ai and hyperscale datacenters are not just neutral infrastructure, but a threat to original works and anyone who uses any of the popular platforms.







**ref: WHY did Robin Gibb's voice sound SO vulnerable?! Let's have a look!
https://www.youtube.com/watch?v=ebH8RfYzzf4
***ref: Bee Gees - Three Song Medley
https://www.youtube.com/watch?v=bmbagLLqeqo

[Brigit]

Also, what part of "Datacenters" don't you understand ?!

if you stop and think for a moment, whose data is going to be collected?

Why would anyone build and power these data centers, capable of minute tracking of every single person on the planet, unless they believed it was for another reason?

No one. No one would build this machine for P. Allen Tier unless they were under a mass delusion that it was all for the purpose of another project.

[Brigit]

In the Space News episode called "Seeing Double - Electric Cosmology," Wal Thornhill addressed the ubiquity of computer models in science in this way:

• Stuart Talbott: "Welcome to Space News from the Electric Universe,
  brought to you by TheThunderbolts Project™
  at Thunderbolts.info
  How do objects in our universe form?
  In the unimaginably vast cosmos at all scales,
  from comets and asteroids in our solar system
  to the vastest superclusters of galaxies
  stretching for hundreds of millions of light years,
  astronomers and astrophysicists imagine gravitational processes --
  and only gravitational processes -- governing these objects' formation.
  
  But the objects we see tell a different story and demand new theoretical pathways.
  
  In our own solar system, one of the most puzzling forms
  is the double-lobed shape of most common nuclei imaged to date,
  a weirdly similar form also seen in nebulae
  and even in the peanut-shaped cross-section of the galactic bulge of our own Milky Way.
  
  In this episode, physicist Wal Thornhill will explain
  why the pairing of celestial objects
  is a predictable effect of a universe governed not by gravity
  but by electromagnetism."
  
  Wal Thornhill: "It is almost 50 years since Hannes Alfvén
  predicted the inevitable death of gravitational big bang cosmology.
  Why is it taking so long?
  The reason seems to be that cosmology has become a dogmatic belief system
  as firmly entrenched as any religion
  within a global fraternity who have been taught
  that computer modeling based on those beliefs will provide the answers
  someday, at whatever cost.
  
  But computer models tend to have a life of their own.
  New parameters, forces, particles and imaginary celestial objects can be introduced
  and endlessly adjusted to match almost any observation.
  
  Gone are the classically trained scientists of the 19th century
  who aimed for simplicity and matched theory against physical experiment.
  
  Today, theorists waste most of the computing power on Earth
  generating virtual reality objects and universes.
  They sit in front of computer screens reflecting back to them
  what they know, about what they know, about what they know.
  
  Meanwhile, the key is to be found in what scientists don't know they don't know,
  which is a consequence of dogmatic training that there is no alternative to gravity
  as the governing force in the universe.
  --Yet no cosmologist will admit that the force of gravity
  remains unexplained in the 21st century...."

He points out that "theorists waste most of the computing power on Earth generating virtual reality..."

This is an extremely helpful pointer which can aid our analysis of the development and use of computer models.

After all, a Large Language Model is a computer model. The LLMs give the impression of modeling the intent, understanding, and intelligence behind language, but in fact, as the OP points out, the LLMs are a probabilistic model using numeric values, purportedly (and they use original works only slightly reworded).

So, to answer Wal Thornhill's observation about the actual "waste of computing power on Earth," -- just how much energy, water, and computing power has been devoted to cloud datacenters in the US? And how much more electricity, water, and computer chip production, in real measurements we all use, would be needed for ai hyperscale data centers?

[Brigit]

I found this documentary which describes an attempt to map all datacenters in the US, and the difficulties in determining their power usage. These datacenters claim "trade secrets" and other protections to avoid stating their power and water usage. In addition to the present datacenters, back of the envelope says the ai datacenters are projected to require the US tomore than double all of the existing power generation in this country.

• Exposing The Dark Side of America's AI Data Center Explosion | View From Above | Business Insider
  https://www.youtube.com/watch?v=t-8TDOFqkQA
  CH: Business Insider
  Sep 12, 2025
  dur. 31:09
  • The explosion of AI across every industry has seen hundreds of water- and power-hungry server farms sprout up across the US. Already, one-third of the world's internet traffic flows through data centers in just one US state: Virginia.
    
    However, until now, there has been no official record of the number of data centers in America, who owns them, or how much electricity they consume.
    
    In an exclusive deep dive into the industry, Business Insider reporters cracked the code and, for the first time, revealed the true cost of the data warehouses feeding our growing appetite for cloud computing and AI. We travelled to Virginia to meet people living in the shadow of 80-foot-high boxes that emit a constant drone, and to the drought-ridden state of Arizona, where some data centers are using as much as a million gallons of water a day to help cool their computer servers.
    
    Business Insider also discovered that the power needs of data centers have forced some states to withdraw from their carbon emissions targets. Power companies are even looking to extend the life of coal and gas plants to help meet the unprecedented demand.
    
    Check to see what data centers might be near your region:

[Brigit]

Also, someone asked the people who are so deeply involved in raising the hundreds of billions in leveraged (read: debt) and investment capital needed for these datacenters,
"What's your product?"
"Data, sir."

"Well it better be damned good data," he said.






ps All I can say is, you better keep your dirty rotten scoundrelly
Tech Bro paws off of "Our House"

• Madness -- Our House
  https://www.youtube.com/watch?v=KwIe_sjKeAY