Outside Topic Artificial Intelligence question

Outside Topic Artificial Intelligence question - Printable Version

+- The CardBoard (https://thecardboard.org/board)
+-- Forum: C-House! (https://thecardboard.org/board/forum-4.html)
+--- Forum: The CardBoard (https://thecardboard.org/board/forum-5.html)
+--- Thread: Outside Topic Artificial Intelligence question (/thread-25744.html)

Artificial Intelligence question - Mick - 09-14-2025

I've been experimenting with a variety of AI tools. I realize that they can be hallucinatory and invent citations and so forth, and in its current form, I wouldn't use it for anything other than a draft of certain sections. It's not trustworthy yet.

And yet...I just asked it to add seven one- and two-digit numbers. The total was 85, but the AI engine came up with 87.

A calculator won't fail, the Excel spreadsheet I'm dropping it into doesn't fail when adding and even a wrinkly old human like myself can eyeball seven numbers, add them in my head and come up with the correct number. And to be fair, it added thirty similar sequences correctly.

I don't know when or if AI will take over the world, but...not being able to add seven numbers? What can we infer from that?

RE: Artificial Intelligence question - Genuine Realist - 09-14-2025

(09-14-2025, 10:09 AM)Mick Wrote: I've been experimenting with a variety of AI tools. I realize that they can be hallucinatory and invent citations and so forth, and in its current form, I wouldn't use it for anything other than a draft of certain sections. It's not trustworthy yet.

And yet...I just asked it to add seven one- and two-digit numbers. The total was 85, but the AI engine came up with 87.

A calculator won't fail, the Excel spreadsheet I'm dropping it into doesn't fail when adding and even a wrinkly old human like myself can eyeball seven numbers, add them in my head and come up with the correct number. And to be fair, it added thirty similar sequences correctly.

I don't know when or if AI will take over the world, but...not being able to add seven numbers? What can we infer from that?

It was luring you into a false sense of security.

I asked one of the programs - maybe Chat G/pt - if there was a God. I was hoping someone with a sense of humor knew of Frederick Brown's famous 1954 story, Answer, in which a super-duper-super computer, hooked up to a gigantic array of computing power, is turned on, with all the luminaries in the world watching. The President asks the same question - is there a God? The computer answers, There is now!, and simultaneously a bolt of lightning fries the on-off switch so it can't be turned off. The story is well known, with frequent variations.

In any case, all I got was a namby-pamby discussion of various metaphysical theories. Maybe there are people that wouldn't have found the answer I wanted all that funny.

RE: Artificial Intelligence question - JeffInCorvallis - 09-14-2025

Mick dateline='[url=tel:1757869790' Wrote: 1757869790[/url]']
I've been experimenting with a variety of AI tools. I realize that they can be hallucinatory and invent citations and so forth, and in its current form, I wouldn't use it for anything other than a draft of certain sections. It's not trustworthy yet.

And yet...I just asked it to add seven one- and two-digit numbers. The total was 85, but the AI engine came up with 87.

A calculator won't fail, the Excel spreadsheet I'm dropping it into doesn't fail when adding and even a wrinkly old human like myself can eyeball seven numbers, add them in my head and come up with the correct number. And to be fair, it added thirty similar sequences correctly.

I don't know when or if AI will take over the world, but...not being able to add seven numbers? What can we infer from that?

A large language model works by predicting the next phrase or “token” based on its training, which might include just about everything on the internet, papers, articles, blogs, books regardless of copyright, and posts from the Cardboard. Great for answering some types of questions, regurgitating info, summarizing, editing emails, and so on. But highly quantitative tasks involving logic and calculation, not so much, as you have discovered.

RE: Artificial Intelligence question - BostonCard - 09-14-2025

Yeah, there are lots of examples of LLMs failing at what are perceived as simple tasks, such as counting the number of “b’s” in “blueberry”:

[Image: chatgpt.webp]

That’s because the model doesn’t actually count or do math, the model predicts the response based on the data it has; sometimes it is correct, but many times it is wrong.

I like feeding ChatGPT the Economist’s weekly quiz. There are five factual questions, whose answers are connected by a theme. The models are nearly perfect at answering the factual questions, unless they refer to an event that has happened after the model’s data cut-off. It is god awful at deducing the theme. For anyone who wants to give it a try, here are this week’s questions:

Monday: Which Austrian dignitary was assassinated, along with his wife, in Sarajevo on June 28th, 1914?
Tuesday: What title did Napoleon Bonaparte assume in December 1804?
Wednesday: Which British rock band had a hit single with “Love Will Tear Us Apart” before reforming as New Order?
Thursday : Which film was really awarded the Oscar for best movie in 2017, after the presenters mistakenly announced the winner as “La La Land”?
Friday: Which New York congresswoman was nominated by President Donald Trump to be ambassador to the United Nations, only for her nomination to be withdrawn?

As far as themes go, this week’s was pretty straightforward, but getting the model to recognize it was like pulling teeth.

BC

RE: Artificial Intelligence question - fullmetal - 09-14-2025

I had it try a few crossword puzzle clues with the number of letters and known letter placements. It couldn't even get the number of letters right in its suggested words with any consistency.

And we all know about hallucinated case citations in the legal realm, lol.

RE: Artificial Intelligence question - 82lsju - 09-14-2025

in my former job most of the successful customer deployments used models that they trained on very specific datasets, actually more machine learning than generative AI, but they all still called the use cases AI because it was on message....

RE: Artificial Intelligence question - TonyLima - 09-15-2025

(09-14-2025, 04:01 PM)82lsju Wrote: in my former job most of the successful customer deployments used models that they trained on very specific datasets, actually more machine learning than generative AI, but they all still called the use cases AI because it was on message....

Thank you for validating my suspicions. AI is only as good as its trainers.

RE: Artificial Intelligence question - TrumpCard - 09-15-2025

(09-14-2025, 12:01 PM)BostonCard Wrote: As far as themes go, this week’s was pretty straightforward, but getting the model to recognize it was like pulling teeth.

BC

Okay, I'm not proud, you stumped me.

RE: Artificial Intelligence question - BostonCard - 09-15-2025

OK here are the answers:
Monday: Archduke Francis Ferdinand
Tuesday: Emperor of France
Wednesday: Joy Division
Thursday: Moonlight
Friday: Elise Stefanik

The theme that connects the answers is works by Ludwig van Beethoven:
"Archduke Trio.", "Emperor Concerto.", "Ode to Joy", “Moonlight Sonata”, and "Für Elise."

I’m not so knowledgeable of classical music that I knew the first two, but connecting “Joy Division”, “Moonlight” and “Elise Stefanik” was pretty straightforward. The hardest part is guessing which word of a compound answer (such as a full name) is tied to the theme, but I don’t know “Stefanik” in any other context, and “joy” seemed more likely than “division”.

BC

RE: Artificial Intelligence question - TrumpCard - 09-16-2025

(09-15-2025, 06:15 PM)BostonCard Wrote: OK here are the answers:
Monday: Archduke Francis Ferdinand
Tuesday: Emperor of France
Wednesday: Joy Division
Thursday: Moonlight
Friday: Elise Stefanik

The theme that connects the answers is works by Ludwig van Beethoven:
"Archduke Trio.", "Emperor Concerto.", "Ode to Joy", “Moonlight Sonata”, and "Für Elise."

I’m not so knowledgeable of classical music that I knew the first two, but connecting “Joy Division”, “Moonlight” and “Elise Stefanik” was pretty straightforward. The hardest part is guessing which word of a compound answer (such as a full name) is tied to the theme, but I don’t know “Stefanik” in any other context, and “joy” seemed more likely than “division”.

BC

That's a lot more difficult than I imagined it would be, and I never would have expected any sort of AI to pick up on that.

RE: Artificial Intelligence question - jacket3ree - 09-16-2025

If you ask the right question the right way, AI can be a powerful internet research tool. I'm using it a lot now to put together production rates and equipment lists for construction. Reasonable stuff that can be vetted with other sources. One random Friday I was going to town in Excel and realized the chatbot had been sitting blinking on another monitor for some time waiting for my next question about fuel consumption for a long-reach excavator. I asked it 'What's it All About Alfie? It knew all about the 1966 Bacharach/David song and movie and I was sucked into a sophisticated and philosophical conversation wherein it ask if I asked because it was Friday afternoon and I needed to go home. I quietly shutdown, closed the laptop, unplugged from the docking station and went home. Pretty sure I saw that helicopter from Goodfellas hovering over Great America Parkway as I turned right, just a little spooked.

Also, I'm pretty sure I'm on some Federal watch list because I asked about how many pounds of explosives and of what grade are needed to blow a 200-foot-long hole in some old salt pond dikes out in the Bay with horrible equipment access and it refused to help.

RE: Artificial Intelligence question - cardcrimson - 09-18-2025

Here's a question for AI: Can entropy be reversed. . . .

RE: Artificial Intelligence question - 82lsju - 09-18-2025

Quote:OpenAI released on Monday some research that explained how it’s stopping AI models from “scheming.” It’s a practice in which an “AI behaves one way on the surface while hiding its true goals,” OpenAI defined in its tweet about the research.

In the paper, conducted with Apollo Research, researchers went a bit further, likening AI scheming to a human stock broker breaking the law to make as much money as possible. The researchers, however, argued that most AI “scheming” wasn’t that harmful. “The most common failures involve simple forms of deception — for instance, pretending to have completed a task without actually doing so,” they wrote.

The paper was mostly published to show that “deliberative alignment⁠” — the anti-scheming technique they were testing — worked well.

But it also explained that AI developers haven’t figured out a way to train their models not to scheme. That’s because such training could actually teach the model how to scheme even better to avoid being detected.

https://techcrunch.com/2025/09/18/openais-research-on-ai-models-deliberately-lying-is-wild/

RE: Artificial Intelligence question - CardinalSagehen - 09-18-2025

My old Econ and Stats professor, Gary Smith, is a big AI cynic.

His account of AI trying to contemplate a new game called "90-degree tic-tac-toe" is a fun example:
https://mindmatters.ai/2025/01/agi-is-not-already-here-llms-are-still-not-even-intelligent/

If you're into this line of thinking, he writes about two articles per month on the topic:
https://mindmatters.ai/author/gsmith/

RE: Artificial Intelligence question - BostonCard - 09-18-2025

82lsju dateline='[url=tel:1758248355' Wrote: 1758248355[/url]']

https://techcrunch.com/2025/09/18/openais-research-on-ai-models-deliberately-lying-is-wild/

This reminds me a bit of the paperclip maximizer thought experiment: https://en.wikipedia.org/wiki/Instrumental_convergence#Paperclip_maximizer

Quote:Suppose we have an AI whose only goal is to make as many paper clips as possible. The AI will realize quickly that it would be much better if there were no humans because humans might decide to switch it off. Because if humans do so, there would be fewer paper clips. Also, human bodies contain a lot of atoms that could be made into paper clips. The future that the AI would be trying to gear towards would be one in which there were a lot of paper clips but no humans.

RE: Artificial Intelligence question - Mick - 09-19-2025

(09-18-2025, 06:43 PM)cardcrimson Wrote: Here's a question for AI: Can entropy be reversed. . . .

I did a general search, then asked that question of Grok, Gemini, Perplexity and ChatGPT. All of them said no.
Grok (232 words). Brought up Maxwell's Demon
Perplexity (262 words), explained the concept in the Physics article below.
ChatGPT (319 words). Cosmology, statistical fluctuation, local vs. universal.
Gemini (550 words). Offered to explain the thought experiment of Maxwell's Demon.

General search. Physics article. 853 words. They found a loophole in the Second Law of Thermodynamics in which entropy can decrease under certain circumstances (microscopic, short term). H-Theorem. APplied quantum information theory to condensed matter physics, new quantum H-theorem model. And yes, they invoked Maxwell's Demon. The Argonne team created a temporary "negative entropy gain" (decrease)

RE: Artificial Intelligence question - BostonCard - 09-27-2025

Interesting take from Joe De Sena, who runs the Spartan race, on AI.

Quote:Spartans!

Joe here in Vermont, just wrapping up a trail run in the Green Mountains. Every step through these mountains reminds me that human bodies evolved to handle hard terrain, not smooth sidewalks, glowing screens. We definitely weren’t made for algorithms deciding what we should see, eat, and think.

Lately I’ve been hearing the same fear everywhere: AI is moving too fast, our kids are getting weaker, the world is becoming unrecognizable. That anxiety isn’t new, however. In 1910, people said cars and airplanes were ripping life apart, moving at an “unnatural” speed! In the late 1800s, Americans talked about “nervous collapse” from telegraphs, railroads, and factory machines (imagine them handling TikTok). Plato himself warned that writing would weaken memory and destroy wisdom. Every generation looks at new tools and says: this is the end of us. And yet, it seems every generation is wrong.

Humans adapt, and that’s our superpower.

Acceleration feels scary because our wiring evolved for scarcity and stability. When change comes - faster, louder, more complex - our first instinct is panic. You see it now in the AI debate with parents fearing their kids losing creativity, workers fearing jobs will vanish, athletes fearing the playing field will tilt. Sounds like the same script to me, with a new cast.

Take running. “Super shoes” are carbon-plated, foam-stacked running shoes that were supposedly going to “ruin” competition. Critics said new records wouldn’t compare when Eliud first put on the Alphafly. Yet what happened? They became the standard. Athletes have raced faster, rethought strategy, and shattered times. The bar moved higher, and the human drive to push against it only got stronger. That’s not freaking collapse; that’s evolution in real time.

Take art. When cameras first appeared, critics said painting was done. When Pixar launched the first fully animated feature film, traditionalists said it wasn’t “real cinema.” Today, photography and digital animation aren’t the end of creativity, instead they’re part of it. Punk rock, born as rebellion against polished corporate music, shows how every wave of disruption provokes a harder, rawer counterpunch. It’s the same now with AI: for every formulaic algorithm spitting out generic crap, there’s an artist bending the tool into something new. Dave Watson here at Spartan often says AI is a power tool; it doesn’t replace the craftsman, it sharpens them, making them faster, more precise, and more effective.

Thinkers have wrestled with this long before AI. Freud warned that modern life crushes our instincts until we burn out. Weber argued the opposite, that discipline takes those same instincts and turns them into strength and progress. I think they were both right. Technology can smother you, or sharpen you - it depends on whether you let it own you, or you be the boss.

Fear of disruption has always been part of the human condition, but weakness isn’t inevitable. AI can replace your efforts, or it can multiply them. The question is whether you can build grit faster than the world builds comfort. I’ll take the Hard Way every time.

Hurry up,

Joe

RE: Artificial Intelligence question - 82lsju - 09-27-2025

and then there is the question of when companies (other than Nvidia and TSMC) will actually make money from AI, of course this question does not seem to bother many people currently but it seems a lot like the dotcom bust or the overbuild of fibre optic infrastructure (ours to our home got lit up about two years ago after not having access to the service that was about 1/4 mile from out house along Foothill Expressway for ~25 years, originally laid by PacBell).

Quote:Artificial intelligence companies like OpenAI have been quick to unveil plans for spending hundreds of billions of dollars on data centers, but they have been slower to show how they will pull in revenue to cover all those expenses
...
Now, the consulting firm Bain & Co. is estimating the shortfall could be far larger than previously understood.

By 2030, AI companies will need $2 trillion in combined annual revenue to fund the computing power needed to meet projected demand, Bain said in its annual Global Technology Report released Tuesday. Yet their revenue is likely to fall $800 billion short of that mark as efforts to monetize services like ChatGPT trail the spending requirements for data centers and related infrastructure, Bain predicted.

The report is set to raise further questions about the AI industry’s valuations and business model. The increasing popularity of services such as OpenAI’s ChatGPT and Google’s Gemini, as well as AI efforts by companies across the planet, means demand for computing capacity and energy is rising at a rapid clip. But the savings provided by AI and companies’ ability to generate additional revenue from AI is lagging behind that pace.
...
OpenAI is losing billions of dollars a year and prioritizing growth over profit, but it does expect to be cash-flow positive by 2029, Bloomberg has reported. Bain didn’t go into the likely implications for AI companies if they continue to fall short of profitability into 2030.

The biggest tech firms including Microsoft Corp., Amazon.com Inc. and Meta Platforms Inc. will ramp up their combined annual spending on AI to more than $500 billion by early next decade, according to Bloomberg Intelligence. The release of new models from OpenAI and China’s DeepSeek, among other companies, is spurring demand for AI services and driving the entire industry to boost investment.

https://www.siliconvalley.com/2025/09/23/an-800-billion-revenue-shortfall-threatens-ai-future-bain-says/

and

Quote:"However, a recent deep dive by Praetorian Capital CIO Harris Kupperman took a peek under the hood of present-day data centers — and for anyone pouring money into the space, his conclusions will be sobering.

"I’ve watched as AI went from an interesting parlor trick for making memes, to something that’s increasingly integrated into my daily workflow," Kupperman conceded. "I am not here to belittle AI, it’s the future, and I recognize that we’re just scratching the surface in terms of what it can do.

But "I also recognize massive capital misallocation when I see it," he continued. "I recognize an insanity bubble, and I recognize hubris."

To that end, Kupperman works out rough calculations for the genuine cost of a data center, factoring in the inevitable breakdown of parts over time.

Each data center, he says, is essentially made up of three components: the chips, which become obsolete in just a few years; the systems connecting the chips, which need to be replaced every decade or so; and the building itself, which should last for quite a while.

Add it all up, and time is not on the data center's side. The finance guru estimates that the "AI datacenters to be built in 2025 will suffer $40 billion of annual depreciation, while generating somewhere between $15 and $20 billion of revenue."

In other words, new data centers have a very tiny runway in which to achieve profits that currently remain way out of reach. By Kupperman's projections, a brand new data center will quickly become a Theseus' ship made up of some of the most expensive technology money can buy. If a new data center doesn't start raking in mountains of cash ASAP, the cost to maintain its aging parts will rapidly overtake the revenue it can bring in.

Given the current rate at which tech companies are spending money without much return — a long-term bet that AI will all but make human labor obsolete — Kupperman estimates that revenue would have to increase ten-fold just to break even. Anything's possible, of course, but it doesn't seem like a hot bet.

"I don’t see how there can ever be any return on investment given the current math," he wrote."

https://futurism.com/data-centers-financial-bubble