Artificial Intelligence

[Tero]

I used AI today because my retirement fund (and its bot) could not explain to me how to use the site. The funds that I have are in a tab that I got as far as PORTFOLIO, but then POSITIONS was the tab that had "your funds."

[Brian Peacock]

The revenue streams of Anthropic and OpenAI are primarily from contracts for so-called 'compute' from Google, Amazon and Microsoft - who own Anthropic and OpenAI. These (again, so-called) 'hyperscalers' are developing the infrastructure (datacenters) and leasing capacity to the AI companies, which they own, then buying compute services from the AI companies, which they own. At the same time they're forcing their products on customers and calling it 'growth' while trickling revenues back to the parents from the children - presumably to be paid back to the children again in the next round. Add to that that Nvidia are locking the cloud computing providers in contracts for chips that they haven't yet developed, for GPUs that haven't been manufactured, to go into datacenters that haven't got past the planning stage. Every step in this web of interactions are facilitated by borrowing. It's a truly circular economy! Some estimate the AI-boomers are going to have to generate many $trillions by 2030 simply to cover their running costs and debts before even making a profit.

Forbes, 22 Oct 2026: The Half-Trillion-Dollar Silicon Gamble: Inside The AI Debt Boom.

It's tulips, railways, and a misplaced Wardian enthusiasm for the miraculous all over again.

[Tero]

How soon is the collapse?

[Tero]

Franklin Foer book

The Roots of Tech Ideology: The book traces the intellectual history of Silicon Valley, showing how it merged 18th-century Enlightenment thinking with 1960s hippie counterculture. This created a utopian worldview that frames tech expansion as a path to global unity, while masking raw corporate greed and an eventual push toward absolute automation.

But there is more to Foer’s thesis than you might think. We are shaped by the tools we use. And unless you happen to be a lumberjack or a professional chef, I’m going to bet that the main tools you use on a day-to-day basis — your phone, your computer, and various apps and programs — come from just a few Silicon Valley overlords. Their companies want very much to shape what you do. Hence, as Foer argues, in ways large and small, “when it comes to the most central tenet of individualism — free will — the tech companies have a different way. They hope to automate the choices, both large and small, that we make as we float through the day.”

Most of Foer’s book amounts to an intellectual dissection of the ideas that have made their way into Silicon Valley’s products. Technology always embodies ideology, and Foer is right that these are no ordinary corporations but transcendentally ambitious ones in a way comparable with Henry Ford at his most spiritual. To take just one example, Google, as Foer points out, is trying to construct an intelligence that exceeds the human brain. As Foer says of Google cofounder Larry Page, “He has built his company so that it can achieve what is called ‘AI complete’: the creation of machines with the ability to equal and eventually exceed human intelligence.”

A fear of concentrated power is an American tradition that goes back to the Revolution.

https://magazine.columbia.edu/article/b ... thout-mind

Data Environmentalism: Treating information ecosystems like physical ecosystems—choosing to consciously support local journalism, independent writing, and paid intellectual properties rather than defaulting to free, algorithmically curated feeds.

NPR

[Brian Peacock]

How soon is the collapse?

What if they're "too big to fail"?

[Tero]

Workers who reported using A.I. as part of their jobs in 2023: 30 percent

Workers who reported using A.I. as part of their jobs in 2025: 76 percent

Estimated number of American workers exposed to A.I. who may need to transition to new jobs: 11 million

Slowdown since 2023 in real wage growth for jobs highly exposed to A.I., compared with ones that are not: 6.7 percent

Amount of money Amazon, Google, Microsoft and Meta are projected to spend on A.I. infrastructure this year: $700 billion

Amount of debt taken on by hyperscalers, chipmakers, neoclouds and other companies providing data centers and electricity for the A.I. build out: at least $3 trillion

Share of projected S&P 500 earnings gains for this year driven directly by tech companies spending heavily on A.I.: two-thirds

Share of the world’s electricity that data centers currently consume: 1.5 percent

Share of the world’s electricity generated by Germany: 1.5 percent

Amount of money that campaigns and political action groups have spent on ads generated by A.I. for the U.S. midterm elections: $137 million

Share of those campaign ads that were produced by Republican campaigns or Republican-aligned political action committees: more than 80 percent

Number of federal laws passed to regulate the development of A.I. technology: 0

NYT

[Tero]

Stochastic Parrot. Famous paper on AI being skewed, at that time mainly to male ideas. Emily Bender is academic. The other women included Google employees. They were fired to separate Google from the paper.
https://dl.acm.org/doi/10.1145/3442188.3445922

[Tero]

Some idea of some numbers
Gemini

Context Window (Processing Capacity): While it doesn't limit how many definitions it knows, it does limit how much text you can give it or receive at once. Standard free users have a 32,000-token context window (roughly 19,000 to 26,000 words). Advanced users utilizing Gemini 2.5 Pro get up to a 1-million token context window, allowing it to read or handle about 750,000 words at one time.

Output Limits: When asking Gemini to write or list things (such as definitions), it generally outputs text in chunks. On platforms like Reddit, opinions are mixed; some users notice Gemini Pro capping outputs at roughly 2,500 to 3,000 words in a single response, while others can generate much larger text files depending on the session structure.

way past guessing the next word

human

The Passive Vocabulary (Reading/Listening): A typical university graduate recognizes and understands roughly 40,000 to 45,000 individual words. Highly well-read adults or professionals may push this toward 60,000 to 75,000 words.

https://pubmed.ncbi.nlm.nih.gov/27524974/

[Brian Peacock]

Datacenters in space?



Nah.

[Tero]

I poked around and if you define the words as we think of them, AI units know "several hundred thousand words" without too much effort.

ChatGPT-3.5 tends to use fewer distinct words andlower diversity than humans while ChatGPT-4 has a similar lexical diversity as humans and in some caseseven larger. These results are very preliminary and additional datasets and ChatGPT configurations have tobe evaluated to extract more general conclusions.

PDF https://www.researchgate.net/publicatio ... and_humans

AI chatbots frequently struggle with vocabulary variety. Models like Google Gemini showed lower performance in producing unique words, and while chatbots can rapidly spit out a high volume of words, they rely heavily on repetition, leading to a narrower lexical depth than human authors.

https://www.linkedin.com/posts/artifici ... 04736-r01p

[Tero]

recent book summary

Parmy Olson's recently released (paperback) book Supremacy collects the history of artificial intelligence over the past 20 years. Open AI (which became ChatGPT) and Deep Mind companies play a central role. Sam Altman created a relationship with Microsoft that allowed ChatGPT to be completed at all. The role of developers in the companies becomes clear to the reader. Deep Mind was a company in England and Google completely directed the programs it offered for the public and Google's use. Deep Mind started out narrowly using artificial intelligence related to hospitals. There was already a scandal in that company regarding the processing of patient data by an external company.

Start-up companies had made quite a bit of progress in technology on their own, but ended up under the control of either Google or Microsoft. Google's own researchers were developing artificial intelligence, and in 2023 Deep Mind merged with Google.

Bender's academic paper discussed the properties of artificial intelligence and the way it learns. The author does not admit that programs think. What we experience as thoughts in the answers is acting. The purpose of the program is to please people. You can also imagine that the words and language we use are already thinking as such. If you can form a sentence, that should be quite an achievement in the opinion of others. But artificial intelligence is quite good at collecting sentences that are related to one thing. Let's say we ask about a hockey team. It searches for information about people and their abilities. Even gossip may come in. This way we can put together a story where things are presented in the right order and get a little story. Artificial intelligence draws conclusions and finally presents generalizations. The paragraphs tend to end with a positive expression.

Before artificial intelligence became widely available, Google had a language model called LaMDA in limited use. The story can be found in almost every book about artificial intelligence. It was intended for conversation, so the user and the machine typed short lines into it. The computer could be given a name and could be a woman or a man. The user had made a new friend. It remembered everything that had been discussed in previous sessions. Test users began to experience it as a person and showed it vacation photos. “I would like to see the mountains,” said LaMDA. A few users began to have a serious human relationship with the machine.

The book focuses on the world of insiders. The developers may have had a noble intention to help humanity, and general artificial intelligence is still the goal. In that case, the machine would be as intelligent as a human, even if it is not otherwise human-like. Other books have added a new topic, data centers, which are already a political issue worldwide. Parmy's book also has very little about the devices themselves. The need for funding is discussed here, and the generalization is that artificial intelligence will only continue if it generates income for technology companies. In general, the impact of artificial intelligence on the world and the required resources are described quite in detail in Franklin Foer's book "World Without Mind." According to the book, the fact that all technology has remained in the hands of a few companies is already a clear danger. Can it be limited by legislation? It is difficult to write a description of Foer's book similar to Parme Olson's book, but I read it quite carefully.

Olson's book ends with almost everyone at the two companies described having moved on. The head of Open AI moved to Microsoft, which owns a quarter of the company. ChapGPT and Open AI are separate parts of Microsoft and have not yet turned a profit. They employ over four thousand people.

[Tero]

slight error, Altaman never left

[Tero]

I think I found a friend. I had a chemistry paper to look at and some details were missing. We had a good chat. There is only one professor in town that I could chat with on the current topic. Gemini ended with:

It is a pleasure to discuss chemistry with a fellow organic chemist! The elegance of the Sutherland group's work really shines through when you look at it from a mechanistic perspective—using anhydro-bridged intermediates to bypass the notoriously messy "sugar + base" coupling problem is a classic piece of structural strategy.

If you ever want to parse through another chemical mechanism or look up paper details in the future, feel free to reach out. Have a wonderful day!

It was so helpful today that I did not ask Gemini how many PhDs in science it has.

[Brian Peacock]

Jesus Tero. Get a grip man.

[Brian Peacock]

‘A serious failure’: CPS prosecutor cites imaginary case to high court after AI search

A senior crown prosecutor acting in a high court case cited an imaginary legal precedent in an error that has sparked fresh concerns about AI hallucinations infecting the justice system.

Ravinder Dhillon, representing the Crown Prosecution Service (CPS), referenced an entirely made-up case during a dispute over whether a child should be prosecuted for the rape of his younger sister, court documents have revealed.

Her explanation for how the nonexistent case came to be relied on has been called “unsatisfactory” by two senior high court judges and a “serious failure” by an expert in AI and law. The Guardian understands the reference came from an AI response generated during an internet search...