THIS JUST IN. I asked my AI-loving nephew to prompt the Claude LLM to write jokes. You be the judge.

I suddenly stumbled on the perfect metaphor for LLM/Not AI.

Many birds can talk. They imitate the speech they have heard in the home where they live (not unlike the LLMs “listening” to the data they steal/scrape). Talking birds operate on exactly the same principal as large language models. No one would say talking birds are intelligent; why would anyone say that about an LLM?

When a talking bird says, “How are you buddy,” it has no idea who it is talking to or what it is saying. It’s just repeating words it has heard. It is not insulting you when it says “Up yours lady,” nor does it know it is talking about itself when it says, “Pretty Bird” or “Polly Want A Cracker.” It is not self-conscious. It understands nothing about what it is saying. Sometimes it randomly says something appropriate, just like an LLM.

With regard to consciousness and vocabulary, Wikipedia says “There is debate… whether…talking parrots also have some cognitive understanding… [S]ome budgerigars have been observed to have a vocabulary of almost 2,000 words.” Which just goes to show a large vocabulary does not necessarily indicate cognition.

Speaking only for myself, I don’t want a talking bird doing my programming, my customer service, my hiring, my strategic planning or my writing. If you do, be my guest. They are entertaining, but not as entertaining as Taylor Swift or Billy Joel.

Another good analogy: LLMs are just Imessage predictive text written large and expensive. Predicting the next word is what they do best and most reliably. Any task that requires more than that is probably inappropriate for an LLM.

(The permanent link for this item)

Unless you were living on another planet in late June, you were exposed to a viral hoax about Jeff Bezos and the water required for LLM data centers.[1]

I would rather dive headfirst into a swimming pool full of double-edged razor blades than offer Jeff Bezos the benefit of the doubt, but he has said nothing (publicly) about LLM data centers and water. He may believe water for LLM data centers is more important than for people, but he hasn’t said it out loud.

One interesting commentary (apparently written before it was clear the story was fake) suggested context would be in order. Here’s a deep dive on the issue of LLM data center water usage from a company that tests water samples. (pun accidental)

I am offering a 50,000 foot view of this issue, demonstrating why America can’t get its head around it.

The President of Microsoft said data centers use about the same amount of water as a restaurant. There are two things he didn’t say: what size data center and what size restaurant. He did, admittedly, say that new data centers would use less, but that distinction could easily be missed. He said nothing about all of Microsoft’s existing water-sucking data centers. Closed loop LLM data center cooling will someday result in minimal draws of water. Someday. We don’t live in someday, we live in now.

The industry is right when it says data centers use a tiny fraction of the world’s water, compared to agriculture and a lot of other industries.

But the people of Gilroy, Calif., for example, don’t live in the world; they live in Gilroy. Same for lots of other people living in water-stressed areas where data centers are going up.

Since this issue can’t easily be understood during the time it takes to watch any single episode of any flavor of Law and Order, most people aren’t going to bother. Their “opinion” will be based on the opinion of their favorite Facebook page, TikTok influencer, Instagram news source or blogger. Or someone they know whose opinion they respect. Or, once in a blue moon a politician, newspaper or foundation that has gone to the trouble of analyzing and balancing the facts, in context, putting humans first.

Well, since this is the way 99% of Americans decide how to vote, I guess the system can’t be all bad; look at the results of recent elections.

[1] Creating a viral hoax on the Internet is easy. The best ones involve quotes attributed to an actual person on an actual date, at an actual event. Since people are too lazy to research, these elements lend sufficient believability to fool most people, especially if it sounds like something the person would say. “A lie is halfway round the world before the truth has got its boots on,” the ancient poet Virgil almost said.

I asked the Claude Large Language Model to compare two Trump news conference transcriptions: September 27, 2020 and April 6, 2026. It concluded that a comparison of Trump’s unscripted and impromptu words showed detectable cognitive decline during those six years.

Claude reported, “This is the most striking quantitative finding. The 2026 speech has nearly 3× more mid-sentence interruptions relative to word count. This reflects a significantly higher rate of losing the thread, restarting, or abandoning clauses — a known linguistic marker researchers associate with cognitive aging.” Full Claude Report

Project Structure, methods and sources. Why and how I did this comparison, why I trusted the generally inaccurate and hallucinating Claude LLM engine’s results and why it isn’t TDS.

This from Wired Magazine

 Google Fired an AI Ethics Researcher

(She was fired for writing about ethics in Large Language Models)

In Dec. 2020, Google artificial intelligence [ed. note: LLM] researcher Timnit Gebru received disturbing news from a senior manager. She says the manager asked her to either retract or remove her name from a research paper (On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?) she had coauthored, because an internal review had found the contents objectionable.

Courts are getting tired of AI/LLM Hallucination-filled lawsuits, and are now sanctioning the lawyers involved.

Paying a $10,000 fine, proving you gave your client a copy of the decision and being referred for disbarment are terrible, but ephemeral. Yes. Disbarment (should it happen) means you lose your livelihood, but doesn’t normally lead to being a laughingstock in legal decisions for the next century or two.

Imagine a case 50 years from now citing your work in Noland V. The Land of the Free as a basis for dismissal.

Will there will still be either human attorneys, or human judges involved in the law? In any case, the LLM for the plaintiff has no shame, so it won’t even be embarrassed when it cites this case as “clearly ruling that AI-generated briefs are acceptable in California’s Fourth Appellate District, Division 44, in the case of Abbott V. Castello.” And the LLM judge will be imagining something else.

AI the time saver. Let an LLM write the brief, then read every word and check every citation anyway. Have you halved your work or doubled it? Wouldn’t it take less time to proofread for citations you already know are real, rather than guessing which is which?

I am starting to think LLMS’ future is in the creative arts, not in the law or medicine (or, for that matter, Air Traffic Control)—anywhere that imagination is useful and not deadly. Of course its art and music will be crap, but at least it will be cheap crap .

And while we’re at it, the law school at Berkeley has prohibited the use of LLMS.

This is a brief quotation from a Ruben Bolling cartoon, used in a critical article (well not really critical)

You can see the whole cartoon here.

It is too bad that I am on the wrong side of history, and there’s nothing I can do to stop the universal adoption of LLM in all areas of life. People  say it’s just like the Industrial Revolution. But I don’t recall steam looms ever killing people, except employees.  

It is one thing to create a machine that produces cheap, low quality cloth, in such high volume and so cheaply that it replaces handcrafted cloth. It is something else entirely to have cheaper, lower quality writing, computer programming, management, customer service, art and medicine… replacing handcrafted human output, without human supervision.

At least someone looked at the product coming out of the knitting mills. Of course they never rejected it, unless it was knitting slop. No knitting hallucinations. No factory-made cloth ever told someone how to commit suicide.

I know for a fact that the statement “a human being reviews our LLM rejections” is a lie. A rubber stamp is not a review.

I submitted a piece of art to a vendor. The LLM rejected it because it “included text.”

I was told that a human had reviewed the LLM rejection. I cannot imagine how that is possible—except that the poor reviewer probably has to look at 100 pieces of art a day, staring at them until visual fatigue sets in, and it is easier to rubber stamp the rejection than actually think about it. It doesn’t cost anyone time and money… except me. Well, OK, not money, just time. And yet my time, even in retirement, is not worthless.

My friend, Tech-Savvy attorney Joe Kashi says it can’t.

Designed to Cross: Why Nippon Life v. OpenAI Is a Product Liability Case

is all very plausible, he says and seems to be a further extension of legal sanctions against  hallucinatory AI legal behavior.

Alas, there is no abstract of the case, human or otherwise. To this civilian, it boils down to whether an LLM can practice law in the traditional sense. So, should we redefine the practice of law for the LLMs, or make the LLMs respect the current definition of practicing law? It should be obvious to the casual observer which way I lean on the question.

Joe also says says that filing  imaginary citations currently produces more legal malpractice suits than any other single error.

The civilian world is looking for tells that identify AI produced artifacts. Of course, as soon as such a tell is widely circulated, AI producers just change the query to include “without using em dashes,” for example.

WordPress offers AI to improve the title of a post; that is, make it friendlier to the AI bots which now do all web searches. My title was “Kindness and Empathy Produce MIT Degree, which AI turned into “Thanking Those Who Made My MIT Journey Possible.” I find “journey” to be an AI tell. Do you?

BTW, the default on WordPress is to let the AI go crazy. Negative option; it changes the name unless you say no.