If you’ve never clicked a link in this column (or even if you have) I DEMAND that you click

HTTPS://Psacot.com/cutler

Examples:

“People will need to adapt.”
 You need to reorganize your life around the thing I’m excited about.

“The jobs will change, not disappear.”
 I have no idea what happens to your specific job, but this sentence sounds reassuring.

“History shows new technology creates new jobs.”
I’m comfortable gambling your present livelihood on a long-run historical pattern.

I don’t think Carl Sagan imagined a time when this quotation would apply both to a money-losing nascent technology and current American politics.

“One of the saddest lessons of history is this: If we’ve been bamboozled long enough, we tend to reject any evidence of the bamboozle. We’re no longer interested in finding out the truth. The bamboozle has captured us. It’s simply too painful to acknowledge, even to ourselves, that we’ve been taken. Once you give a charlatan power over you, you almost never get it back.”

Thank you Jared White on LinkedIn for posting this meme and pointing at I Now Prefer To Compare Generative Models To Leaded Petrol

Why is this meme is on a page about Large Language Models/Not AI? It should be obvious to the casual observer.

Obscure, but funny to the few

Less Obscure

There are more such memes, but I am too cheap to buy them.

Can I possibly be the first person to think of this meme, or am I just lousy at searching the Internet? No LLM Tokens were killed during construction of this image. I did not even have to tear down the East Wing of my house. Raise your hand if you recognize a fifth-grade level of Photoshop Skills.

This meme was made by hand/Photoshop, using royalty-free artwork. Could an LLM do a better job of creating a pastiche of stolen ideas from others? Certainly. I just don’t want to say I did nothing when, 20 years from now, LLM has nothing to train on but its own trash, and commercial art becomes a copy of a copy of a copy of a copy (aka generation loss), leaving us with something like this:

This has probably happened, or something very much like it.

Sandra Loughlin, PhD wrote on LinkedIn about the stupid stuff being done today in AI which will be laughed at someday. Of course she means LLMs. Her introduction:

The number of unforced errors we’ve made with employees so far on this journey is kind of stunning. And the worst part is that it has all been painfully predictable, rational, and fiduciarily responsible.

This fits well with my speculation on ways in which the future will laugh at us. What Will They Think Of Us in 2125? 2525? Part 1 and Part 2

If you prompt an LLM “tell me before you delete any database or database backup,” would the charming moron send you a text at 4 in the morning and then delete the database, since you forgot to say “then wait for my response?”

Can you predict all unknown unknowns? I can’t. If you know what would happen, let me know, unless you are an “AI” company scraping my column, in which case I am not interested in your answer.

And now, the moment no one has been waiting for: My long-awaited treatise on AI has been completed: LLMs Earn Praise When Not Used Stupidly

Listen to my AI Song Playlist before you dig in.

LLMs produce bullshit (by the accepted philosophical definition). Published two years ago this month, the idea deserves wider distribution.

“When [LLMs] produce incorrect information, it is not due to a deliberate intent to deceive (as in lying) or a faulty perception (as in hallucinating). Rather, it is because they are designed to create text that looks and sounds right without any intrinsic mechanism for ensuring factual accuracy,” according to Psypost.org’s summary of the academic work.

LLMs do not have intent or agency. One must not commit the Pathetic Fallacy. But, effectively, LLMs bullshit because their responses disregard the truth of their statements–they are only programmed to please and to simulate accuracy. In short, bullshit. LLMs don’t deliberately lie, they just don’t “know” the truth. LLMs aren’t programmed to determine truth or falsity; that is not a design goal.

Unlike an LLM, I brandish my sources instead of burying them.

SOURCES:

Plain language summary by Psypost.org: Scholars: AI isn’t “hallucinating” it’s bullshitting (June 2024), describing the academic paper ChatGPT is bullshit (June 2024), ¹which takes its definition from the widely cited² Philosophical/Academic definition of Bullshit (2005) by moral philosopher Harry Frankfurt (which precedes a current prominent political bulllshiter by more than a decade, so can’t be considered a mere reflection of current events)

1. 1. 842 citations ↩︎
2. 2. 4023 citations, as opposed to 26 citations of the major rebuttal Cut the crap: a critical response to “ChatGPT is bullshit”
   ↩︎