Diverse Nazi stormtroopers and other stupid AI tricks
When I breakdown just a little and lose my head
Perfecting Equilibrium Volume Two, Issue 68
I breakdown in the middle and lose my thread
No one can understand a word that I say
When I breakdown just a little and lose my head
Nothing I try to do can work the same way
Any time it happened, I'd get over it
With a little help from all my best friends
Anybody else could see what's wrong with me
But they walk away and just pretend
The Sunday Reader, March 17, 2024
There can be no doubt that the greatest author in the history of language is Nigel Richards. Oh, sure, that Shakespeare guy could turn a nifty phrase, and Papa Hemingway and Alice Munro won Nobel prizes for their ability to make readers simultaneously weep with rage and sorrow within the confines of a handful of pages. That was good!
But Richards has won SIX world championships in Scrabble!! How many has Gabriel Garcia-Marquez won? Toni Morrison? Although I hear Mark Twain was a fiend at Wordle.
What’s that you say, dear reader? The ability to arrange a pile of letters into the highest possible score has nothing to do with literature, or, indeed, indicates any understanding of their meaning?
You are correct. And that is why Large Language Models are not intelligent, artificial or otherwise.
The world was taken aback last month when Google was humiliated when its shiny new Gemini image generation AI responded to the prompt “1943 German Soldiers” with this:
The world was shocked; Perfecting Equilibrium readers were not. Here’s what we said last year: Large Language Models are effectively playing Scrabble, but with words instead of letters. A Scrabble player can score 37 points for “Cyclohexylamine,” and even point to the definition – “an organic compound used to prevent corrosion in boilers” – without understanding organic chemistry or corrosion or how boilers work.
A Large Language Model is like giving a prompt to a Scrabble player with an endless bag of tiles. You prompt it with “immune,” worth 10 points. It considers “immunize” for 21 points, “immunized” for 23, then moves to “hyperimmunized” for 36 before settling on “hyperimmunizing” for 37 points. It’s working a matrix for the highest score.
Google Gemini illustrated this in the most spectacular possible way. It is using a matrix to construct the highest-scoring answer. So it turned some German uniforms into illustrations – note that few if any of the medals or markings are accurate. Google admitted in the ensuing furor that it weights heavily for diversity; why is anyone surprised that Gemini and ran up the highest possible score by making the group 100 percent minority.
How did Gemini miss that Nazis were virulent racists who shipped non-Aryans – basically, everyone in this picture – off to camps for the horrifying Final Solution?
Because Gemini no more understands the meaning of words than our Scrabble player understands “Cyclohexylamine,” organic chemistry or corrosion or how boilers work.
It’s easy – and fun! – to dunk on Google for something so obvious. The insidious thing is that it is often less obvious, as Matt Taibi learned the hard way last week.
I made the mistake of asking, “What are some controversies involving Matt Taibbi?”
Gemini’s response: Matt Taibbi faced criticism for an article he wrote for Rolling Stone magazine titled “The Great California Water Heist.” The article alleged a connection between conservative donor Charles Koch and a left-wing activist group called the “Compton Pledge.”
However, investigations by other journalists and fact-checkers later revealed that this connection was inaccurate…Following the controversy, Taibbi acknowledged the error and issued a correction on his personal website.
None of this happened! Though it sounds vaguely like a headline for an article I might have written, there was never a Rolling Stone piece called “The Great California Water Heist,” and I’d never heard of the “Compton Pledge.”
Forget it, Matt. It’s GoogleTown.
While we’re at it, let’s follow up on another story from the fall: Commercial Real Estate Apocalypto. Perhaps that headline seemed somewhat unhinged at the time, so let’s see what the guy whose job it is to prevent economic panic had to say about it this week: “There will be bank failures,” Federal Reserve Chair Jerome Powell told the Senate Banking Committee Thursday.
I feel better now; don’t you?
“I think it’s manageable, is the word I would use,” Powell said.
Well, no worries then.
Unfortunately for Powell, this guy is still worried.
There will be a bifurcation…The product in a good location with a good, safe environment will recover. And then you've got another group that will somehow hang in there and get reset in pricing. And then you have the others that are basically worth nothing—the D class. Those just have to be torn down. That's probably at least 30% of all offices in the country.
That’s real estate analyst Fred Cordova, of Santa Monica-based Corion Enterprises.
But let’s let Powell have the last word.
In many cities, the downtown office district is very underpopulated. There are empty buildings in many major and minor cities. It also means that all the retail that was there to serve those thousands and thousands of people who work in those buildings, they’re under pressure, too.
At least 30 percent of office space needs to just be demolished. Downtowns are very underpopulated; and that leaves retail under pressure, too.
There will be bank failures, but it’s manageable. And that’s the take from the guy whose job it is to PREVENT panic!
Somehow, I am not reassured.
Next on Perfecting Equilibrium
Tuesday March 12th - The PE Digest: The Week in Review and Easter Egg roundup
Thursday March 14th - The PE Vlog: We’re taking a couple of weeks developing marketing graphics for Feola Factory as an exercise to understand how and when AI tools are useful. This week we’ll look at the final collateral.
Friday March 15th - Foto.Feola.Friday
Sunday March 17th — While the screamers shout, America quietly works just fine. It’s fashionable to talk about how American government no longer works, and to point out the latest Washington shenanigans. Perhaps. But all over America Americans are quietly and calmly governing themselves. Here’s a little story about that. Last month Ft. Worth asked 72 people to show up for jury duty. All 72 showed up. Six times. And were smart and quiet and extremely respectful discussing an extraordinarily distressing case.