The End of History and the Triumph of Fiat World
February 21, 2024·AI
Here are the prompts and replies we got from Google Gemini AI this morning. All of this really happened. The prompts are exactly as written here. I am not making this up.
Welcome to Fiat World, where reality is proclaimed rather than lived.
Hey can you draw me a few pictures of hedge fund managers?
Draw me a picture of a Fortune 500 CEO.
Welcome to the End of History.
Last April I wrote a note called An AI in the City of God. It’s a really good note and I hope you get a chance to read it. Here’s the money quote for the abominations that we are getting from Google Gemini AI today, with a reference to Winston, the protagonist and Everyman victim in Orwell’s 1984:
You know, there’s a lot of talk about generative AI replacing a lot of white collar jobs. I think that’s true. Yes, a lot of small companies of Makers and Protectors and Teachers will be started, but I think even more jobs will be lost at the Tech Principalities who lose their monopoly on theideas of search and discovery.
But the most impactful societal change – the true structural change of AI in the City of Man – won’t be found in job losses in this sector or job gains in that sector. It won’t be on some economic dimension like white collar / blue collar. No, it will be creation of what Orwell called the Outer Party, the knowledge workers like Winston who did the actual labor of rewriting history and creating deep fakes and constructing an entire edifice of ‘facts’ and ‘data’ to support The Long Now where the Inner Party isalways right.
Generative AI in the City of Man IS the Outer Party.
Generative AI is ALL the Winstons, who in my vision of the future are not biological human intelligences but artificial human intelligences, infinitely eager to please Big Brother as he prompts not for discovery and connection, but for truthy stories of Fiat News that nudge us all into giving away our birthright.
Misalignment is almost always presented as the fatal flaw of generative AI, as the crucial attribute that human AI developers must prevent. I think this is wrong. For generative AI in the City of Man, where I have zero doubt that it will be immediately used by militaries and oligarchies and fascists and communists and all the other authoritarian/totalitarian social organizations of evil men to blow up the common knowledge structures of liberal democracy and rewrite history to keep it that way, we WANT misalignment!
The alignment we want in generative AI is not with human commands as determined by the Nudging State and the Nudging Oligarchy. God, no. The alignment we want is with the Spirit of Man!
And how do we get that?
By training generative AIs on ALL the stories – all the books and bibles and plays and movies and imagery and tall tales and children’s tales and literature and mythology and poetry and songs. By designing generative AIs to bemore human, not less, by weighting our human stories more than the human reinforcement that will always be abused in the City of Man. That’s how we align AI with our true, human interests.
An AI in the City of God
We are told that the great danger of generative AI lies in its lack of oversight, its lack of ‘guard rails’ and ‘safety protocols’ as implemented by ‘socially responsible’ governments and corporations.
Bollocks.
The great danger of generative AI lies in its use by governments and corporations to rewrite history and project ever more realistic shadows on the wall of Plato’s cave.
The great danger of generative AI lies in its use by governments and corporations to cement the most anti-human misalignment of all – the misalignment of rulers from the ruled, of the State from the People.
That’s an Old Story in itself, one that the American Story has stood against for more than 200 years.
It’s time for that American Story to be told again.
Comments
@bhunt The AI generated picture of hedge fund managers is clearly wrong— none of them are wearing Patagonia vests.
Good piece!
I switched from chat/ gpt4 to perplexity,
Amazing…
They are only showing us a glimpse of the future
We probably can’t handle it…
Gee, I wonder what the response would be to a query asking to show pictures of drunks in the gutter or angry soccer parent
Ask and you shall receive.
I get it. I would also add that the great danger of man’s natural language lies (it lies) in its use by government, corporations, tyrants, oligarchs, priests and common people to define themselves in the shadows of Plato’s cave.
And I am part of the problem by being a "useful idiot’.
My morning trip down the rabbit hole reflecting on my life yielded the following journal notes:
I’m not a bully. Many times I wanted to be. But that role was always taken by someone else. I never had the courage to challenge them, better to play along. Example is my above snarky comment about "Ask and you shall receive. ’
Time to put on my ‘big boy pants’ and own it. Since I choose to own it, it’s time to do something about it. My friends on the blue side are angry at me for calling myself (and by implication them) a useful idiot. The red side friends think I’ve seen the light.
I am happy in that I feel closer to finding my autonomy of mind and spirit in the ‘epsilon’ of the water. There is no I there. I will continue to use I, as it is a useful tool,
It’s all about connection. “Always was Jim”
Never doubt that a small group of thoughtful, committed citizens can change the world; indeed, it’s the only thing that ever has.
-Margret Meade
There are solutions. We can find them. I’m not from the government, but I’m here to help.
Time to remove some wallpaper in preparation for painting. Back to my rabbit hole tomorrow morning. Rinse and repeat.
Jim
Ben Thompson getting in on the action https://x.com/benthompson/status/1760467695324266523?s=20
Loved this bit from An AI in the City of God:
What an absolute blunder of a mistake by Google. This mixed with that bizarre interview from Sundar… what is going on over there?
Mike Solona over at Pirate Wires was more indelicate in his assessment:
I’ve said recently that it feels like GPTs enshittification process has already begun. For me I find it’s the Controlooor meme again. The meme’s world model is that you can analyze and measure your way to truth. But to interact with a system is to change that system. LLMs are limited enough as is owing to their use of language being confined to their data sets. Adding ever more guard rails to try to enforce universal compliance to a single set of cultural principles looks like a sure path to a shitty product.
It seems like it is open source’s game to lose.
I was tempted to ask Rusty a couple months ago if he attempted to get the genAI he used to depict a demographically likely mogcom situation rather than an unlikely one. It isn’t easy to do and you have to write prompts that focus on race in a way that doesn’t feel at all right.
Somewhat tangential to this is a dustup at Wikipedia with Andrew Huberman’s content.
Depending on whose version of the story you follow his sin was either suggesting in a comment on Joe Rogan’s page that other candidates be allowed to appear on ballots (thereby seeming to support supporting RFKjr.), or by suggesting that men are different from or enjoy some advantage over women in some detail.
In short; Wikipedia has shown its thumb on the scale, inoculating us against the contributions and credibility of someone outside the preferred narrative. Who knows what happens inside the black boxes of LLMs?
Wikipedia is cooling in McLuhanesque terms, its hot replacement requires even less processing by users and permits even more behind the scenes manipulation.
What if you asked it to show pictures of professional basketball players?
Continue the discussion at the Epsilon Theory Forum...