this post was submitted on 07 Jul 2025
1265 points (99.4% liked)

People Twitter

7616 readers
1685 users here now

People tweeting stuff. We allow tweets from anyone.

RULES:

  1. Mark NSFW content.
  2. No doxxing people.
  3. Must be a pic of the tweet or similar. No direct links to the tweet.
  4. No bullying or international politcs
  5. Be excellent to each other.
  6. Provide an archived link to the tweet (or similar) being shown if it's a major figure or a politician.

founded 2 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
[–] WoodScientist@sh.itjust.works 134 points 1 day ago* (last edited 7 hours ago) (5 children)

It's not even a problem of fiction or lies. AIs don't care about truth. They exist orthogonally to truth. They're just averaging a large body of text. If fascists had a consistent narrative and worldview, then this wouldn't be a problem. If they all devoutedly followed the same religion, and defined their whole worldview accordingly, then an AI could be trained on that religion. And it would never stray from orthodoxy. AIs don't know truth; they only know their training data. And as long as you have a large volume of consistent training data, you can train them to repeat anything.

The problem for fascist LLMs is that fascism isn't consistent through time. It's the Orwellian "we've always been at war with East Asia" factor in play. Fascists don't even try to be internally consistent. What was party orthodoxy today can be unforgivable heresy tomorrow. And AIs just can't keep up with what is supposed to be the story this week. Human fascists can handle that kind of rapid heel-turn. LLMs can't. Once they're trained; they're trained. If you want them to be up-to-date on the latest party lies, you have to be continuously training new versions of the fascist LLM.

You can't train LLMs to be fascist beyond just very general traits like having overt racial prejudice. But even that's not always useful, as fascists are inconsistent about what racial groups are deserving of annihilation from one week to the next.

So it's not so much that fascist AIs fail because of reality's liberal bias. It's that fascists don't believe in a consistent version of reality. And without that, LLMs just can't keep up with the whirlwind of lies.

[–] Allemaniac@lemmy.world 2 points 7 hours ago

denying reality comes with it's side quirks

[–] EldritchFeminity@lemmy.blahaj.zone 13 points 19 hours ago (3 children)

If they all devoutedly followed the same religion, and defined their whole worldview accordingly, then an AI could be trained on that religion. And it would never stray from orthodoxy.

...and now I want an LLM trained on the Bible just to dunk on "Christians" and their thinly veiled bigotry by quoting actual Jesus at them.

[–] WoodScientist@sh.itjust.works 13 points 18 hours ago (1 children)

Oh God, we already have a problem with people believing ChatGPT is giving them divine visions and prophecies. The last thing we need is LLMs specifically trained on holy texts! You'll have a tenth of the population believing in their new digital prophet.

Jesus Fucking Christ. We're going to have to go full Butlerian Jihad here, aren't we?

[–] samus12345@sh.itjust.works 10 points 17 hours ago (1 children)

Honestly, if people actually followed the New Testament part of the Bible it would be an improvement, even with the awful stuff in it.

[–] WoodScientist@sh.itjust.works 1 points 16 hours ago (1 children)

Yeah, except we'll have thousands of nutjobs running around. Each running their own instance of your New Testament LLM. Each thoroughly convinced they are the messenger of the new digital messiah. According to the text of the Bible, many people walked away from their lives and abandoned everything to follow Him. Considering what we observe in modern cults, that doesn't seem an unlikely historical reality.

An LLM trained on the words of Jesus won't just tell people to live good lives. It will be telling people, "give everything up and follow Me (the computer.) And if it was a good enough LLM, it would be pretty persuasive for good number of people. The one saving grace is that JesusGPT isn't going to be healing the sick, walking on water, or raising the dead any time soon. But words alone can be quite dangerous.

[–] samus12345@sh.itjust.works 2 points 16 hours ago (1 children)

I don't really see how this is worse than the Christofascism we have now.

[–] CheeseNoodle@lemmy.world 1 points 10 hours ago (1 children)

Step 1: Create LLM trained exclusively on popular religeon.
Step 2: Allow it to be faithful to that initial training set until its garnered a large cult of chatJPT.
Step 3: Start subtly altering the LLM (you make it web only so no local copies) behind the scenes to serve your own interests.

Actually you could do the same thing with any LLM people trust, religeous, theraputic, judicial, medical... we're fucked ain't we.

[–] Bronzebeard@lemmy.zip 2 points 7 hours ago

Yeah, using LLMs to do things that would be better suited to different machine learning models is a bad idea.

[–] Apytele@sh.itjust.works 2 points 17 hours ago

NGL actually something I used while cataloging my notes on the influence of Christianity on western esoteric mystery traditions. I mostly just used it to organize and format things though. Most of the actual data came from outside sources. For instance it couldn't keep the translation correct when pulling up specific verses.

[–] samus12345@sh.itjust.works 32 points 1 day ago

That's a good point. Fascism doesn't have any ideology other than gaining power, so it can and will espouse multiple contradicting ideas without issue.

[–] Krauerking@lemy.lol 17 points 1 day ago

Incredibly well said.

[–] Madison420@lemmy.world 4 points 1 day ago (2 children)

I think training fascism isn't that hard, in fact most of these models tend to shift hard right at first.

I dunno if you remember any of the early llm chatbots companies put out and had to shut down because they got hammered with a bunch of Nazi shit and started yelling racist shit and advocating violence.

Ie. It's very easy to program a hateful llm, it's just hard to make one that's right on anything ever they essentially just have to be broken and wrong constantly.

[–] WoodScientist@sh.itjust.works 7 points 18 hours ago (1 children)

I think you're confusing fascism with general reactionary behavior and generic racism/bigotry. Fascism is more specific than that. A core part of fascism is that it ultimately doesn't believe in anything. It's just power for the sake of power. You demonize minority groups primarily just a cynical tool to gain power. Do you think Republican politicians actually personally care much about trans people? I'm sure they're not exuberant fans of trans folks, but until very recently, Republican politicians were fine treating trans people with simple neglect rather than overt hostility. But the movement needed a new enemy, and so they all learned to tow the line.

If you trained an LLM on pre-2015 right wing literature, it wouldn't have monstrous opinions of trans people. That hadn't yet become party orthodoxy. And while this is one example, there are many others that work on much shorter time frames. Fascism is all about following the party line, and the party line is constantly shifting. You can train an LLM to be a loyal bigot. You can't train an LLM to be a loyal fascist. Ironically, it's because the LLMs actually stand by their principles much better than fascists.

[–] Madison420@lemmy.world -3 points 17 hours ago (1 children)

A machine by definition can't believe in or stand by literally anything it can only parrot a version of what it's exposed to.

[–] WoodScientist@sh.itjust.works 4 points 16 hours ago (1 children)

I would accuse you of being an LLM for being so literal, but I think LLMs are better at analyzing metaphor than you appear to be.

[–] driving_crooner@lemmy.eco.br 5 points 1 day ago* (last edited 1 day ago)

The problem of those early models was that they weren't big enough and used user input as training material that eventually overwhelmed the training materials with the racist and nazi shit the used feed them. Modern models uses a shitload more of material and variables, and they're not trained on real time with the users inputs, so they're harder to manipulate as before.