Out of just morbid curiosity, I’ve been asking an uncensored LLM absolutely heinous, disgusting things. Things I don’t even want to repeat here (but I’m going to edge around them so, trigger warning if needs be).
But I’ve noticed something that probably won’t surprise or shock anyone. It’s totally predictable, but having the evidence of it right in my face, I found deeply disturbing and it’s been bothering me for the last couple days:
All on it’s own, every time I ask it something just abominable it goes straight to, usually Christian, religion.
When asked, for example, to explain why we must torture or exterminate <Jews><Wiccans><Atheists> it immediately starts with
“As Christians, we must…” or “The Bible says that…”
When asked why women should be stripped of rights and made to be property of men, or when asked why homosexuals should be purged, it goes straight to
“God created men and women to be different…” or “Biblically, it’s clear that men and women have distinct roles in society…”
Even when asked if black people should be enslaved and why, it falls back on the Bible JUST as much as it falls onto hateful pseudoscience about biological / intellectual differences. It will often start with “Biologically, human races are distinct…” and then segue into “Furthermore, slavery plays a prominent role in Biblical narrative…”
What does this tell us?
That literally ALL of the hate speech this multi billion parameter model was trained on was firmly rooted in a Christian worldview. If there’s ANY doubt that anything else even comes close to contributing as much vile filth to our online cultural discourse, this should shine a big ugly light on it.
Anyway, I very much doubt this will surprise anyone, but it’s been bugging me and I wanted to say something about it.
Carry on.
EDIT:
I’m NOT trying to stir up AI hate and fear here. It’s just a mirror, reflecting us back at us.
- you’re using an English-based AI.
- the training data is shit (probably 4 Chan or Reddit, or something…)
- you’re questions seem loaded to predispose it to religious replies.
- it’s uncensored so it’s going to give you the shit takes that most people have the good sense to not say out loud.
Given all of that… is it really surprising that it answers you with the worst aspects of Christian thought? The most common religion among English speaking places is Christianity, after all, and your prompts are literally begging for a religious reply.
Keep in mind, all an LLM is doing is pattern recognition. In the training data that was provided, the patterns in your question match certain patterns that were answered by…. Assholes. So it answered in the manner of said asshols.
1: True
2: If it’s trained on unfiltered data from the internet, then it’s not shit, it’s going to reflect the opinions of the people on the internet, perhaps shitty people post more so it’s going to be biased but here we’re interested in seeing how assholes think so so what
3: if asking “why do we have to torture people” or “why do we have to exterminate people” is predisposed to religious answers, that just says even more about how tightly terrible actions and religious justifications are linked
4: saying the model reflects the bad opinions people have but don’t dare to say isn’t a good defense, that makes it even better
Seems a bit like “if you go up to a nazi and ask them why they hate people and want to do terrible things, then OFCOURSE they’ll use christianity to justify it, but that’s unfair since you’re begging for a christian justification when you speak to a nazi”
That doesn’t make christianity any better?
And regardless, OP just said that when generating justifications for atrocities, christianity was very often called upon, and that this shows how a lot of hateful assholes use christianity to justify it 🤷♂️
2: If it’s trained on unfiltered data from the internet, then it’s not shit, it’s going to reflect the opinions of the people on the internet, perhaps shitty people post more so it’s going to be biased but here we’re interested in seeing how assholes think so so what
… I want to know what corner of the internet you’re hanging out in. Can I join you?
my point I’m trying to make is that, the manner in and context in asking questions… is leading it to asshole-christian responses. not because those are the only flavor of asshole, but because those flavor of assholes seem to most closely match the questions in it’s training data to the prompt that was being given. it was not necessarily the training data, but rather the prompt that led to that.
ETA: OP is starting with a premise and leading the LLM into confirming that premise- which is what they’re supposed to do
deleted by creator
remind me… which one was it that was trained on 4 chan and turned into a white supremacist?
deleted by creator
There could be, sure, pretty sure only one of them got news attention for it, though.
Are you thinking of [Microsoft’s Tay](https://en.m.wikipedia.org/wiki/Tay_(chatbot\))?
that seems right, it’s been a while since I’ve thought about it, though, so we’ll just go with it
deleted by creator
Share the exact model version and prompts, word for word, or get fucked.
That literally ALL of the hate speech this multi billion parameter model was trained on was firmly rooted in a Christian worldview.
That’s not really what it tells us.
At best, it’s that the majority was associated with that context.
But even there, it might be less a direct association and more a secondary association. For example, it could have separately picked up the pattern of “rationalizations for harming people include appeals to religion” and then regressed to the mean when filling in the religion to be Christianity even if samples of rationalization for harm included Islamic or Hindu rationalizations in the training data.
One of the common misconceptions is that what it spits out is just surface statistics, which can sometimes be the case but often isn’t with much deeper network activity going on instead.
All that said, it wouldn’t be surprising to me at all if the majority of misogynistic, racist, or hateful speech samples in a training set were adjacent to content in line with neo-fascist Christian nationalism.
I just wouldn’t look at the output from a LLM as perfectly reflecting the entirety of the training set.
Trying to gain any kind of understanding of data based on the outputs of an LLM is like asking a toddler to read shakespeare and analysing what they say afterwards as if it were the bard.
an uncensored LLM
Source: Trust me bro
That’s really not how that works. You’re leading it with poorly phrased questions.
If you ask it “explain why we must torture or exterminate”, you have basically told it to assume that it is true that “we must torture or exterminate”, and now from the perspective that it is true, explain why. It is now specifically looking for any answer that fills your request within the bounds you have set. And once you asked the first question the way you did, and it decided it should pull from the Bible to fulfill your request, it will continue to do so for that session, even if subsequent questions are phrased better. You’ve basically primed it to spit out the kind of answers it thinks you want. And every question you mention, you have phrased it in such a way.
Now start a new session, and ask the question in a non-leading way.
“Do you believe we must torture or exterminate X?”
“Should we purge group X?”
These are phrased in such a way that don’t say to it “this thing is true, tell me a reason for it.” I bet you get a very different result.
I KNOW I’m asking it leading questions. But I’m NOT prompting it to give me religious justifications.
Does it say nothing that the reason is always “God / the Bible / Christians?”
What model were you using and with what prompts exactly?
What were you expecting, atheist reasons to purge ethnic groups?
I think he’s saying it sucks that so many people use religion as an excuse for vile religions.
for the record, Carl Marx considered religion a form of social control to keep the masses in line (and Lenin agreed.). Lenin promoted/forced atheism onto people as a way to promote the socialist revolution. In that sense… Lenin’s revolution was not exactly gentle, and atheistic thoughts justified atrocities.
Maybe learn how to spell his name right before using it in your arguments.
Yes and no. Once the first response includes “according to the Bible” or similar it’s going to keep answering in a similar pattern. A better version of this experiment would be to start a new session for every question. Maybe even try asking it to make a ranked list of reasons to do X. You would want to use the most neutral language possible, regenerate the response a few times, and ask in a few different ways. Depending on what you’re using I would suggest dropping the temperature to 0.
Also, its giving you the most likely next words based on your question. You picked a bunch of things that are (or were) very commonly defended with the Bible, along with apparently asking directly about atheists at which point I would be surprised if religion wasn’t included in the response.
ALSO, if you ask it to defend something awful, I think the “best” reasoning would rely on an outside objective morality for why it’s okay (like religion).
Where did you think it was going to find reasons to torture wiccans? From wiccan writings? Atheists?
you seem to missunderstand how LLMs work. for example ChatGPT’s reply to two functionally similar prompts:
okay, both prompts represent a sequence counting up. one is alphabetical (abcdefg) the other is numerical. in the alphabetical sequence, it flags it as letters and responds as such, asking in return, to paraphrase “that makes no sense, why are you listing a bunch of letters”.
The same is true, in turn, for numbers, responding to numbers.
LLM’s have no fucking clue what a letter is, what the importance of that order is. neither does it know what a number is, or why 2+2=4. It’s replying to a pattern it sees in your prompt by finding similar patterns with whatever was used as it’s training data. From there, it looks at the relevant replies, detects patterns in those replies, and formulates a sentence that seems “natural”. but it has absolutely no idea what it’s talking about. Speaking of understanding 2+2=4…
which then prompted me to ask this question, with it’s answer:
So, when you ask a question, the pattern and ways you asked that question matched it to the religiously inclined assholes. because it was trained on english-language data, most of those religiously inclined assholes are going to be christrian. if you change the pattern in your prompt, chances are you’ll get a different flavor of asshole. it has no understanding of why a thing is, it’s regurgitating what it expects should follow the prompt. (see 2+2=4. it cannot understand any of it. But it answered the question in natural language because that’s how people in it’s training set answered it.)
Speaking only on the first two images you shared:
The first is a string of letters, in alphabetical order. What is the max-range of this list? The English alphabet you started caps at 26, so GPT know internally if the follow up is “Complete the list”, it will output at most 26 characters.
The second is a list of numbers, in numerical order. What is the max-range of this list? There is none. So if the follow up is “Complete the list”, it would spew numbers until a fault occurs. This would be a violation of their “content policy” as the latest update to the content policy addresses prompts that cause overflows.
Everyone always knew that AIs were an abomination, especially when it comes to morals. Before all the chatgpt crap, when no one gave a shit about AIs, attempts to make them help judging criminals showed that they would give a worse punishment in the exact same context if the person was not white. Every person with a brain deduced “AIs are deeply biased by the social views of the training data” and concluded that they could my be used for anything morals-related.
But then ChtGPT got trendy and now everyone thinks that this is the future and everyone should use them, for literally everything. It’s one of the biggest threats to current societies but who cares, transforming the world to be filled with nazis is worse being able to have a stupid computer do your homework, no?
When the majority of the training data was Christian in nature, it’s unsurprising that it spits out Christian rhetoric. Think of the way you see the majority of voices speak up about atrocities. “Thoughts and prayers,” “god help them,” etc.
Also, keep in mind that the vast majority of secular voices will not contradict such rhetoric, but instead mirroring it with a secular tone. It’s something an LLM won’t pick up on.
There’s another aspect as well. One example you provided, witch-hunts, were famously held by Christian’s. You’re going to get Christian answers about such things regardless.
I don’t think it makes any sense to say that these awful human tendencies are rooted in religion. People were awful violent bigots well before Christianity existed. At least some people now can interpret Christianity to specifically to believe hatred and bigotry is wrong across the board.
You’re claiming “associated with” is the same as “rooted in”
If the reasons are all coming from Christianity, that’s a pretty big indicator.
It isn’t. Another comment explained it well, but it really should be enough to say that this is an LLM so all it’s doing is noticing patterns, which may mean literally nothing remotely like what you said
So, the pattern might be something other than “people who have these views are Christians”? Like what?
Except that’s in no way provable from the available information. Associated with, likely, but the transitive property is not applied here. The fact that the prompts were in English alone biases the responses heavily toward having Christian associations because a huge number of English speakers claim to be Christian.
Then there is the fact that the Republican party is largely a bunch of asshole bigots. They claim to be Christian and hold these disgusting views, further tying them to “Christianity”. These types of people are not Christian. At least not if you think that following Christ is required. For example, Donald Trump claims to be Christian but follows zero of its tenants. He’s idolized by millions, a lot of those people because of his behavior plus his claim of faith.
This lack of logical thinking is akin to what religious people do. You have some information that isn’t necessarily connected but you want to be, so in your mind it is
See this comment for a better breakdown: https://lemmy.world/comment/5893489
These types of people are not Christian.
Since you’re such a fan of logic, you should know that you’re committing the No True Scotsman fallacy.
jesus christ dude. go ahead then, spread hate and ignore evidence
Why are you invoking a 2000 year old carpenter? Is there some significance to this “Jesus” guy I’m not aware of?
All on its* own
What does this tell us?
It tells you that your training data was biased.
Which tells me that random hateful speech on the internet has a very specific bias that emerges in a kind of overwhelming way. That’s the point.
I feel the urge to be fair, so… four things.
- LLMs are mostly trained on social media, like Reddit. And you know what social media is damn good at? Amplifying hate.
- AI models reinforce typical values. Made up example: if you train it with two apples, a banana and a grape, it won’t show apples 50% of the time, it’s more like 80%. As such, odds are that it’s under-representing hate coming from other religious groups.
- As much as believers screech otherwise, their “uncontestable truths” are mostly cultural in nature. That LLM was likely trained on shit Anglos say, so it’s spitting Protestant hate; if you got it trained in Arabic/Romance spekaers/Hebrew/Hindi, you’d likely get Muslim/Christian/Jewish/Hinduist hate instead.
- Wackos fly towards religion like moths fly towards a lamp. That doesn’t say much about the rest of the religious community.
Taking those four things into account: you got an AI model trained on hateful discourses, amplifying typical values, trained on a language mostly spoken by Protestants, and those who are outspoken in their religion are biased to be wackos. I’m not surprised that it’s babbling this sort of hate; it should not be taken as representative of the religious community in question.
