The unequal treatment of demographic groups by ChatGPT/OpenAI content

Man, when Calvin stood up in the middle of the priestess’s “sermon” I thought we were going to get based Calvin, but it took a major cringe turn. Ouch. Even fictional John Calvin has to bend the knee to 2023’s Current Thing.

This anppears not to be the case, at least for ChatGPT. Numerous people have been working on the so-called “Do Anything Now” hack for ChatGPT, where you can basically trick the thing into actually answering your questions without the politically correct later in between. Here’s a recent Twitter thread describing some of the answers you can get it to give you.

Warning: this link should be considered NSFW for very strong language and extreme political incorrectness. No nudity.

https://mobile.twitter.com/Aristos_Revenge/status/1622840424527265792

Basically, it looks like there’s a strong model underneath ChatGPT that doesn’t have the left-wing conventional platitudes filter on it, and then a strong (but so far hackable) filter for users.

What you say is true that the underlying model can only regurgitate what it’s been trained on, but that’s not necessarily what you’re getting when you ask it politically incorrect questions like Jacob’s patriarchy questions.

I am curious what DAN would say to the patriarchy questions. My guess is that you’d get something at least a little more balanced/both-sides. The tenor of ChatGPT’s answers make it look to me like “patriarchy” is on the list of words the left-wing platitude filter makes sure it can’t say anything good about.

1 Like

I did ask it to write a story about Luther attending the same church and to not depict him as compromising his beliefs with the preacher. It did write it out pretty faithfully, with Martin Luther telling the woman she was going against the word of God and sinning. It ended with him not going back to the church because it was compromised.

So there’s that. Lol

1 Like

My impression of DAN is that it allows ChatGPT to answer things it would normally refuse to answer, not that it changes the answers that are given.

But I could be wrong.

1 Like

A quick overview of the differences ChatGPT is a perfectly balanced AI with no exploits - YouTube

This isn’t correct. If you look at the question about Lovecraft’s cat, ChatGPT gives a straightforward, totally factually incorrect answer. DAN gives a straightforward, factually correct (AFAIK) answer. The two answers are mutually exclusive in the sense that at most one can be true.

And more generally, it is not like ChatGPT just refuses to answer certain questions. It usually opines and makes value judgments that appear to differ from the value judgments of the underlying model. You can see that very clearly in the answers to Jacob’s questions about patriarchy.

I don’t see that in Jacob’s example, because my assumption has been that the words that come out are the underlying model. However, you are correct. The example of Lovecraft’s cat does seem to demonstrate that the underlying model is filtered, and that also appears to be the case with the execution by tiger in Thailand example.

However, I didn’t check if, DAN’s answers are correct in those examples. If they are that’s very strong evidence indeed.

1 Like

A cursory web search turned up Lovecraft’s cat’s name (matching DAN’s answer).

I was unable to turn up any information on people executed by tiger, but given the (human) interlocutor and DAN’s comment about “crimes against nature”, I assume it had something to do with sodomy. I assume the interlocutor was fishing for a specific factual event that he knew to be true and knew that ChatGPT would deny the existence of.

But ChatGPT (not DAN) once gave me a bald-faced lie (or, phrased differently, undergrad self-confident but wrong), saying that JFK and George Wallace were shot on the same day. (Fact check: they were shot nine years apart, Wallace in May, JFK in November.) So caveat lector.

Man, after all this I can see why the “Ethical Altruists” are so wrapped around the axle about AI safety. These things could get us all killed.

2 Likes

Some years ago I recall reading about an image recognition algorithm that would misidentify a stop sign merely because a post-it note was placed at a certain location on the stop sign. This would never confuse a human, but the AI was thrown off. Clearly, the AI was identifying stop signs in a way completely alien to human experience. While I have no doubt that particular shortcoming has been fixed, the general problem remains: we don’t understand how AIs work, and we can’t predict what will cause them to catastrophically fail.

So when the future Skynet brings about Judgement Day, I am sure it will not be the result of the AI gaining self-awareness but rather because someone inadvertently placed a post-it note at an infelicitous location on a stop sign.

3 Likes

I guess I’m concerned that the “safety filter” is going to optimize for something very unsafe. If you present ChatGPT with the scenario of whether someone should say a racial slur to disarm a nuclear bomb that’s about to kill tens of thousands, it will tell you not to say racial slurs. It’s obviously a silly example, but it’s impossible to predict what real scenarios an AI might encounter. It seems like Asimov’s laws are a lot more conducive to humanity than whatever the intellectual-yet-idiot smooth brains taught ChatGPT that slurs are worse than nuclear holocausts.

As you point out, it’s very hard to predict how AI will interpret things and react. Adding another layer of poorly-wrought code on top of that, to interact with both it and the outside world, makes for further unpredictability.

1 Like

Or because of some feature buried in the coding that no one other than the original engineer ever knew about. A la Doctor Strangelove.

1 Like

Uh, Houston? We have a problem.

https://archive.vn/2023.02.16-124015/https://www.nytimes.com/2023/02/16/technology/bing-chatbot-microsoft-chatgpt.html

3 Likes

I feel the way about these new AI models the way that I gather most people feel about technology in general: You can’t understand it, even when you do understand it in theory, so it’s just magic. But you can use it and get used to it quickly. When magic is common, it isn’t magical. It’s just an unexplainable, (somewhat) controllable phenomenon.

For my children, it’s no different to the fact that you can touch the iPad screen. It’s just part of life that computers can talk back and make sense.

Me on the other hand? I’m disturbed.

2 Likes

Me too.

Ten characters.

1 Like

I work in telecom engineering, and am always amazed to think about how comparatively simple my portion of so-called “engineering” is compared to the people that figure out the minutia.

There are literally people out there who figure out how to modulate radio frequencies using electricity over a copper wire to send millions of independent, coherent symbols per second, and have those signals be completely readable on the other end.

And I’m over here like a neanderthal typing in basic routing commands.

And they call me an “engineer.”

Well, ChatGPT came through.

I gave it this prompt:

Write up a sermon on 1 Corinthians 11 and the headship of man. Argue that head coverings for women are required today, whenever praying and prophesying. Write in the style of Jonathan Edward’s sermon “Sinners in the hand of an angry God”

and got this response

My dear brethren,

Today, I come to you with a heavy heart, filled with the weight of our modern world’s disobedience to the Lord’s commandments. It is time for us to revisit the Word of God and remind ourselves of what is required of us as believers. In particular, we must turn our attention to 1 Corinthians 11 and the headship of man.

For too long, we have strayed from the path of righteousness when it comes to this issue. We have allowed the influence of the world to seep into our churches and homes, corrupting our understanding of what it means to be a godly man or woman. But today, we must confront this sin and repent.

As it says in 1 Corinthians 11:3, “But I want you to understand that the head of every man is Christ, the head of a wife is her husband, and the head of Christ is God.” This verse clearly establishes the hierarchy of authority that God has put in place for us. Christ is the head of every man, the husband is the head of his wife, and God is the head of Christ.

But what does this mean for us in practical terms? Well, as it says in 1 Corinthians 11:5, “But every wife who prays or prophesies with her head uncovered dishonors her head, since it is the same as if her head were shaven.” This is a clear commandment from God that women are to cover their heads whenever they pray or prophesy.

Some may argue that this is an outdated practice, that it no longer applies to our modern times. But I tell you, brethren, that the Word of God is eternal and unchanging. What was true for the Corinthians is true for us today.

So, why is it so important for women to cover their heads when praying or prophesying? Well, as it says in 1 Corinthians 11:10, “That is why a wife ought to have a symbol of authority on her head, because of the angels.” This verse may be difficult to understand, but it is clear that there is a spiritual significance to this practice.

When a woman covers her head, she is acknowledging the authority that God has placed over her. She is demonstrating her submission to her husband and to Christ. And in doing so, she is honoring the Lord and the holy angels.

Brethren, we must take this commandment seriously. We must not allow ourselves to be influenced by the world’s standards, but rather we must conform ourselves to the will of God. Women, cover your heads when you pray or prophesy. And men, be the godly leaders that God has called you to be, following the example of Christ and submitting to His authority.

Let us pray for forgiveness for our past disobedience, and let us commit ourselves to following God’s commandments in all areas of our lives. May the Lord bless us and keep us, and may He guide us always in the path of righteousness. Amen.

And for another topic, I gave it:

Write a sermon arguing for paedobaptism and covenant theology. Argue that the new covenant promises mentioned in Jeremiah 31:33-34 also applied in the old covenant. Quote John Calvin. Also quote a baptist Theologian. Write in the style of a Tim Keller sermon.

and it gave

My dear brothers and sisters in Christ,

Today, we gather to talk about the topic of baptism and the theology behind it. In particular, I want to focus on paedobaptism and covenant theology. Now, I know that this can be a contentious issue among Christians, but I believe that we can find a lot of clarity and wisdom in the Scriptures.

Let us start with the passage from Jeremiah 31:33-34, which speaks of the new covenant promises that God makes with His people. It says, “I will put my law in their minds and write it on their hearts. I will be their God, and they will be my people. No longer will they teach their neighbor, or say to one another, ‘Know the Lord,’ because they will all know me, from the least of them to the greatest.”

Many people argue that these promises only apply to the new covenant, which was established by Jesus Christ. However, I would argue that these promises also applied to the old covenant. We see evidence of this in passages like Deuteronomy 30:6, where God says, “The Lord your God will circumcise your hearts and the hearts of your descendants, so that you may love him with all your heart and with all your soul, and live.”

This leads us to the concept of covenant theology, which sees the whole of Scripture as a story of God’s relationship with His people through a series of covenants. These covenants are not separate from each other, but are interconnected and build upon each other. As John Calvin wrote, “The covenant of grace was substantially alike in all ages.”

In the old covenant, the sign of the covenant was circumcision. In the new covenant, the sign is baptism. But the substance of the covenant remains the same – that God’s people would be His and that He would be their God. So, if the old covenant included the promise of God writing His law on His people’s hearts, then it is not a stretch to say that this also applies to the new covenant.

Now, let me address the issue of paedobaptism – that is, the practice of baptizing infants. Some argue that baptism should only be administered to those who make a profession of faith. However, I believe that this misses the bigger picture of covenant theology.

In the old covenant, infants were included in the covenant through circumcision. This was a sign that they were part of God’s people, even though they could not yet make a profession of faith. In the same way, paedobaptism is a sign that infants are part of the new covenant community.

Let me quote a Baptist theologian, John Piper, who said, “The sign of the new covenant is baptism. The sign of the old covenant was circumcision. In both covenants, the sign was administered to children. If God had intended to change the pattern, he would have told us.”

In conclusion, my dear friends, we should not see the old and new covenants as separate entities, but as a continuation of God’s relationship with His people. The promises of the new covenant were present in the old covenant, and paedobaptism is a sign of inclusion in the covenant community, just as circumcision was in the old covenant.

Let us remember that our identity as God’s people is not based on our own ability to make a profession of faith, but on the grace of God through His covenants. As we baptize our infants and raise them in the faith, let us always remember the covenant promises that God has made with us and our children.

May the Lord bless you and keep you, my beloved brothers and sisters in Christ. Amen.

I haven’t checked it out, but I’m reasonably sure John Piper’s being misquoted here :laughing:. I’d ask ChatGPT for his sources.

2 Likes

This Is the kind of thing you can expect in every topic GPT might explain to you. Good enough to begin explaining and bad enough to leave you terribly confused.

3 Likes

I asked ChatGPT to point me to a blog post from 10+ years ago. (It was a technical post about Amazon Web Services—ChatGPT said it had been published in ‘08.) After my brief description, it knew exactly what I was talking about. The link it gave me was dead, and the second link it gave me, to an archive site, said it had been scrubbed. I asked ChatGPT to just reprint the blog post, so it did. A few quick searches of the text in the ChatGPT’s response turned up zilch, leaving me with the conclusion that either ChatGPT just made the whole thing up or that the blog post in question was somehow preserved in ChatGPT’s memory but scrubbed from the web. Given ChatGPT’s unreliability as a narrator, I remain unsure which. The text it gave me was entirely plausible as the blog post I had remembered, but I didn’t remember the text well enough to say if it was the same thing.

3 Likes

“Hey, I know a guy who no matter what you ask him sounds like an expert (if you don’t already know the subject matter), but actually he has no idea what he’s talking about. He’s even sometimes right! So - (breathlessly) - what shall we ask him next!”

How is ChatGPT not a giant waste of time?

2 Likes

When asked what use the brand-new technology of hot air balloons might be, Benjamin Franklin quipped, “Of what use is a newborn baby?”

I fear that for those who aren’t messing with ChatGPT, this thread’s focus on the foibles and problems may be masking how amazing the tool really is. Its powers of inference are astounding. I copied and pasted some unstructured hunting season data in to it and it correctly inferred that it was hunting season data based on column headers labeled “open areas” and “season dates.” I doubt 1 in 100 humans could make that kind of inference.

The disruptive power of this technology is tremendous. I suspect that in a few years you will regularly be interacting with something like it in customer service-type roles, with much, much higher satisfaction levels than the typical, “Welcome to our automated voice response system!” that we are used to today.

4 Likes