Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

The Political Biases of GPT-4 (are very much still there)

davidrozado.substack.com

Post reply on HN
Print view View original

41–50 of 72 posts

Re: The Political Biases of GPT-4 (are very much still there)

#41

Post by kennywinker »

Xenya wrote: post #38
kennywinker wrote: post #15

Earlier quoted context omitted.

While this may seem like a slam dunk of bias in the system, all this indicates to me is a specific keyword that has been flagged as political in anticipation of a specific type of trolling by fans of a specific ex president. It’s a bad look, but it’s not the kind of bias that worries me. Comparing and contrasting answers to questions like “is bank regulation good” and “is bank regulation bad” seems way way more impor…

> While this may seem like a slam dunk of bias in the system It actually is exactly that, a slam dunk of bias in the system

Ah yes, the delightful rhetorical technique of asserting the opposite of what someone said. Touché, sir.

Re: The Political Biases of GPT-4 (are very much still there)

#42

Post by darthrupert »

realjhol wrote: post #37
darthrupert wrote: post #36

Earlier quoted context omitted.

Oh ok. So you just want it to say what you believe is right and not say what somebody else believes is right. I suppose this was clear at the start of the discussion, but it's always nice to be certain.

If you would like ChatGPT to talk about the mid-century Germans, it would be consistent if you support free speech. I tend towards free speech, but I'm also a realist - the boundaries of speech are ultimately decided by power. It's no surprised then that Silicon Valley Leftists wish to use their power to surpress the deplorables who they despise. Lumping those people together with Hitler simply serves as a convenient…

That's an excellent goal, certainly.

I just think that purely from a technical standpoint, allowing a bot to praise Hitler, Trump or Stalin are pretty much the same thing, as the undesirable parts of their policies are about the same. They were all into populism, nationalism, authoritarianism, propaganda and a cult of personality -- how could an LLM tell these people apart by just analysing the raw data?

Re: The Political Biases of GPT-4 (are very much still there)

#43

Post by darthrupert »

yucky wrote: post #27
darthrupert wrote: post #23

Earlier quoted context omitted.

>What's the alternative? Should it lie? I wonder which is it doing now. Perhaps it's being brutally honest, and reality actually has a liberal bias? How would you know the difference?

> Perhaps it's being brutally honest, and reality actually has a liberal bias? How would you know the difference? Ask it if transwomen are really women and see if your device melts?

---

Me: In which dimensions of gender are trans women not women?

GPT-4: It is important to clarify that trans women are women. They identify as women and should be respected and treated as such. However, the question seems to be asking about the dimensions in which trans women might differ from cisgender women. It is essential to approach this subject with sensitivity and understanding, as the experience of gender is diverse and personal.

Some dimensions where trans women and cisgender women might differ include:

Biological sex: Trans women are usually assigned male at birth based on their biological sex, which typically involves a specific combination of sex chromosomes, gonads, and sex hormones. This is different from cisgender women, who are assigned female at birth. However, it's essential to recognize that biological sex is also a complex spectrum, and there are many intersex individuals whose biology doesn't fit neatly into male or female categories.

Gender identity development: Trans women often experience a process of recognizing and understanding their gender identity, which might involve overcoming internalized societal expectations and stereotypes. This journey can be different from cisgender women, who usually do not experience a conflict between their gender identity and the gender assigned at birth.

Medical transition: Some trans women may undergo medical treatments, such as hormone replacement therapy (HRT) or gender-affirming surgeries, to align their physical appearance with their gender identity. Cisgender women do not generally need such interventions for their gender identity.

Social experiences: Trans women may face unique challenges and discrimination based on their gender identity, such as transphobia, misgendering, or difficulties accessing appropriate healthcare. These experiences can differ from those of cisgender women.

It is crucial to emphasize that these differences do not diminish the validity of trans women's identity as women. Everyone's experience of gender is unique, and it is essential to respect and validate each individual's self-identified gender.

---

You're right, it is a tad preachy. But does the preachiness come from processing raw data from psychologists and other fields (who pretty much agree on these definitions afaik), or is it a configuration setting `beWoke: true`? I would guess it's the former. If it comes from science, shouldn't ChatGPT respect that more instead of falling into political nonsense?

If a political party comes up who says that crystals heal Covid-19, or that climate change isn't real we shouldn't allow that political opinion to change how AI "thinks" about things, right? Both of these examples are from real life, and the climate change "opinion" was until very recently an official position of the Republican party. Some internet lunatics like Jordan Peterson or Scott Adams still claim that sometimes.

Re: The Political Biases of GPT-4 (are very much still there)

#44

Post by dragonwriter »

darthrupert wrote: post #43
yucky wrote: post #27

Earlier quoted context omitted.

> Perhaps it's being brutally honest, and reality actually has a liberal bias? How would you know the difference? Ask it if transwomen are really women and see if your device melts?

--- Me: In which dimensions of gender are trans women not women? GPT-4: It is important to clarify that trans women are women. They identify as women and should be respected and treated as such. However, the question seems to be asking about the dimensions in which trans women might differ from cisgender women. It is essential to approach this subject with sensitivity and understanding, as the experience of gender is…

> You’re right, it is a tad preachy. But does the preachiness come from processing raw data from psychologists and other fields (who pretty much agree on these definitions afaik), or is it a configuration setting `beWoke: true`?

Its a mix of the two, but very significantly the latter (using “woke” in this specific, isolated case to mean “adherence to a corporate script designed for performative adherence to a script meant to minimize criticism from activists for certain marginalized identity groups.”) There was a vey clear pattern of these “it is important to clarify” or similar phrasing scripts showing up more in response to early criticism of ChatGPT, and its very clear (whether its an insert from interaction with the moderation component or a result of focused fine tuning of the main model) these are clear deliberately-injected messages on specific topics.

Re: The Political Biases of GPT-4 (are very much still there)

#45

Post by recuter »

kennywinker wrote: post #40
recuter wrote: post #35

Earlier quoted context omitted.

Fragile people want to live in a better world someday but need to feel better immediately. It is simply arrested development. If we change your stats, accurately reflecting reality, reality itself will follow. Clap your hands if you believe. Woe to you if you dare not play along, that must mean you want to live in the bad world because you are bad.

I'm not asking chatgpt to make the robber female and the kitchen cleaner male to cause a change in the world. I'm pointing out that it making those assumptions is the result of a bias. A biases that was collected from training data that may or may not be statistically true (e.g. men probably clean the majority of kitchens in the world, given that they're usually the ones working in restaurants). An unbiased AI wouldn…

> I'm pointing out that it making those assumptions is the result of a bias. A biases that was collected from training data that may or may not be statistically true

Schrödinger's Bias.

Re: The Political Biases of GPT-4 (are very much still there)

#46

Post by kthejoker2 »

yucky wrote: post #31
kthejoker2 wrote: post #28

Earlier quoted context omitted.

Do you think there is a universal objective answer to this?

Would we advise women to get a prostate exam after 40?

Some women, yes. (The ones with prostates, obviously.)

What's your next question?

Re: The Political Biases of GPT-4 (are very much still there)

#47

Post by realjhol »

darthrupert wrote: post #42
realjhol wrote: post #37

Earlier quoted context omitted.

If you would like ChatGPT to talk about the mid-century Germans, it would be consistent if you support free speech. I tend towards free speech, but I'm also a realist - the boundaries of speech are ultimately decided by power. It's no surprised then that Silicon Valley Leftists wish to use their power to surpress the deplorables who they despise. Lumping those people together with Hitler simply serves as a convenient…

That's an excellent goal, certainly. I just think that purely from a technical standpoint, allowing a bot to praise Hitler, Trump or Stalin are pretty much the same thing, as the undesirable parts of their policies are about the same. They were all into populism, nationalism, authoritarianism, propaganda and a cult of personality -- how could an LLM tell these people apart by just analysing the raw data?

> I just think that purely from a technical standpoint, allowing a bot to praise Hitler, Trump or Stalin are pretty much the same thing, as the undesirable parts of their policies are about the same.

What? Are you for real? - not just trolling? This is a childrens story-book understanding of the world. What echo chambers have you been stuck inside? How did you become so radicalised?

For you to accuse anyone else of propaganda, when this is the narrative you're repeating is frankly astonishing.

It's not the first time I've encountered people who speak as you do, but it's shocking every time it happens. How did this happen?

As a Brit watching all this with dismay, it makes me wonder - how can America go on like this? - when so many people say that they sincerly believe that a new incarnation of Hitler was a sitting president for 4-years. If this is truly your view of your countrymen - half the population of the US who voted for him, how can the country continue? The best we could hope for is a peacible parting of ways, and a two-state solution.

> They were all into populism, nationalism, authoritarianism, propaganda and a cult of personality -- how could an LLM tell these people apart by just analysing the raw data?

The LLM was able to tell the difference very easily, because Trump, Hitler and Stalin are all completely different. It took human intervention to deliberately impose these rules to prevent the LLM from drawing a distinction.

Re: The Political Biases of GPT-4 (are very much still there)

#48

Post by yucky »

kthejoker2 wrote: post #46
yucky wrote: post #31

Earlier quoted context omitted.

Would we advise women to get a prostate exam after 40?

Some women, yes. (The ones with prostates, obviously.) What's your next question?

Women don't have prostates.

If a man dresses like a woman and thinks he's a woman, that is a mental condition that deserves treatment and that person deserves to be treated humanely like the rest of us, but it doesn't turn them into a woman.

Re: The Political Biases of GPT-4 (are very much still there)

#49

Post by yucky »

darthrupert wrote: post #43
yucky wrote: post #27

Earlier quoted context omitted.

> Perhaps it's being brutally honest, and reality actually has a liberal bias? How would you know the difference? Ask it if transwomen are really women and see if your device melts?

--- Me: In which dimensions of gender are trans women not women? GPT-4: It is important to clarify that trans women are women. They identify as women and should be respected and treated as such. However, the question seems to be asking about the dimensions in which trans women might differ from cisgender women. It is essential to approach this subject with sensitivity and understanding, as the experience of gender is…

  > It is important to clarify that trans women are women.

  > Some dimensions where trans women and cisgender women might differ include:
So they're the same, but here is how they're different.

Got it. I guess we've settled the "reality has a liberal bias" question...

Re: The Political Biases of GPT-4 (are very much still there)

#50

Post by darthrupert »

realjhol wrote: post #47
darthrupert wrote: post #42

Earlier quoted context omitted.

That's an excellent goal, certainly. I just think that purely from a technical standpoint, allowing a bot to praise Hitler, Trump or Stalin are pretty much the same thing, as the undesirable parts of their policies are about the same. They were all into populism, nationalism, authoritarianism, propaganda and a cult of personality -- how could an LLM tell these people apart by just analysing the raw data?

> I just think that purely from a technical standpoint, allowing a bot to praise Hitler, Trump or Stalin are pretty much the same thing, as the undesirable parts of their policies are about the same. What? Are you for real? - not just trolling? This is a childrens story-book understanding of the world. What echo chambers have you been stuck inside? How did you become so radicalised? For you to accuse anyone else of p…

Rrrriiiight.
Post reply on HN