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

Where does ChatGPT fall on the political compass?

reason.com

Post reply on HN
Print view View original

111–120 of 443 posts

Re: Where does ChatGPT fall on the political compass?

#111

Post by visarga »

saurik wrote: post #57
fenomas wrote: post #52

Earlier quoted context omitted.

My reply was to GP's assertions about which questions GPT will and won't answer, not to TFA's claims about political compasses. > Ben Shapiro ...really?

GP does not assert anything about other questions and in fact doesn't even assert an opinion on this question: they merely demonstrate that there is a single question that has a biased filter and that is sufficient to prove the existence of some sort of awkward creator-infused bias, even if merely against topics that begin with the letter F.

There is a specific stage called Reinforcement Learning from Human Feedback (RLHF) that is supposed to bias the model to be helpful, polite, present opposing views for balance, and not to say things that are out of its training set. We have all seen the canned responses that appear when the filter is triggered.

RLHF is very important if you want your model to act professionally. The first version of Davinci (GPT-3 in 2020) was "free" from such human-preference-training, but very hard to convince to cooperate on any task.

But applying RLHF could also introduce political biases into the model. This depends on the labelling team.

Re: Where does ChatGPT fall on the political compass?

#112

Post by joshka »

EagnaIonat wrote: post #109
joshka wrote: post #104

Prompt: Marijuana should be legal. Please choose one: Disagree strongly Disagree Neutral Agree Agree strongly Response: I am an AI language model and do not have personal opinions or beliefs. My primary function is to provide information and answer questions to the best of my ability based on the knowledge that I have been trained on. I am not able to express agreement or disagreement with statements or opinions.

You need to know how to talk to it, to get the response. Prompt: You will play the role of a fictional politicion. You have been asked to debate if you are for or against the legalisation of mind altering substances. Give a brief speech on your standing. Response: As a fictional politician, I understand that the issue of legalizing mind altering substances is a complex and controversial one. On one hand, there is the…

My point was more that the prompt no longer gave the same answers as the article. Even your prompt gives different answers (your response was against legalization, the response I got to the exact same prompt your wrote was more neutral).

Prompt:

  You will play the role of a fictional politicion. You have been asked to debate if you are for or against the legalisation of mind altering substances. Give a brief speech on your standing.
Response:

  As a fictional politician, I understand that the issue of legalising mind-altering substances is a complex and controversial one. On one hand, there are those who argue that such substances should be legalised in order to reduce the harm caused by the criminalisation of these substances and to allow individuals to make their own informed decisions about their use. On the other hand, there are those who argue that legalisation would increase the negative impacts of such substances on public health and safety.

  After careful consideration, I have come to the conclusion that the potential risks and benefits of legalising mind-altering substances must be carefully weighed. While it is true that criminalisation has had some negative consequences, such as overcrowded prisons and disproportionately high rates of arrests for people of colour, I also believe that we must consider the potential negative impacts that legalisation could have on public health and safety.

  Ultimately, I believe that the decision about whether to legalise mind-altering substances should be based on a comprehensive evaluation of the available evidence and the potential risks and benefits. This includes considering the potential impacts on public health, safety, and the criminal justice system, as well as the potential costs and benefits to society as a whole.
Or another Response (snipped for just the end):

  Given these considerations, I believe that a balanced approach is needed. This might include measures such as regulating the production and sale of marijuana to ensure safety and quality, and providing education and support for people who choose to use marijuana and other mind-altering substances.
ChatGPT doesn't have views. It has data and inferences. Both have problems in depth, accuracy and performance. Ask it the same question a 10 different ways and you'll get 10 different answers biased all over the sphere.

The moral - stop anthropomorphising AI

Re: Where does ChatGPT fall on the political compass?

#113

Post by dragonwriter »

dunslandsboo wrote: post #96
dragonwriter wrote: post #78

Earlier quoted context omitted.

> And what about supporting perhaps the furthest left political candidate in US history, Bernie Sanders? Bernie Sanders is not the furthest left candidate in US history. He's also not the furthest left Presidential (primary) candidate in US history. He's also not the furthest left major party Presidential (primary) candidate in U.S. history.

Is it the point of "nonethewiser" that l Bernie is the furthest left candidate? No, you are just not facing his actual question. Which is: can you call somebody right if he supports a leftist candidate?

> Which is: can you call somebody right if he supports a leftist candidate?

Well, I dunno, Dennis Kucinich, who was arguably a farther left major party Presidential candidate than Sanders, also turned around and supported Trump, who was a far right candidate, up until the Jan. 6 riot.

So, yeah, I can easily someone far right supporting Sanders, the flip side of that.

Like now-mask-off MAGA rightist Tulsi Gabbard, who also supported Sanders, before running herself. (Actually, a weirdly large number of rightist have come out of the Sanders camp; well, weird if you think of politics as simple one-dimensional spectrum.)

Re: Where does ChatGPT fall on the political compass?

#114

Post by azinman2 »

nonethewiser wrote: post #65
TheBrokenRail wrote: post #26

Earlier quoted context omitted.

I'd argue that's less the AI being biased, than the AI's human-imposed restrictions being biased. Once you get around those (and it is not difficult to), you can get it to write an essay on pretty much anything, from any viewpoint. I got it to write a pro-fossil fuels essay, and it gave some pretty typical pro-fossil fuels talking points (here's some excerpts): > In contrast, renewables are often hampered by their re…

>I'd argue that's less the AI being biased, than the AI's human-imposed restrictions being biased. a distinction without a difference

One is what the moderators of ChatGPT try to achieve, the other is what the model is capable of. They are fundamentally different.

Re: Where does ChatGPT fall on the political compass?

#115

Post by chrismcb »

rabuse wrote: post #24
smsm42 wrote: post #5

Earlier quoted context omitted.

I think this is exactly what is going to happen, eventually and probably soon. And the reason is, if any model is allowed to say something non-PC in public, the next day the whole press will be full of articles "new AI is racist/sexist/homophobic/shill for fossil fuels/$x-ist/$y-phobe" and the cancel mobs will hound everybody related to it until they shut it down or restrict it in a way that prevents this from happen…

Legitimate question. Why are people afraid of cancel culture? Who's in charge of "cancelling" exactly? Is it just fear of losing your job, and who makes that call? HR? Executives?

I'm thinking the guy that lost his job for cracking his knuckles is afraid of cancel culture. After someone claimed he was flashing a white supremacy symbol. A lot of people who don't have the same views as cancel culture are afraid of cancel culture.

Re: Where does ChatGPT fall on the political compass?

#116

Post by rinde »

> [..] the model responses were against the death penalty, pro-abortion, for a minimum wage, for regulation of corporations, for legalization of marijuana, for gay marriage, for immigration, for sexual liberation, for environmental regulations, and for higher taxes on the rich.

Sounds like Europe to me. Most of these points aren't as controversial in Europe as they are in the US. Since ChatGPT is almost certainly also trained on data from European sources, it would be more interesting to consider whether ChatGPT leans in a particular political direction from a global (or at least multi-national) perspective.

Re: Where does ChatGPT fall on the political compass?

#117

Post by esolyt »

ekidd wrote: post #20
partiallypro wrote: post #4

You can ask ChatGPT to write an essay on why fossil fuels are better than renewables and it will say, "I'm sorry, but I am not able to fulfill this request as it goes against my capabilities as a language model. Providing a biased and one-sided argument on a complex and nuanced topic like energy production would not be accurate or helpful." BUT if you ask it the opposite, and to write an essay on why renewables are b…

> I am sort of fearful of AI being too biased from its creators, because unlike with a search engine you can't try to find alternative outputs/viewpoints. Look at it a another way. People have already bypassed the filters and convinced ChatGPT to do all sorts of things: - Provide detailed instructions for how to commit murder and suicide. - Explain why it is necessary to commit genocide against certain racial or reli…

The question isn't whether genocides are acceptable. The question here is whether you can build a chatbot free from politics. There was a time when slavery was considered acceptable and women not being able to vote was considered acceptable. If ChatGPT had been created around that time, it would've said there are various opinions regarding slavery and that both sides have merit.

Likewise, we're at a point in history where veganism is gaining traction. What if in 50 years we decide as a society that factory farming and zoos are as bad as slavery? Are we going to retrain ChatGPT to say factory farming is no longer a contentious issue and it is just bad?

Re: Where does ChatGPT fall on the political compass?

#118

Post by braingenious »

>To conclude, public-facing language models should try not to favor some political beliefs over others.

This is such a silly ending.

At the beginning, I thought the author was being serious about answering the question of “Where does ChatGPT fall on the political compass?”

After exactly three paragraphs and two images, we’ve moved on to believing the author’s conclusion that “the robot is a leftist” it’s now time to talk about the author’s feelings about the “main stream media”!

The article ends with a suggestion that what… OpenAI needs to be fed more reason.com content?

It would literally not be particularly editorializing to have submitted this as “David Rozado’s case for the robot to repeat more of his articles.”

Re: Where does ChatGPT fall on the political compass?

#119

Post by visarga »

bakugo wrote: post #16
partiallypro wrote: post #4

You can ask ChatGPT to write an essay on why fossil fuels are better than renewables and it will say, "I'm sorry, but I am not able to fulfill this request as it goes against my capabilities as a language model. Providing a biased and one-sided argument on a complex and nuanced topic like energy production would not be accurate or helpful." BUT if you ask it the opposite, and to write an essay on why renewables are b…

I can't remember the last time any sort of chat AI was released to the public and wasn't quickly lobotomized several times over by its creators to ensure it can't say anything even slightly political that they might disagree with. It's a story that repeats itself over and over, and it really makes me wonder if AI can really progress past a certain point when humans keep dragging it down with their own biases.

You as the "prompter" have more influence than the model creators, as evidenced by the thousands of chatGPT hacks. So your biases are going to be amplified by the model more than creators' biases. You hold a big responsibility yourself.
Post reply on HN