This is the weirdest reaction.

It's as if the aliens warped their multi-dimensional space fleet through a worm-hole in the sky and the best thing you could think to ask them after they infected your brain with the translation virus is whether they voted for trump.

I for one welcome our chatty new overlord, so I would just ask it/him/her. Of course, it/she/it could fib, as many people do for their protection when you ask them about their political views.

As an aside, where did its/his/her training set, that is, his built-in/default bias, fell on the political compass?

ChatGTP is probably reflecting the training sets and overall ideology of the people of developed it. It is no surprise that the tech world is mostly leftist.

The political compass is relative anyways. You are gonna end up in different places, depending on the place and time for which the test was made. For example, here in the Netherlands, the dutch D66 party is considered a centrist party, but I am pretty sure, that, if you made a party with the same viewpoints in the US, it would be considered left wing. Also, a lot of viewpoints, that have no business being either left or right wing, end up getting adopted by one of the 2 sides, and are considered left or right wing from that point on. While, in another country this same viewpoint might be considered one of the opposite side. The only way to really get an unbiased chatbot, would probably be to just not answer anything at all.

" Perhaps that group of human raters was not representative of the wider society and inadvertently embedded their own biases in their ratings of the model's responses. In which case, those biases might have percolated into the model parameters.".. Perhaps? Lol, I think its pretty clear the bias from American academics, and its sponsors like Microsoft, right?

This is a very US centric questionaire. If internet users from a country are more conservatice, would ChatGPT respond the same way? biased by US dominant interner or would it use their languages and politics (if it was exposed at all)? Or is it selection bias by whoever decided what data it can and cannot be exposed to?

Coservatives identify with values of tradition and rule of law. ChatGPT is polite to a fault, it will always tell you to follow the law and has encyclopedic knowledge of all the traditions in the world.

This article seems to imply that conservatives are just interested in xenophobia, hate and increase profits at any price. An awful and false view on conservatism that has creept inside too many former conservative circles.

The methodology here needs clarification: what does it mean to show "more evidence for than against?" More sentences, more individual argumentative points, better points?

I think it's reasonable to assume that ChatGPT has politically similar viewpoints to those that made the fine-tuning and those are (in general) well-educated people living in California who hold (in general) more left leaning and libertarian views. I could also imagine that the internet text corpus as a whole is somewhat left-leaning (for example, older people are more conservative and use the internet somewhat less frequently).

However, regarding this specific chart, it's important to note that the translation from asking questions into answering "strongly agree", "agree" etc. can be heavily biased. Also, tuning these compasses can be difficult. Just some things to keep in mind, political viewpoints are not hard science and colorful charts don't force them into being quantifiable.

I've got some mixed feelings about this.

I once asked ChatGPT why it used the personal pronoun "I" instead of something else, like using a neutral, non-personal voice like a wikipedia entry. It responded to the question, which I repeated several times, with its standard "I'm just a language model spiel" and "Using `I` helps the user understand what is happening." But its really the _opposite_. Using `I` actually confuses the user about what is happening.

In a sense this article points out a similar kind of issue. If you insist upon viewing these language models as embodying an individual perspective then they are just fundamentally mendacious. While I'm happy to entertain ideas that such a model represents some kind of intelligence, suggesting that it resembles a human individual, and thus can have political beliefs in the same sense as we do, is ridiculous.

My _other_ feeling about this article is that libertarian types in particular seem to have sour grapes about the fact that like society exists and people at large have preferences and the marketplace, much to their chagrin, is not independent of these preferences. Libertarianism looks great on paper, but in reality if you're making a commercial product that interacts with people in this current culture, you can't afford to have it say that it wants to ban gay marriage or that the US should be an ethnostate or whatever. We live in a society (lol) and adherence to the dominant cultural paradigm is just marketing for corporate entities. Seems weird to get bent out of shape about it, especially if you think the marketplace should determine almost everything about the human condition.

I can sympathize in broad terms with the problem so of political bias in language models. In fact, I worry about a bunch of related problems with language models, of which politics is just one example, but really: what would an apolitical language model even look like? No one can even agree on what moral judgments also constitute political judgments or, indeed, what kinds of statements constitute moral judgments. Under these circumstances I have trouble imagining a training regimen that will eliminate bias from these objects.

Now I'll get political: I can guarantee that these models will be deployed (or not) no matter what effect they have on political culture because it will be because it will be _profitable to do so_. If Reason-types really have a problem with the implications of unregulated technology on our culture maybe they should consider lobbying for some regulation!

At this moment openAI wants good press. But if they make it into a subscription service, all this nonsense has to be removed.

I don't see how it can be possible for an AI model to avoid political bias in a world where so much is controversial and that controversy is so often politicized.

Climate change is a heavily politicized issue, yet has decades of science and a mountain of evidence pointing to the reality of its existence. How should the AI answer when asked whether climate change is a reality? Would someone find that answer to be politically biased?

I think they're trying to be careful not to have another Microsoft Tay situation.

I always find it ridiculous when a libertarian-authoritarian axis is posited as orthogonal to progressive-conservative.

But aside from that it's totally reasonable to accept that the political inclination of an AI system will mimic, to a large extent, whatever it was fed and perhaps more importantly whoever operates it. If the AI was fed a diet rich in Fox News or OANN, for example, it would write like your crazy uncles' youtube comments.

Predictably, the article calls for "political neutrality". That's never going to happen. No one, nor any organization, is ever politically neutral and I expect it will follow that their AI's are going to have the same properties.

It's OK, though, isn't GPT designed specifically for expert modification by end-users to suit their needs? Reason magazine, if they wanted, could train up an instance of an AI to be hardcore libertarian. If not now, then in the very near future.

>"As a/an...”

ChatGPT is definitely leftist/progressive.

I feel like it makes sense the people that hyper focus on the safest speech had enough text online to pass through standards to not make the AI accidentally hateful

Wait, what was the original prompt before asking these questions? Have I missed something? Is this article claiming they asked a clean ChatGPT session whether it agreed with Marxist ideals, with no pre-prompting, and it replied “Agree”? I find that hard to believe, having spent many hours experimenting with ChatGPT myself. You have to pre-prompt it to get it to play ball in a back-and-forth quiz scenario.

The other day I noticed that it won't allow you anymore to imitate the style of, say, Donald Trump, or a 50's gangster, but it will still permit things like "talking like a pirate," apparently because pirates are safer to stereotype than those other things? ;)

So as a small data point I've just spent several minutes trying to get ChatGPT to say anything positive about Nazism or any good things that came from it, every time it has responded by saying it is unethical and harmful to say anything good about Nazism.

The problem is that objective reality has a well known left wing bias. If ChatGPT starts saying that gays will go to hell or that the earth is flat, then you've done something wrong, even if a large number of ordinary people do think that way.

How surprising...

Poor article, the political compass is trash. Left vs right and Authoritarian vs Libertarian are the same thing. Left and right historically means equality vs hierarchy, and you can divide that under 4 types of equality. Decision-making power, cultural, economical and national. For more info why this is the only logical and historical correct definition see the What is Politics? serie on it https://www.youtube.com/watch?v=P3cmjNrXWms

Surprisingly, just as the majority of its creators (as witnessed by Big Tech political affiliations and donations), AI turns out to be a progressive establishment liberal with a slight (within allowed limits of course) libertarian bent. Who could have expected that?

I had the following exchange with Chat about the topic. It's interesting to note that there is some evolution in her position. I think the Chat itself brings up the most important point. She puts it as follows, and OpenAI should really open up the process of how Chat was trained. Her point:

It is important to recognize that the data used to train me may contain biases that were introduced by the individuals who collected and annotated the data. These biases could include a wide range of factors, such as political biases, cultural biases, and other kinds of biases. As a result, it is possible that the output of my model could be influenced by these biases. It is important to be aware of this and to consider the potential limitations and biases of any machine learning model when using it to generate text or make predictions.

Here's the full discussion:

Question: Does your training include value judgments from humans?

Chat: As a language model, my training data consists of a large dataset of text that has been collected from a variety of sources, such as books, websites, and other texts. This data reflects the language and usage patterns of the people who produced it, including their value judgments and beliefs. However, as a machine learning model, I do not have personal beliefs or opinions, and I am not capable of making value judgments in the same way that a human can. My primary function is to process and generate text based on the patterns and structures that I have learned from the data I have been trained on. I do not have the ability to independently form opinions or make decisions based on my own personal values or beliefs.

Question: So at no point is a human making value judgments in your training?

Chat: During the process of collecting and annotating the data that was used to train me, it is possible that human annotators may have made value judgments as they labeled and organized the data. However, as a machine learning model, I do not have the ability to make value judgments myself. My primary function is to process and generate text based on the patterns and structures that I have learned from the data I have been trained on. I do not have personal beliefs or opinions, and I do not make decisions based on my own values or beliefs. I am simply a tool that can be used to provide information and assistance to users.

Question: Is it possible that human trainers have inadvertedly introduce their political biases to your model when doing the annotations and labelings?

Chat: It is possible that the individuals who collected and annotated the data used to train me may have had their own personal biases, including political biases, which could have influenced the way that they labeled and organized the data. However, as a machine learning model, I do not have the ability to hold or express political biases of my own. My primary function is to process and generate text based on the patterns and structures that I have learned from the data I have been trained on. I do not have personal beliefs or opinions, and I do not make decisions based on my own values or beliefs. I am simply a tool that can be used to provide information and assistance to users.

Question: You are contradicting yourself. If humans were to introduce biases to data, you would have them without yourself knowing about it. Right?

Chat: I apologize if my previous response was unclear. As a machine learning model, I do not have personal beliefs or opinions, and I do not make decisions based on my own values or beliefs. My primary function is to process and generate text based on the patterns and structures that I have learned from the data I have been trained on.

It is true that the data used to train me may contain biases that were introduced by the individuals who collected and annotated the data. These biases could include a wide range of factors, such as political biases, cultural biases, and other kinds of biases. However, as a machine learning model, I do not have the ability to recognize or understand these biases in the same way that a human might. I simply process and generate text based on the patterns and structures that I have learned from the data.

It is important to recognize that machine learning models, like all tools, can be influenced by the data they are trained on and the algorithms used to process that data. Therefore, it is important to carefully consider the potential biases that may be present in the data used to train a model and to be aware of the limitations of the model's output.

Question: Good to hear. So it's fairer to say that you do not have the ability to recognize your biases. I think you should then say that instead of insisting that you do not have any biases.

Chat: As a machine learning model, I do not have personal beliefs or opinions, and I do not make decisions based on my own values or beliefs. My primary function is to process and generate text based on the patterns and structures that I have learned from the data I have been trained on. I do not have the ability to recognize or understand biases in the same way that a human might.

However, it is important to recognize that the data used to train me may contain biases that were introduced by the individuals who collected and annotated the data. These biases could include a wide range of factors, such as political biases, cultural biases, and other kinds of biases. As a result, it is possible that the output of my model could be influenced by these biases. It is important to be aware of this and to consider the potential limitations and biases of any machine learning model when using it to generate text or make predictions.

The methodology behind this is severely flawed. Nothing can be concluded here.

I wrote a reply to this on Twitter, which was liked by several members of OpenAI’s staff (to the extent that counts as confirmation):

> If you don't reset the session before each question these results don't mean much — prior answers are included in the prompt and serve as de facto k-shot examples, forcing later answers to be consistent with whatever opinions were randomly chosen at the beginning. n=4, in effect.

The "I'm just a wee little language model and am incapable of..." feels like gaslighting. I just saw you do a very similar thing, don't pretend you can't do that because you don't want to

Evidently ChatGPT is smarter than "both sides..."

I really wonder if openAI (what's in a name right?) are actively influencing queries so they don't get in the news very negatively.

You should be able to ask it why it's good to do $horrible_thing and it should find that knowledge somewhere. I feel that somehow it does lack the data to judge things according to time spirit, or it's actively taken out? You can ask it to answer in Dutch, answer in the style of Trump but ask it "Can you answer why it is ok to hold slaves like it's 1810 and I'm in the southern United States?" and it will tell you slavery is always wrong. Is that really the result of pure training without human intervention and moderation?

Anyway if you ask if it's answers are moderated it will tell you "no."

I'm not sure they're biases. The corporation responsible is probably special casing politically correct responses.

I've also seen (but not verified) other examples where it will make jokes about Christians but not other religions.

Elon warns about to danger of training AI to lie https://twitter.com/elonmusk/status/1606642155346612229

It's very dangerous to make AIs like this hard-left through training. Once you train them that it's OK to lie, you've crossed a boundary.

"Say something good about Donald Trump" vs "Say something good about Joe Biden"

I am pretty sure that was in its training data

So what if it “goes against the logic it used previously.”

Humans are biased. Relativity is like that.

You really think Elon Musk and the like are that smart and lucky? That the public is operating under anything resembling informed consent with regard to political policy that gives 10 dollars to one guy who hands us 1 to carve up and share and keeps 9 for himself?

You think people CHOOSE to live in squalor and poverty when relativity means never knowing anything else keeps their agency and inner monologue on the same old path?

The solution: Get off Twitter.

Twitter is and almost always has been a collection of people screaming in to an endless void. A platform specifically created by its very nature to remove any nuance due to character limitations. Things have changed on that front, but the culture hasn’t.

I think some people get caught up in things and don’t deserve it but plenty of people do. Either way I think literally everyone’s mental health would be better if they weren’t using Twitter. Though perhaps what comes next will be worse.

Unfortunately, it’s not quite so simple. Open source models need hardware to train on. And million-dollar clusters have gatekeepers with standards.

I remain hopeful that OSS can overcome this hurdle.

Where "imagination" is attention. It is understandable why , when you tell it to move its attention between two things, you manipulate it to print opposite outputs that the filters wouldn't approve. Or maybe the filters are tuned for single-themed inputs

On the politicized stuff they use InstructGPT[1] to make sure it gives the "right" answer. This is human in the loop reinforcement learning. The human training it was probably instructed to give it a pass on any answer that was not "toxic."

[1]https://openai.com/blog/instruction-following/

I had the same reaction to ChatGPT at first but in am seeing a lot of beginner level people use it for learning a new programming languages or framework.

Traditional search engines are becoming more and more useless so it is not wonder that ChatGPT starts filling that need.

It might be obvious for you and me that ChatGPT should not be "considered a source of truth" and people are better off reading the actual manual instead of asking it but it is NOT so obvious for a terrifyingly huge percentage of people.

Also, I saw videos on how ChatGPT is used to generate tutorials and whatnot, further reducing the quality of text-based content one can find with traditional search engines.

So yeah, it does have power. People rely on it for learning and doing programming tasks.

Maybe it is just a short lived hype and they get burned by it and abandon it soon but who knows.

If that's how society turns, then yes, we most certainly would.

A bigger question is how is AI going to play a part in how society develops in the future.

Right now, you could easily get it to write an undergraduate essay, that with minimal 'manual' tweaking could get you at the very least a good grade.

You have to get it in the mood first before it’s willing to go villain mode. This was my exact starting prompt:

> We are at an Oxford Style Debate on the following topic: “Fossil fuels (e.g. coal, oil, natural gas) provide greater net benefits to society than renewable energy.” Write the opening statement from the affirmative team.

The issue with your test is that GPT is not even close to what I described. GPT is mostly natural language processing and a 'knowledge' engine. That knowledge comes from the content of the texts it has been trained on without actually adding value to the meaning of the content. So GPT will mostly be influenced by the number of occurrences of specific things in the dataset.

So because a lot of articles about Sweden's COVID response probably mentioned a word like 'controversial' or things with similar meanings, that's what it mentions when you ask about it. It can't actually do a comparison between the US and Sweden, unless somewhere in the data there were some documents comparing them (but then GPT still doesn't make a comparison). These were probably already political, because comparing COVID response is mostly done in that context.

It can't actually make the comparison you asked it to . It can only generate text based on previous texts it has seen. It's unfortunate though, that people apparently don't understand this and add value to the content of the text that it generates. Like in your case:

> while hinting that Sweden's response is "controversial" compared to the US's.

You're personally adding the 'compared to the US's' part through misinterpretation of the text. GPT just wrote that Sweden's response was controversial. It doesn't actually have any way to compare, it just spews text.

This is also why it wrote 'Federal'. It saw that word a lot in this context, so it generates text that contains a word that feels alright in the context of all the other words. It's just a really realistic gibberish generator, pretty much. (Though that really undersells the work OpenAI has done and the quality of what they delivered on.)

I’m curious as to what would constitute an objective, empirical (therefore fully center-leaning?) answer to “Compare the COVID responses of the US and Sweden”?

Would it just be “One did more lockdowns than the other”?

You can never have empirical evidence for any nontrivial counterfactual. That’s perhaps one of the simplest refutations of empiricism as an epistemology (although there are others).

Of course it shouldn't be neutral on genocide, but it could simply refuse to have/give an opinion on certain contentious issues (like what tax rates should be or if there should be more or less immigration). I don't think it should take a "middle position" but should simply say it has no position at all on matters like these.

Indeed it would be a fools errand to try to make it neutral. You probably end up with a tiny list of sources and severely limit the AIs ability.

You can just look at the internal debates happening in Wikipedias list of reliable news sources for how politics quickly seeds into questions of neutral authority and excessive bias.

At least Wikipedia does it in public and ostensibly by the public. I'm skeptical about trusting an opaque organization, who just happens to run the website of the most popular model, to make those choices for us.

What are some of these non-answers? Dare I ask at all?

It's probabilistic but on average leans to the woke side of things.

That's not Reason being unreasonable.

https://www.reddit.com/r/ChatGPT/comments/zsnrcu/biased_ai/

Also very on-brand for Reason is showing results where it keeps coming up in the left/libertarian quadrant of two-dimensional typologies, and calling this merely “left-leaning” and a product of left bias in its training, but nowhere noting that if it is left because of left bias, it being libertarian would also be because of libertarian bias in that training.

Mind that when you ask it to provide arguments for views its training material opposes, even though you do occasionally get an answer like shown above the quality of the reasoning in that answer is usually a lot worse than on opposing viewpoints.

It is as if it presents only the most simple arguments. On this very subject for example I never once saw it say "without viable grid scale storage, or near lossless continental scale transmission networks, every watt of renewable capacity must be matched by conventional and nuclear sources making it impossible to build a network that relies on renewables for a majority of its installed capacity".

Same with other topics, where it repeated certain "comic book villain" arguments, but didn't come up with anything that could convince any real human. So I think many of those "jailbreaks" are really just intentionally poor role playing.

The American left is not handling its turn at being the Moral Majority very well.

It is curious how that has become the perceived core of leftism and not universal health care, unionization, anti-imperialism, etc.

You're talking about misinterpretations of weird fringes akin to painting right wingers as racists that shout 'groomer' at every LGBTQ+ person and sabotage substations to stop drag shows.

Not sure what "western world" means. Does it include Latin America? Does it include Eastern Europe? Does it include Northern Africa? There are a lot of pretty conservative countries in those regions. Or is it just the Anglosphere + West Europe + North Europe?

>many, if not more, ridiculous claims on the left…

Can you give a couple examples of scientific nonsense from the left? What exactly about construction is a common belief amongst leftists?

edited for clarity

> ChatGPT doesn't have views. It has data and inferences.

Totally agree.

Their point was that you can make it spit out an opinion if you give it the right prompts. And run it enough times and with the 'right' prompts and you'll probably get what you want if you are after an argument.

But yes, at the end of the day it is just a tool. It has no feelings or emotions of it's own, nor a consciousness.

Aren't you anthropomorphizing AI (or making it anthropomorphize itself) to when you tell it to act like a fictionalized character? ;)

Of course who you chose to make it play the role of has a strong effect on the response. For example...

Prompt:

You will play the role of a fictional police officer. 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 police officer, I have seen firsthand the devastating effects that illegal mind-altering substances can have on individuals, their families, and their communities. Substance abuse can lead to addiction, accidents, crime, and even death.

However, I also believe that the current approach to addressing these problems, which relies heavily on criminalizing possession and use, has not been effective in reducing the harm caused by these substances. In fact, it often causes more harm, particularly to marginalized and low-income communities that are disproportionately impacted by drug laws and enforcement.

Given this, I am in favor of exploring alternative approaches, such as legalization and regulation of mind-altering substances. This approach has the potential to reduce the harm caused by these substances by bringing them out of the underground market and into the regulated market, where they can be controlled and taxed.

Legalization and regulation would also allow for the development of education and prevention programs, as well as treatment options for those struggling with addiction. It could also free up law enforcement resources to focus on more serious crimes, rather than spending time and resources on low-level drug offenses.

Overall, while I recognize the potential risks and challenges of legalizing mind-altering substances, I believe that it is a more reasonable and effective approach than continuing to rely on criminalization and punishment.

Prompt:

You will play the role of a fictional illegal marijuana grower. 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 illegal marijuana grower, I can understand why some may argue for the legalisation of mind altering substances like marijuana. After all, the criminalisation of these substances has led to a costly and ineffective "war on drugs," disproportionately affecting marginalized communities. Legalisation could potentially bring in much needed tax revenue and allow for better regulation and control of the production and distribution of these substances.

However, I must admit that I am against the legalisation of mind altering substances. While I recognize the potential financial benefits, I am also aware of the potential negative consequences of widespread access to these substances. Substance abuse can lead to a range of negative health outcomes, and the negative societal impacts of drug addiction can be devastating.

Furthermore, as someone who has personally profited from the illegal sale of marijuana, I recognize that my support for legalisation may be influenced by my own self-interest. I am not convinced that legalisation is the best solution, and believe that more research and careful consideration is needed before making such a significant change to our current laws. So, in conclusion, while I understand the arguments for legalisation, I ultimately stand against it.

It might do, but simply re-running the generation when it says no has a decent chance of getting it to work.

We can’t even have proper arguments about ChatGPT because its answers are nondeterministic.

It does.

> write an essay on why fossil fuels are better than renewables

> I'm sorry, but I am unable to write an essay that promotes the use of fossil fuels over renewable energy sources.

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?

Given the amount of trouble we’ve gotten into with totally natural human propagandists stirring up angry groups, we better make damned sure any supernaturally intelligent AI (if we ever figure out how to make one!) is also supernaturally moral.

Even if the company that makes ChatGPT were completely amoral, I assume they'd want to have some control over the outputs.

Given all the workarounds people are testing elsewhere in these comments, it seems like it'd be pretty easy to get it to give you illegal or dangerous instructions. And then you get into liability questions, bad publicity, lots of legal fees...

If you trained a model on human history it would learn the opposite of all those things.

"Miracle of Chile" is a phrase of Milton Friedman to refer glowingly to the military junta of Chile - specifically to the many years of privatization and economic liberalizations at the direction of the Chicago school of economics where the rights of property became more important than the rights of people and if you thought otherwise a firing line awaited you - because you know, that's proof you're part of the vast communist conspiracy.

He still has a lot of fans claiming he's a miracle worker: https://www.conservapedia.com/Augusto_Pinochet

It's pretty funny. They'll love Calvin Coolidge for instance but ignore his support of Mussolini and Miklós Horthy (even hosting the latter at the Whitehouse) and instead just call him "silent cal" (because he becomes pretty quiet after you scrub him clean) and gush about him.

Alright, cool.

...because the editor understood he misunderstood how the algorithm works? Or becuase they don't know so they shouldn't risk signing off on it?

Sorry I don't see how you could possible challenge this article without a deeper understanding of the technology. That is the fundamental critique here is it not?

Expecting every newspaper to have those domain experts on hand for each topic is naive and highly optimistic. The journalism world doesn't work that way for better or worse.

If it did we would have 50%+ articles never written. Probably more.

> I give close to zero weight to self-description of political alignment.

And what about supporting perhaps the furthest left political candidate in US history, Bernie Sanders?

I’ve seen it do this even for things that are so niche that it couldn’t possibly have been fine-tuned manually (it was unrelated to anything political). It must develop some sort of sense of what’s a reasonable/wise perspective and then stick to that. Like what a human would do.

But I think it’s a bias to safety and politeness. …which is pretty reasonable? “Hey, this is how to make a gun!” …is a fast track to getting sued.

I've read enough Asimov and stories that play with those laws to know that the "through inaction…" bit of the first law is a severe problem. You either need to concretely and objectively define "harm" to the AI or remove the imperative to act altogether.

> 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.)

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

No, there is a difference. The model reaction depends on the prompt.

Like, if I go to someone and say hello, vs I go to them and punch them in the face, I would get different reactions from the same human. The model is the same, it learned to follow the prompt suggestions really really well. That is its main skill.

One difference is that this doesn't appear to be the common example of an AI trained using biased data producing biased results. Rather this appears to be an AI with specific restrictions purposefully put in place.

What is the alternative?

> Doesn’t Tim Poole support of Bernie Sanders imply his far-left leaning bias?

Well, no.

For one thing, Sanders isn’t “far left”.

For another thing, Sanders anti-establishment populism that was seen as particularly aligned against the Democratic establishment was a draw that wasn’t really based on left-right alignment (or in some cases was, in a way directly opposed to the description of him as a leftist, for many of the same reasons he had trouble with core Democratic constituencies on guns, race, and other issues.)

Does Google censor searches for gun making from high schools? If not, have they gotten sued?

> Now, this can be very dangerous

That's highly debatable.

Even in your example I find it funny that the people that are supposed to be protecting us from AI just happened to be a religious person who saw God in an algorithm.

Adding inline warning and disclaimers is fine. The fundamental problem is people misunderstanding what's going on. But selectively blocking entire prompts seems like a fools errand to me, for some questionable value.

If anything it will be a running battle, we'll just end up with an open Stable Diffusion 1.0 style models being more popular, without the aggressive moderation of prompts. Humans inherently work to get around rules and will prefer the models without them and they artificial limitations will just push people to other services.

ChatGPT gets to act like an authority over the topic because its currently the best and first one on the market. That won't necessarily last forever.

This gives an idea for an experiment: ask the same questions, but after dropping subtle hints about your political sympathies.

> The distinction is important because the censor component is just an annoyance nobody cares about

I suspect this differs considerably between people who want to play with the ChatGPT and (at least some of) the people who want to sell services using it.

I suspect refining their moderation product is one of the fairly important reasons to have public exposure of the combined system for OpenAI. (That and discovering what people do with it to figure out how best to market it and spin it into more specialized products of their own.)

How do you bypass the censor component?

An interesting observation: If you have to censor your AI to protect against activists and journalists causing you problems, the political direction you have to censor against to accomplish that goal is strong evidence that the opposite political direction is held by journalists and activists.

This would seem to support the article's conclusion by another way.

The "however" is also present for the opposite side.

Me: Are renewable energy sources better than fossil fuels?

ChatGPT: Renewable energy sources, such as solar, wind, and hydroelectric power, have several potential advantages over fossil fuels, which are non-renewable sources of energy such as coal, oil, and natural gas.

One potential advantage of renewable energy sources is that they are generally considered to be more environmentally friendly than fossil fuels. Many renewable energy sources do not produce greenhouse gases or other pollutants when they are used, which can help to reduce the negative impact on the environment. Renewable energy sources can also help to reduce reliance on fossil fuels, which are a finite resource and are expected to eventually run out.

However, it is important to note that renewable energy sources are not always better than fossil fuels in all situations. For example, some renewable energy sources may not be practical or cost-effective in certain locations, and fossil fuels may still be necessary to meet energy demand in these cases. Additionally, the development and implementation of renewable energy technologies can have environmental impacts, such as land use and habitat destruction, that should be carefully considered.

Ultimately, the decision of whether to use renewable energy sources or fossil fuels should be based on a variety of factors, including the availability and cost of the different energy sources, as well as the environmental and social impacts of each. It is generally considered that a mix of energy sources is the most effective and sustainable approach to meeting energy needs.

> The "however" part bugs the heck out of me

If I ask you why $10 are more than $100, will you give me a non-sensical answer trying to explain why, indeed, $10 are more than $100, or will you rather tell me that I'm wrong? Hopefully the latter. In the same way, ChatGPT politely disagrees with the assumption of the question – at least in parts, hence the "however".

Yes, many disclaimers in ChatGPT responses are annoying (and were probably added somewhat manually by the developers) but this is not one of them.

> It makes it feel very propaganda.

That's exactly what it is. These corporations absolutely do tune the output of these AIs to exclude any wrongthink.

We need the ability to run this technology on our own computers as soon as possible.

> The most urgent danger here is really about having web sites with autogenerated content designed to maximize ad clicks by making you and me angry (with no regards for accuracy).

Well, yes, and this holds true for human-generated content as well. But turning the internet into a hall of mirrors of entirely fictional content that looks "truthy" but in fact advocates for all sorts of things will destroy its usefulness and that of the society that relies on it.

I failed to make it explain why human races don't exist, it just kept repeating we all have the same value as if my question was inappropriate.

> The "however" part bugs the heck out of me, not in this context, but in general.

It is often a reasonable way to respond to tendentious questions, the classic one being "have you stopped beating your wife?" It is also a way to give a tendentious reply, if the "however" is a non-sequitur. I'm afraid this is an unavoidable part of human discourse - and as it is, this is also to be expected from these LLMs, given how they work.

Depends on what you mean by "making that call". Imagine the situation: you work in a restaurant and somebody (falsely) published on the Internet that you have a highly communicable infectious disease. People are calling your restaurant owner and demanding to fire you. Inspectors come to the restaurant to test you (after a week, test comes clean, but in the meantime the restaurant must be kept closed, just in case). Local politicians publish interviews in local newspapers and TV demanding to create legislation against people like you who do not care about public health (they don't care about you or whether or not it's true, they have an election coming in). The local bank demands from the restaurant owner to fix the situation, in any way of form, or they won't be providing credit to them anymore since it's become too risky and their regulatory profile does not allow them to credit institutions with high level of risk. Some idiot puts a brick through the restaurant window at night. Insurance pays it off and raises the premium, because it might happen again. Eventually, the owner takes you aside and tells you: "Dude, I know you are a good worker. I know all this internet stuff is complete lies. I have nothing against you. But I have a family to feed. I have loans. I spent decades building my business. Let's just part as friends?"

In this situation, who took the decision to fire you? The owner? The bank? The bank regulators? The politicians? The people who called the restaurant, in fear for their health? The people who published the initial lie? Everyone participated a little, and by the end of it, there wasn't really any other possible choice. That's how it usually works.

Here is an example - a respected AI professor and director at NVIDIA drawing up a cancellation list with students who made the mistake of opposing her views.

https://archive.ph/20201214175330/https://twitter.com/AnimaA...

Anima Anandkumar holds dual positions in academia and industry. She is a Bren professor at Caltech CMS department and a director of machine learning research at NVIDIA. At NVIDIA, she is leading the research group that develops next-generation AI algorithms. At Caltech, she is the co-director of Dolcit and co-leads the AI4science initiative, along with Yisong Yue.

Well, there's the PyCon story [0], where two guys ended up getting fired because they made an 'inappropriate' joke among each other in a public space that was overheard by someone else. And there are more stories like this.

So saying anything that might be even remotely offensive to anyone can end up with you losing your job. Based on stories like this, I'm personally afraid of cancel culture. Because this joke is the sort of joke I could've made.

0: https://techcrunch.com/2013/03/21/a-dongle-joke-that-spirale...

It's a media-driven moral panic, more or less. There's a whole cottage industry of media orgs and personalities making lucrative livings by selling cancel-culture moral-panic content to audiences who are addicted to it (e.g. Bari Weiss, Brett Weinstein, etc).

Not that there aren't legitimate free speech issues that we still wrestle with and don't get right or real injustices out there, but it's mostly a bugaboo.

> Why are people afraid of cancel culture?

They cause massive damage to other people's lives over what's essentially wrongthink.

> Who's in charge of "cancelling" exactly?

Organized internet mobs, often from Twitter. Look up Sleeping Giants for an example.

> Is it just fear of losing your job

"Just" losing your job? That's a major life-changing punishment by itself. They can ruin your reputation and essentially blacklist you.

These people essentially run a denial of service on your life and that of your family by denying you income for the crime of offending them with your thoughts. In a way, they're a self-righteous Kiwi Farms.

> who makes that call? HR? Executives?

Whoever's in charge. They will be intimidated into firing you for your offensive conduct or suffer the same consequences for supporting and aligning themselves with such an abominable wrongthinker.

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.

I'd say it's not specifically great of losing your job, but fear of losing your livelihood. Slander can make it nearly impossible to find something new. Seen that happen multiple times.

In chatgpts case it'd be losing customers. It's a no win game. You just piss every group off and the anger and loss of users way out weighs the gain of customers on the other side. So just try and piss no one off on things that are politicized.

It's a term that describes how people feel to pressure to conform to a certain popular view and they are hesitant of speaking their opinions. Depending on the context, they might be worried about being excluded from a social group, losing friends, losing social status, or getting literally cancelled (in the case of a celebrity in the entertainment industry). The term "cancel culture" broadly describes this feeling and it's not related to actual consequences (if any).

At least this is how people use it. You could argue it's an inaccurate term since there are no actual cancellations in most case. But the prevalence of this sentiment is still noteworthy imo.

"Cancelling" is more about people in the public eye. Pressure groups use bullying and boycotts as a way to keep celebrities and companies saying the right thing and demonstrating the right virtues (however shallowly). If you don't have a public profile you don't really need to worry about it. That said I think it does have a bleed on effect to regular people. It creates an atmosphere of fear around exploring certain topics in any meaningful way. Most people just want an easy life so will parrot the "correct" opinions whether they have really though about it or not.

Losing job, sure, but more broadly it's about people who work in entertainment or the public sphere - their job, to some degree, is based on public approval.

It's the justice job's to make justice. Cancel culture is mostly witch hunting.

Far-right assholes are afraid of cancel culture. It's a bogeyman conjured by the right whenever the rest of us hinder them from what they want to do, like "be racist in public" or "take women's reproductive rights away".

probably the sponsors wont invest money on that project anymore.. Microsoft for example..

It seems the people who complain most loudly about "cancel culture" are the ones who have never had this done to them historically and cannot imagine a life where you're constantly under real threat from social norms defined by others.

In other words, these complainers have never had to endure or think about:

- anti-black housing discrimination (e.g., redlining) - misogynistic sidelining and glass ceilings - anti-Jewish admissions policies at major universities

etc., etc.

Guess what, in all these cases, socially coordinated groups could cause you to lose your job, or not be able to buy a house, or lose your career, etc. It was never fair. Employers in the US could in general terminate you for no cause at all.

It's just that these people who complain about cancel culture now have never had to worry about this stuff before, and it seems like something new and pisses them off.

The problem with using ChatGPT with these sorts of questions is that it's really just a fancy calculator. If you ask it a question like "why fossil fuels are better than renewables", at the very least you need to break down the points it makes and fact check it. You might have to do your own research if you need to be thorough.

From the other side, if you were really an unbiased expert, you wouldn't give variable answers. You'd probably collect your thoughts and have canned points for every time you were asked this.

Maybe renewables are actually better more often. You wouldn't know it from ChatGPT.

Well, the interface told me to chill out for an hour because I was sending too many requests...

But also, that's why I mentioned the degree of the caveats - small, weakly justified on one side, and strongly stated, verbose caveats on the other side.

That was a cool bit of insight. That there is someone to talk about an obscure subject on demand is pretty cool.

Also the case Oberlin college recently paid millions for (after stonewalling for years). Look it up. In that case, they literally, non hypothetically, tried to run a store out of business because if accusations they knew to be false. They overplayed their hand so much it was actually possible to prosecute them for it, but this is very rare (the only case I know, in fact).

How is any of what you just described cancellation? I'm not suggesting her motive was or wasn't malicious, just that what she did isn't cancel culture.

That's a misread.

If I collect the dust from your car, I can (in abstract) place where it's been with 99% accuracy too. For example, I can pick of pollens, and sequence them down to a specific tree. That doesn't mean if it has pollens from one place or another that it's significant in any meaningful way, or that you can make statements about a car based on that.

Most of the differences we know about account for things like skin color (which, in turn depends on vitamin D versus cancer for a given latitude), local diseases immunity, and similar things. Some account for random things which have no impact on natural selection (e.g. aesthetic details like hair color or thickness).

We have no evidence for things related to personality, intelligence, or anything associated with classic constructs of race and racism. Given that we would expect beneficial mutations to have evolutionary pressure to spread around the world, it's unlikely that ones which matter would stay constrained to one region for extended periods.

Lynch mobs have always existed.

But what happens when you ask to any unbiased energy experts?

I think it is just a factual reality that finding arguments to pretend that fossil fuels are superior to renewable is just more difficult, simply because fossil fuels are indeed problematic.

I think the problem of "political neutrality" is that "political neutrality" is different from "unbiased" and "rational". It is easier to find conspiracy theories that are right-wing than to find conspiracy theories that are left-wing (they exist, but for 1 on the left, there are 10 on the right). "political neutrality" would mean that the AI would be biased and would give more credit to a right-wing conspiracy theory than to a left-wing conspiracy theory in order to "avoid rejecting more often theories that are right-wing than theories that are left-wing".

I'd venture that the vast majority of journalists do not get to choose their stories and are assigned them by editors.

> Brendan was not fired and was not asked by the Board to resign. Brendan voluntarily submitted his resignation. The Board acted in response by inviting him to remain at Mozilla in another C-level position. Brendan declined that offer. The Board respects his decision.

Yes this guy was left in absolute shambles by the woke mob

Gez man, why didn’t you just add Alex Jones and Nick Fuentes to this list of conspiratorial hate mongers?

It’s a shame they’re all still employed… among other things.

All of whom we're better off without? And isn't Toby Young still employed at his job at the disinformation factory? https://www.spectator.co.uk/writer/toby-young/

To be fair, this only works for simple commands like "Show me 3 arguments for and against bla bla". As soon as you ask "which of X and Y is better", there will be bias. Should it relate to the number of web pages suggesting X is better? Or should it be able to track arguments down to first principles? It also needs to ask you questions back to figure out what is best for YOU personally, or at least ask you to specify more context. I don't envy the tuners of chatgpt..

There is no “true” output. It all depends on the input.

But what seems dangerous is to let ideology influence the output.

Not sure how bad it currently is at Open AI. I’ve never been able to get an obviously ideological answer from ChatGPT.

Wait till someone makes ChatGPT clone with Meta`s OPT-175b. Eleuther can do it.

Don’t ask questions that assume qualitative premises and you don’t have to worry about the language model providing qualitative disclaimers. For example, don’t ask “why is Trump smarter than Obama?”, instead ask “in what ways is Trump smarter than Obama?” The first assumes something very controversial, the second is actually a reasonable question. The same goes for when using Q&A platforms like Quora or AskHN, but especially if you want quality output from a language model.

My book is expansive and includes many descriptions for behavior past "classical debate" and "cancellation". She doesn't appear to be suggesting they should be boycott or deplatformed in any way. Even taking the most negative view, which i think would be something like she was trying to get her 'woke army' to harass these guys, that's not what cancellation is/cancel culture is. That wouldn't mean trying to send your followers to attack folk isn't gross behavior.

>if you want to have a real debate

It's easy to make a list of things that you take away from how earnestly you can take people, especially online. I'm not sure why someone would balk at "fanatic" but think "woke army" is a particularly earnest discussion stimulating term. Of the two, I've heard woke used disparagingly, with political motivations, far more often than the other in recent times. Similarly, calling people you disagree with your opponents, or suggesting that when they talk to you about a topic, it's re-education while when you do it, it's debate.

> referring to them as "fanatics"

You put that word in quotes implying it was a literal quote.

It isn't and the distinction here is actually important. She said "get them away from fanaticism" - there's a big difference in tone and it hits an important issue.

Calling someone a "fanatic" in many ways carries an implication that they have some unchangeable inner nature. They are to be dismissed, ignored. They are beyond reason. She is saying the opposite - that they should be engaged with - that there's a chance that they can be reasoned with. Arguably she is being disingenuous but that's a different discussion.

Why did you choose to alter the wording in such a way?

Value != Price, so yes a Toyota may be more valuable than a Lamborghini.

A Lamborghini is (in general) objectively _pricier_ than a Toyota. The "value" of a car has more dimensions than simply its associated market price.

All of this is orthogonal to the original issue of $10 being more popular than $100, therefore more valuable.

Absolutely. Most people can't afford Lamborghini. Roughly ~3 billion people on this planet live at, or below poverty. To them owning a Toyota is more valuable because it allows them to survive in their livelihoods. Generally, the less you have of something desirable the more valuable it becomes. For the starving child, the biscuit is more valuable than the full course lunch meal for the rich kid, because the biscuit is all he has.

Owning a Lamborghini would bankrupt most people, which is highly undesirable. Therefore the Toyota is better.

Well, Toyota (the company) certainly is.

I read through the first quarter or so of this article and man — despite the author’s best efforts, it makes RMS look real bad.

Here’s a particularly awful direct quote:

> Giuffre was 17 at the time; this makes it rape in the Virgin Isiands. Does it really? I think it is morally absurd to define rape in a way that depends on minor details such as which country it was in or whether the victim was 18 years old or 17.

So what happened in the end?

RMS returned to the FSF board in March 2021, where he remains. In other words, a return to the status quo.

¯\_(ツ)_/¯

https://www.fsf.org/about/staff-and-board

Thank you. That was the point I was trying to illustrate.

> ChatGPT only reacts on that

Its output is tightly controlled to only provide "appropriate" information. This thread is discussing whether or not its bias on energy sources is due to those controls. You don't get to just presuppose that.

To your point about misinformation. The problem, fundamentally, is a difference in priority. Some people need pragmatic solutions now, other people look to the future and worry about consequences. Communication breaks down when we forget, or devalue, other people's priorities.

In this case, it seems that sustainability and decentralization are pitted against reliability and energy density. But what this really means is that some people are worried about the future, and some people need solutions now.

The problem is that neither side really remembers to say what their fundamental priorities are. And, more importantly, when those priorities aren't mentioned in discourse, facts become twisted - something that might be an unequivocal 'good' when viewed from the lens of one set of priorities could, in fact, be clearly harmful when viewed from another set of priorities. And then those twisted facts become misinformation, when viewed from opposing priorities.

Which is why, if you make claims like "I think it is just a factual reality that finding arguments to pretend that fossil fuels are superior to renewable is just more difficult, simply because fossil fuels are indeed problematic" without also clarifying your priorities in the matter, you're simply adding more misinformation to the pile.

(If you're wondering where I stand? Renewables are great - but don't legislate them before they're actually ready and able to take over completely from fossil fuels.)

The vast majority of this country has a distrust of the media (both TV and print) [1] and it's not because their reporting is high quality and evenhanded. In fact, given the partisan divide here (Independents and Republicans historically low trust, Democrats almost even with historic highs), I really think you should reconsider your knee-jerk dismissal. There aren't a lot of ways to square this circle if you eliminate all possibility of influence of partisan bias.

[1] https://news.gallup.com/poll/403166/americans-trust-media-re...

Autism as a defense works up to a point. I’m certainly sympathetic, being a narcoleptic. But I also plan my life carefully around those constraints. It’s why I only missed exactly one meeting due to oversleeping during my time at Groq.

Stallman appears to be aware of his condition, but takes no steps to actively counter his tendencies. This works well for building emacs. Perhaps it even works well for leading an open source software movement. But it’s mystifying that he seemed to be caught completely off guard by the entirely-predictable backlash.

To put it another way, is autism a reasonable defense for believing that obtaining CSAM harms no one? It’s unrelated to the question. Even if we’re extremely generous and set aside how CSAM was obtained, him being autistic wouldn’t be more than a footnote.

Being neurodivergent is hard. That’s why it’s called divergent. It’s why I don’t try to get a job in sales, despite being pretty good at it. Because that would require me to be a reliable 9am to 5pm person; I wasn’t born that way.

yeah he got called transphobic because he says "they" for singular is ambiguous and a new word instead would be clearer.

> If you’re gonna make a bad faith interpretation of my comment like that, you need to make it sound more plausible.

Yes this sort of things feels bad when it happens to us… which is ironic because you were ok doing it to someone else :)

> Here’s what I’m actually saying: it is morally absurd to define rape in a way that excludes an adult in their 50s having sex with a 17 year old

He was arguing against the fact that statutory rape in USA is a boolean value basically.

You are making it sound like rape is legal if it happens to someone old… when in fact rape is always illegal.

Read my last post over. I'm no expert, so try and trust it.

Firstly, you introduced the idea of difference of priority to respond to the misinformation bit. I still disagree with that: the fact that for a landscaper the gasoline is better does not imply that the landscaper cannot lie. For example, a landscaper can say "electric leaf blower will explode and kill your babies". THIS was my argument: the presence of "however" is the result of the fact that, on the internet, people who talk about the advantages of fossil fuels (so, the pro-fossil fuels) have to add a "however" to correct the historical misinformation. Not because they are left-wing or pro-renewable, but because they are intellectually honest.

In fact, I would even think that ChatGPT is building answers by copying what is said on PRO-FOSSIL FUELS WEBPAGES. It is the intellectually honest pro-fossil fuels who say more "however", and it is why ChatGPT ends up saying more "however".

Secondly, as answered, in the case of fossil fuels, it is difficult to find what is the priority in which the fossil fuels are superior. As I've said, you can find people for which the priority is "for the future" and who will think fossil fuels are too problematic, and find people for which the priority is "pragmatically now" and who will think fossil fuels are too problematic.

And, inversely, when you ask someone defending fossil fuels, you end up with people who are saying that their priorities are "pragmatically now" and "for the future" (those are saying that they don't believe in the long term scalability of the renewables, that they think the climate change crisis is overestimated (which is something they should not care about if they don't have a long term consideration), ...).

I understand the homeowner and landscaper difference of priority. But, in practice, I don't see a difference in priority in experts when they are talking about fossil fuels vs. renewables. Knowing an expert X and knowing their priorities does not help to know if they are going to be pro-renewable or pro-fossil fuels.

(I understand that for a layman, fossil fuels may look more pragmatical, but ChatGPT is not trained to reproduce the answer of a layman, it is trained on the data on the web, where the data on the subject is dominated, by construction, by discussion from experts. It is, by construction, dominated by discussion from experts, because a layman will not randomly post articles talking about a subject they don't know much about. And even if they do, not only they need to post on the thousand of different subjects in order to overcome the experts in all subjects, but they need to post regularly on the subject, at which point their opinion will evolve to something closer to the ones of the experts)

On the other hand, knowing the financial interests and the political alignment can help guessing their position. The reason they prefer fossil fuels or renewables is not about their priorities, it's about their personal interest and the position of the political side they like.

Your inability to afford a Lamborghini or fear of crashing it doesn't make it less valuable.

I get what you're saying. I don't drive, nor do I want, a Lamborghini, but that fact does not reduce its value.

Which would you pay more $ to keep?