Anti-ChatGPT | Hacker News Reader

Anti-ChatGPT

github.com

99 pointsomarfarooq75 comments

I understand that the blacklist is probably just an example of things to hide, but it's interesting to note that if the filter was effective, it would block literally any and all content, including information criticizing and fighting against the nefarious media manipulation highlighted in this repo, but also the repo itself.

(note: this comment has a strong political bias, may reduce your lifespan, and the note you're reading might be considered "meme" content)

edit: Okay maybe it's not that interesting and that's the exact point the project was trying to make by showing censorship of Assange's tweets and I feel stupid.

Why wouldn't they just predict confidences and then set a threshold?

A more political piece of media isn't necessarily less honest or more manipulative. Similarly, a very supposedly "neutral, both-sides" opinion can be very manipulative. The best thing you might be able to do is filter out specific hand-picked subjects (like other commenters said QAnon or specific conspiracies), but I doubt you need more than basic keyword matching for that. Anything further feels to me like you're just tightening an echo-chamber.

That isn’t what censorship is.

The name "anti-ChatGPT" is rather weird, no? The project is a use of ChatGPT to block content that is not necessarily AI related or AI generated, and that the developer claims (probably sarcastically) not to like.

> With Anti-ChatGPT, you as an individual have the power to fight the AIs trying to manipulate you.

But it appears to do nothing of the sort. The example given immediately below this is a Twitter comment by a EU official being blocked by the script, as well as several other posts engaging with it. I don't have any reason to think these people are AIs or are using AIs.

So how is this anti-ChatGPT?

Maybe it just blocks talking in an overly convoluted way? ChatGTP winds up with that by default, but anyone talking in a very officious or even just extremely thought out careful manner could come off like that as well. I can see some merit to blocking that out, it’s not something that everyone wants to read, and in fact, lots of people’s own BS detectors are tuned the same way.

The fact I can't tell if this is a parody or a serious project says a lot about the state of the internet today.

Maybe it’s a bit of an art piece.

Maybe it’s a bit of both.

Or like most things good or bad depending on the way it’s used.

The poor proof-reading rather limited my ability to take it seriously. "ADS" in all caps...

Ads is short for advertisement. Therefore ADs (he doesn't mention ADS in all caps) doesn't make sense. But author might not be native speaker.

Dev Aggarwal seems like an Indian name (to be clear: from culture of India). His profile at [1] suggests he is one of the main programmers behind the social network Dara Network. This 'Anti-ChatGPT' feature could be tested and/or used on said project.

[1] https://dara.network/dev/

Pretty sure Google capitalizes AD in embedded content

lol @ filtering "politically biased content". If you apply this filter you're not neutralizing biases, just adding a bias towards the status quo. The status quo being by definition not progressive, that unfortunately means you've given yourself a conservative bias.

You've also offloaded the decision about what is political material to an AI - which has a biased view of what is politically biased based on its training data.

Sorry but this just sounds salty and like an anecdote (hn edition). If the status quo was conservatism you would expect to see that reflected in elections globally wouldn't you.

I took it as more about definitions: conservatism (in theory) is all about conserving the status quo. It's in the name.

As indeed we do.

No, what we see is popular parties with watered down centric programmes drawing in the same core voters for decades. Parties that actively advertise resisting change almost never win any elections, unless the political system you use as reference sucks so bad that it's a defacto duopoly.

Idk if you’ve noticed, but centrists aren’t usually pushing radical reformations of our society to make it better. Centrism is just “let’s have the illusion of progress without adopting any change that threatens existing power structures.”

I don’t agree. "Stop this newfangled bullshit and let's go back to the way things used to be" is a very common political message.

Are we not seeing a rise in far-right leaders across the globe? Italy, Israel, Poland, Hungary... not to mention growing conservative movements in however many countries not currently governed by far-right populists. (America is certainly one of them. I believe France is another.)

They don't win elections because of their conservative takes but because of protest voters. If you need a concrete statistical example: in the state of Hessen, the far right party AfD gained by far the most new voters from the social democrats which are firmly center left; ten times more than from the largest conservative party. This is an alternative to not voting or destroying your ballot paper because it actually reflects in polls shown on TV (which do not show absent voters). They just want to piss off other parties, it's not motivated by support for the (barely existing) far right's programme for the largest part.

Obviously you also have people who actually support the far right ideologies but lot of these voters were recently found to be willing to switch to a new far left party as soon as it runs at an election (catapulting it to 20 % off the cut at a federal level); same motivation. Protest votes

I'll be honest, I'm inclined not to believe this in the slightest. (Far-right voters voting far-left, or vice versa? En masse? In some kind of ill-defined protest?) But if you have any stats I'd love to see them.

Source for the far right to far left statistic is Insa Consulere, one of the large public opinion institutes. These companies are so reliable that the live results on election days used to announce a winner on TV are actually from their analyses at polling stations throughout the day, not from government institutions. If the figure should change it's because these people have changed their mind afterwards, but as it stands right now, a lot of far right voters consider it a better alternative. Percentage wise, out of all parties, the far right loses the most to this new far left party. This was already predicted before any statistics were in because they just want to flip the bird to established parties.

https://www.berliner-zeitung.de/news/sahra-wagenknecht-parte...

The voter movement from center left to far right is from the official election results from a few weeks ago.

Huh, interesting. (And thanks for the link!) I wouldn't foresee American voting blocs behaving like that in a million years, but German voters apparently have it in them.

That's because politics in the US isn't pluralistic but a static duopoly that's treated like a team sport by a lot of voters. In Germany the opposing player to the far right is the Greens for example because the two parties are both benefiting from the cosmopolitism vs communitarism cleavage that drives a lot of political debate. Voters will hardly switch between these two parties specifically. But if you are in a de-facto two party system where the opposing party is at the opposite end of each cleavage then you'll barely have any voter movement. It's entrenched this way for decades

Most democracies are pluralistic with more parties which is why the Germany anecdote makes more sense than the US one

> Far-right voters voting far-left

Considering the horseshoe effect, this is totally plausible.

https://en.m.wikipedia.org/wiki/Horseshoe_theory

> The horseshoe theory does not enjoy wide support within academic circles; peer-reviewed research by political scientists on the subject is scarce, and existing studies and comprehensive reviews have often contradicted its central premises, or found only limited support for the theory under certain conditions

The horseshoe theory 100 % applies to Germany and most certainly to other pluralistic countries too because the far left and far right are fishing in the same pond, for socioeconomically weak voters. They thus end up also sharing some views that you wouldn't expect to resonate on both sides like reducing immigration.

The people who desperately dispute the horseshoe theory are usually politically extreme themselves and hate the thought of having such similarities with the opposite end.

Yikes. Centrist yells at “socioeconomically weak” cloud.

I’ll just say this: horseshoe theory is wrong because when there are similarities they’re inconsistent and shallow. If that wasn’t true, political scientists would be able to write papers about the consistent and deep similarities, but they can’t so they don’t.

fyi good article on where the ends of the horsehoe have bent toward each other I think: https://www.bbc.com/news/world-55957298

Using traditional (and IMHO more correct) definitions of conservative and progressive, you are of course correct. However, at least in the US those terms have pretty different meanings to most people IRL. For example, the "conservative" position on Roe v. Wade was to change the status quo, and the "progressive" position was largely to stick with the status quo. Conservatives largely want to change the status quo on funding the Ukraine war, while progresses don't. I'm sure there are other examples, but anyway if you limited yourself to the traditional definitions, you'd be utterly perplexed by such a thing.

I think the person you're replying to is using the term in its academic form. Conservatism seeks to conserve [the status quo]. This transcends any colloquial gobbledygook.

You could be right, although given that it is explicitly about filtering contemporary political bias, that would be an unhelpful definition to work with. Also many progressives nowadays will say the same thing about issues at work for example. When people say things like, "let's keep politics out of the workplace" they will say that is impossible because that reinforces the status quo, which is a political viewpoint.

Wanting to undo progress isn’t progress. Going back to “the way things were” (not a real thing) is a pretty central idea to conservatism both in theory and in practice in the US. To your example: calling roe v wade the status quo because it was the law from 1973 to 2023 seems like a stretch to me.

As for the ukraine war example, yeah that one kinda breaks the mold. My best understanding of it is because the conservative position is to seek national strength, rather than the less concrete strength of unity. Sometimes that means war (e.g. iraq and afghanistan, nominally to prevent future terrorist attacks) and sometimes that means not-war as in the case where the benefit of keeping ukraine free, to the US, is unclear.

Ultimately i think you’re right - as a political ideology, conservatism is more fickle than just “don’t change things” - but since that impulse is definitely in there i think my original point stands

> calling roe v wade the status quo because it was the law from 1973 to 2023 seems like a stretch to me.

It sounds like our disagreement is more around "where do we draw the line" of "what is the status quo." 50 years of being the law seems like the status quo to me. For most people that are alive, that's the way it's always been (for anyone born later than 1955, that was the law when you were facing the decision of abortion).

There are thousands of years of history, so if the line of status quo isn't drawn at "how are things today" then we have to pick some arbitrary point and draw it there, and IMHO that makes the distinction between the two largely meaningless as there will always be an argument for why the line is in the wrong spot on any given issue, and moving the line on that issue flips the sides of which person is conservative and which is progressive.

But largely, it sounds like we are agreement. If the labels are more about "impulse" rather than position, then I agree.

> For most people that are alive, that's the way it's always been (for anyone born later than 1955, that was the law when you were facing the decision of abortion).

Unfortunately for us a ton of lawmakers and at least 3 of the members of the supreme court were born before 1955. Which is the problem.

Vernor Vinge predicted this - the "net of a million lies", filtered for human consumption by AI tools.

You are just scratching the surface on something very big here. Media filtering is going to define the AI era and shape our every experience.

Already has been.

This is why I think locking GenAI models down is fundamentally bullshit - it simply enables greater censorship and centralizes power.

I don't think hiding manipulative content is the right path here. Instead of burying your head in the sand, it would be better to make people aware of the manipulation and expose the techniques and messages they use to that end. What do they want you to believe, and why? That puts you in a much more powerful position.

> I don't think hiding manipulative content is the right path here.

True; shadowbanning would be more effective as it doesn't require a user to install and maintain their own filter(s).

> I don't think hiding manipulative content is the right path here. Instead of burying your head in the sand, it would be better to make people aware of the manipulation and expose the techniques and messages they use to that end. What do they want you to believe, and why? That puts you in a much more powerful position.

Fair take but would you argue same regarding ad blocking and adblocking circumvention? You can only focus on so much. There are people in OSINT community who exposed IRA projects. But who's gonna read up on that? Primarily those who can benefit from it. Very few people would read into that out of sheer curiosity. I'm a very curious person (ha-ha!) and I only have so much time. If I were to respond to all my curiosity signals I'd get no work done. And, most working adults on this planet (they got 'em voting rights therefore interesting target) have very little leisure time.

It would be extremely useful to know what people want you to believe and why, but how on earth do you figure out someone's intent?

Attempts to manipulate public perception are usually launched as campaigns with lots of content pushing the same message, often in different ways. In the case of mass media, for example, you could detect manipulation efforts by finding commonalities across multiple articles from one media outlet or by finding commonalities across multiple media outlets. It's possible you could distill this down to something like "From October 14 through October 23, content from media outlets X, Y, and Z seem to be correlated and are pushing the same message, and that message is ---".

Having that knowledge would be extremely powerful, effectively neutering manipulation efforts by identifying them as such. A person is far less susceptible to manipulation if they are aware of the belief that the manipulator is attempting to instill in their mind.

In reality you can't, but most people seem to think you can. It can be enraging trying to debate things even on HN because people often ignore the actual content of what you're saying and try to figure out what you secretly think or meant. My guess is the way people "accomplish" this with AI is by training it to make major assumptions based on other things. It will have an abysmal error rate (just like it does now) but the vast majority of people won't ever notice, and the person speaking will never even know they were censored, let alone have a chance at clarification or God forbid, explaining the nuance.

You should be able to build a profile of a person based on adtech, credit score, and social credit score/surveillance. With enough data on everyone we could easily determine intent with AI.

Though for some reason minorities always have questionable intentions, but whatever, it’s basically a utopia already.

Or, even simpler - replace it with the 1-2 word version of the message so that it's obvious and blatant. Like the 'They Live' movie. Long as it's a decent AI at reasonable speed, that sounds incredibly useful.

Edit: if those apple glasses take off, and we can live-edit text in the real world (if it's done, it'd probably be for sign translation etc) - those 'They Live' glasses could probably become an actual reality.

This doesn't work because the average internet user isn't an hn user. How many people here do you think would fall for an sms phishing campaign? And then ask yourself why they are as common. It's because this is a bubble and the average internet user doesn't even know what ssl is or has heard of the term astroturfing before. This is being abused all the time and it won't change.

I strongly disagree. Interacting with manipulative content both harms you, whether or not you believe you are "immune", and only serves to give the author of the content more views and influence. In fact, this sort of content often relies on being able to engage people who are don't like it and speak against it.

There are no easy solutions here other than encouraging others not to engage and read manipulative content, or producing your own counter-manipulative content. Consider, no reasoned argument or exposing of techniques is going to make a Qanon victim change their mind. Arguing with them just further entrenches their belief as they see you taking it seriously.

https://en.wikipedia.org/wiki/The_medium_is_the_message

There is very little point in trying to 'fix' content streams by filtering, censoring, or 'making them safe'. The content is just bait, the consequences you are attempting to avoid by filtering or improving the stream of content are inevitable and unavoidable so long as you sell your attention to that content provider at all.

The mechanisms that are used to massage you into consuming more and more of that media stream are the things you should be concerned about, not what the specific message encoded into any specific piece.

Example: You scroll tiktok for an hour. You might see pro hamas videos, anti hamas videos. You might just see videos of puppies. Whatever the the algorithm shows you, the effect of the specific videos it shows you is so small as to be irrelevant compared to the way the application causes rewards to be triggered in your brain and encourages you to continue to use the app. Worrying about the details of specific videos it shows you is like worrying about whether your Oxycontin chewables are cherry or blue raspberry.

I love that this is demonstrated by the filter completely redacting literally every tweet in the example.

A start of a “Coasean filter” as discussed back in 2007.

> “…technological device he calls a “Coasean filter,” that would compare all incoming marketing messages with the particular consumer’s current utility as determined from an analysis of the consumer’s current location, her response to prior messages, her own communications, etc.”

https://goldhaber.org/attention-marketing-a-coasean-filter-a...

Reminds me of that black mirror episode where people could block others from their sight, making them inaudible, and appearing as blackened figure.

also a subplot in Ghost in the Shell: Standalone Complex. in a world of ubiquitous bionic eyes, Laughing Man manages to hack the eyeballs of everyone who looks at him to replace his head with an animated gif.

The name is misleading

Weird project name but cool concept. Expect we'll see a lot more of this concept…

Yeah the project name is kinda clickbaity, ironically. Love the idea though

Filtering for clickbait and faddy gimmicks, I can no longer access this project's website.

If I turned this thing on, it would be obliged to hide its own existence from me, on the grounds that it's:

- potentially trying to spread misinformation

- sounds like clickbait

- contains politically biased content

How is it trying to spread misinformation or politically biased?

>privacy is a myth and big tech companies are using us as data to feed us ADs, viral content to reduce our timespans, and even rig elections

Anti-ChatGPT's author is explicitly opposed to the level of power that big tech companies have in today's society, and explicitly promotes individual privacy. You might completely agree with these opinions, but they are nonetheless political opinions.

Now, what separates a publicly expressed political opinion from "political bias"? I can only think of 2 responses:

1. A publicly expressed political opinion is less biased to the extent that the opposing view is also publicly expressed. (The Anti-ChatGPT README fails this criterion.)

2. There is no objective, meaningful distinction -- that is, there is no way to distinguish "publicly expressed political opinion" from "political bias" in a way that opposing sides could agree on.

Is there a third possibility?

You're talking about the author of the project. Sure he might be politically biased. But how is the tool itself biased?

Is ChatGPT a died in the wool conservative?

Maybe it would be fairer to say that if I turned this thing on, it would be obliged to filter out its own README. If it doesn't, then it (the tool itself) is politically biased. (If it does, then no contradiction arises and no accusation of political bias can be made -- though IMHO some irony does arise.)

I assume it doesn't filter out its own README, but I haven't actually tried doing this, so I could be wrong.

I can answer the second part: opposing ai is political.

It isn't opposing AI. It is AI.

AI deployed to counteract the impact of AI: "With Anti-ChatGPT, you as an individual have the power to fight the AIs trying to manipulate you."

Whether the product itself opposes AI or is AI or both is immaterial, since the description of the product on the repo takes a "politically biased" position.

Right. All statements about anything are completely biased. As Nietzsche points out, the preference for being alive is a bias. All of life is predicated on this bias, without which animals would simply stop eating or moving or breathing. It's all bias all the time, everywhere. And as you point out, even a statement of pure fact demonstrates bias as it was bias that led to it being expressed at all rather than merely standing mute.

Nobody wants to filter bias. Nobody. What they want to filter is content that annoys them. They call it bias because they don’t know what bias means.

Also “objectification?” There is no such thing. It’s not possible to remove a person’s agency with your words. When people say “don’t objectify people” they mean something like, don’t express disrespect for people.

Adults aren’t trained or encouraged to think philosophically, anymore. So political discourse is now more akin to the grunting of apes.

People are more trained to think philosophically than they ever have been at any point in history. If people don't think the way you think is correct you should worry more about your ability to explain.

Humans are apes so I don't know what you are hoping for. That's not me claiming to have transcended, it is me rolling my eyes at you for thinking you have.

> Also “objectification?” There is no such thing. It’s not possible to remove a person’s agency with your words.

Objectification isn’t the act of removing a person’s agency, it’s treating a person the way would treat an object. So it is absolutely possible to do that with words.

I think your reply was intended to be facetious, but I take it to be not only true, but to highlight that what Anti-ChatGPT actually intends to do is not to filter out all bias (which would leave nothing, as you observe), but rather only to filter out political bias that "deserves to be filtered out". And as soon as we realise that, we realise that the notion of "deserves" depends almost completely on a person's own political opinions.

So the big question is: Whose political opinions will determine what Anti-ChatGPT calls "bias"?

The moment we even consider that this question is necessary to ask, we realise that the underlying political nature of Anti-ChatGPT can't be escaped.

Sure, I'm sure Jordan Peterson would applaud you for worrying about who controls the censorship apparatus. I'm not being sarcastic, it's a concern, but I think you are bringing this and I didn't detect any of this attitude in the anti-gpt page. Maybe I missed it.

“AIs trying to manipulate you” isn’t a statement of fact, it’s an opinion about facts. AI exists, and content is created with it. Another opinion on the same facts would be “ai vying for your attention”.

So this is political in both the sense that everything is political, but also in the much more concrete sense that it is political.

All it does is filtering political biased content.

The example available here [1]; line one being "Kejriwal and Liberals whining that he was penalized for demanding PM’s Degree". I'll preface: I got no beef with these factions, don't even know them. It is just if I see someone attempting journalism and calling one side 'whining' I'm quite simply put not interesting.

If it were to block all political content, that'd be BS since anything can be deduced to politics or (vaguely) related to. Wikipedia English defines politics as: "Politics (from Ancient Greek πολιτικά (politiká) 'affairs of the cities') is the set of activities that are associated with making decisions in groups, or other forms of power relations among individuals, such as the distribution of resources or status." [2]

Also, there's a difference between biased and opinionated.

[1] https://gooey.ai/compare-large-language-models/?example_id=e...

[2] https://en.wikipedia.org/wiki/Politics

I think filtering based on words like "whining" (an emotionally-charged word, for which less emotionally-charged synonyms exist) is the easiest case of a general problem space that is dominated by hard, and maybe unsolvable, cases.

Suppose there are two large groups, whose members loathe each other. The groups are large, so each of them says and does a large number of things. Suppose a news organisation with many viewers publishes articles about both sides which are completely factually accurate and avoid emotionally charged language, but they choose to write only about the things said and done by group A that most people think are good and nice, and the things said and done by group B that most people think are bad, stupid or embarrassing. Is this news organisation biased? Is a particular article by it biased?

For a while email was perfect. Then spam came and it was unusable. Then spam filters came, and email is outstanding again (the end).

For a while the web was great, then everything was an ad. Will the ability for consumers to adblock be the technology free us from capitalist web enhsit?

With Google going to war on adblock... questionable.

Reply on news.ycombinator.com