News for hackers. Stuff that matters.
Re:AI Dungeon will block certain words, review content flagg... by Traubenfuchs on Wednesday April 28, 2021 @07:58AM (#26968331)
This is the peak intersection of "thoughtcrime prevention" and "corporate self harm for the sake of wokeness". It paints a grim picture of our future with castrated, sterilized and limited AIs that rat us out if we even want to talk about something that goes against what those in power deem acceptable. All of that for the sake of "protecting minors" -which minors, the virtual ones dreamt up by GPT3? Ridiculous.
Re:AI Dungeon will block certain words, review content flagg... by qayxc on Wednesday April 28, 2021 @08:43AM (#26968673)
Nah, it's a "damned you do, damned if you don't"-situation.

There's simply no winning this - if you don't apply preventive measures of any kind, your product will deteriorate into a homophobic, racist, sexist ultra-cringe clusterfuck in the blink of an eye [0].

All it takes is a handful of bored teenagers or middle-aged basement-dwellers to turn your product into something the vast majority of your clientele finds appalling.

If you are a company you have the right to choose what you want your product to be and they simply don't want certain headlines pointing at their product.

This has as much to do with "wokeness" as hardcore porn being banned from the Apple app-store, i.e. nothing. It's company policy and you can like it or not, but don't overrate this.

It's nothing new in the slightest and just like publishers can decide to not put books in print that they don't want to be associated with, this company tries to eliminate content they don't want to be associated with - that's all.

[0] https://www.cbsnews.com/news/microsoft-shuts-down-ai-chatbot...

Re:AI Dungeon will block certain words, review content flagg... by the8472 on Wednesday April 28, 2021 @08:55AM (#26968773)
Is AI dungeon used to train newer models/update the current model? If not then there's no risk of another Tay.

If they use a frozen model then one person using it to generate some NSFW content shouldn't make the output more NSFW-laden for anyone else.

So I don't see how they're "damned if they don't".

Re:AI Dungeon will block certain words, review content flagg... by qayxc on Wednesday April 28, 2021 @09:05AM (#26968856)
> So I don't see how they're "damned if they don't".

The specifics of their approach don't even matter - all it takes is one random Twitter user or media outlet to report on "immoral" output generated by someone to damage the brand image.

That's what they want to avoid and that's why they are trying to take preventive measures.

Re:AI Dungeon will block certain words, review content flagg... by wernercd on Wednesday April 28, 2021 @10:42AM (#26969907)
People make penis shaped blocks in Minecraft all day long... and that doesn't tarnish the Minecraft brand. Why? because people understand that it's not Minecraft the company doing so - or condoning the practice. And that's just scratching the surface of what can be done in games like that which allow "creative expression".

Why would you think that this company needs to Thought Police their gamers when thousands of others are able to host creativity based games without being dystopian?

Re:AI Dungeon will block certain words, review content flagg... by ljp_206 on Wednesday April 28, 2021 @04:27PM (#26973766)
For the record, I'm not sure where I land with regards to this change by AI dungeon. That being said, I find that arranging blocks in obscene shapes (let's even call it, generally, 'creating or arranging potentially objectionable content in-game'), an act that will likely happen offline, or on a privately owned server, is different than a game/toy/service where output is requested, generated, then sent to the user. (Trying to keep that description simple since I don't really know if that's the right way to describe it)

When comparing the Minecraft scenario to the AI Dungeon scenario under the purview of "doing creative things with games," I don't find the two scenarios to be much different. However, for one of those things, it's a possibility within the vast play space of Minecraft - you could even create genitalia and display profanity with just alphabet blocks - but for the other scenario, depending on how you see it, it's content that is generated, served, and potentially stored by AI Dungeon. I'd be concerned about the publicity if I was AI Dungeon as well.

Re:AI Dungeon will block certain words, review content flagg... by qayxc on Wednesday April 28, 2021 @03:09PM (#26972922)
> Why would you think that this company needs to Thought Police their gamers when thousands of others are able to host creativity based games without being dystopian?

Wow. You act as if that's something new :D

There's even an actual job for doing just that: it's called an editor. You know, the people who work in media, publishing houses, TV, radio, papers, magazines, that kind of thing. It's what they do. Everywhere. Editing content. Removing comments, changing words, making sure the views represented in the publication match the intentions and policies of the investors and parent organisations, etc.

You act as if something is new and evil just because it's done with computers. It isn't. It has been the case for as long as papyrus and cave paintings have been around. Back then it was don't anger the chief/king/Pharao, today it's don't anger the Twitter mob or the hand that feeds you. Same difference.

Games are a medium just like magazines or online blogs and comparing blocky dongs in Minecraft with explicit child pornography says a lot about your understanding of "creative expression". Really makes me wonder sometimes.

[dead]
Re:AI Dungeon will block certain words, review content flagg... by qayxc on Wednesday April 28, 2021 @04:13PM (#26973618)
So right to Godwin's Law we go then - yeah.

I get the feeling it's not me who is the radical here.

There is a difference between wanting to remove content that a company don't want to associate itself with and genocide.

For you to even suggest there's a connection says far more about your twisted logic and dysfunctional thought process than mine.

Also AI dungeon is a multiplayer game so yeah - YOUR freedom of expression is limited by OTHER players freedom of not being subjected to certain content. If you don't understand that, work on your social skills.

Re:AI Dungeon will block certain words, review content flagg... by Dylan16807 on Thursday April 29, 2021 @01:17AM (#26977713)
> Also AI dungeon is a multiplayer game so yeah

What? No it isn't.

Technically there's a "multiplayer" mode but that's fundamentally just people taking turns. Restricting certain kinds of content for that reason is nonsense, akin to restricting the kind of story you can write in google docs because multiple people might have been given access to a document.

[dead]
Re:AI Dungeon will block certain words, review content flagg... by qayxc on Saturday May 01, 2021 @09:55PM (#27011729)
So you don't do anything below XK-class end-of-the-world scenarios, so you?

Your massive overreaction shows no difference to your perceived enemies. This is a game and the creators want to limit the game world and the actions therein to exclude certain content.

This isn't about control over what people think. There are social norms and these change over time.

The "facts" in your message are the same kind of facts that your typical Karen shouts into the phone when they call 911 because there's a negro walking through their neighbourhood. Same level of paranoia, same unhinged out of proportion response.

When was the last time RMS was literally hunted down by a mob? Because if he wasn't, I have to ask who the real fragile flowers are - the ones that get easily offended or the guys like you who see freedom being destroyed and the 4th Reich rising because a private company wants to remove child pornography from their product?

Ridiculous.

Re:AI Dungeon will block certain words, review content flagg... by dang on Monday May 03, 2021 @04:22PM (#27030172)
Please don't post flamewar comments to HN. We're trying to avoid that here.

https://news.ycombinator.com/newsguidelines.html

(Also, please don't karen: https://news.ycombinator.com/item?id=26880612)

Re:AI Dungeon will block certain words, review content flagg... by wernercd on Sunday May 02, 2021 @05:45PM (#27019344)
> this isn't control over what people think

I don't think you understand that banning people over what the Thought Police find "bad" is LITERALLY working to control what people think.

> facts that your typical karen

and you are the typical ignorant fool who complains of "godwin" then resorts to "karen hates black people" drivel. Ignorant of the same calls "Karen" made a hundred times before but all of a sudden skin color matters to everything (hint: there's a word for people who only care about skin color... the funny little hypocrisy of the Woke Crusaders is the racism that permeates every move they make to fight racism by making EVERYTHING about race even when it's not).

> Ridiculous

Considering the other stories about AI dungeon over stepping? I'm more convinced I'm right about the Woke'saders and their march down the same path of others throughout history. It is ridiculous... but that's never stopped the religious in their fervor.

> Child porn

If that's really waht's there? report it to the police and take REAL action... not some half assed Woke'sader steps.

But "think of the childrend" is a close second to "ZOMG THE NAZI'S!" when it comes to justification for overstepping in the Name Of Good. So Sayeth the Wokesaders.

Re:AI Dungeon will block certain words, review content flagg... by dang on Monday May 03, 2021 @04:23PM (#27030178)
Please don't post flamewar comments to HN. We're trying to avoid that here.

https://news.ycombinator.com/newsguidelines.html

Re:AI Dungeon will block certain words, review content flagg... by contravariant on Wednesday April 28, 2021 @09:26AM (#26969068)
If they think that blocking certain words is going to prevent that from happening then well, good luck with that.
Re:AI Dungeon will block certain words, review content flagg... by jerf on Wednesday April 28, 2021 @04:29PM (#26973780)
It does prevent the news article from saying "And they knew about it and they did nothing!"

I'd be very worried about this in their shoes and acting very similarly.

Re:AI Dungeon will block certain words, review content flagg... by ThalesX on Wednesday April 28, 2021 @04:50PM (#26974020)
But then don't you have to answer to all the media queries?

"They knew about this problem, they even attempted to half ass fix it, but it's obviously their mind is on the subscriber money, and not preventing blatant CHILD EXPLOITATION!"

Then you throw some more money at a yet unsolvable task, and then in 2 years, when some other outlet gets triggered, you have to defend yourself once again.

"They've know about this problem for 5 years and it's still rampant in their community! Shame on you JERF INC.!"

Re:AI Dungeon will block certain words, review content flagg... by jerf on Wednesday April 28, 2021 @09:04PM (#26976265)
You're not entirely wrong, but if you make the story unattractive enough, the media will go do a hatchet job on someone else. They've only got so much time and they've got a lot of outrage to drum up^W^W^W^Wengagement metrics to drive, so you jst put enough in to try to make them hit somebody else instead.
Re:AI Dungeon will block certain words, review content flagg... by ainiriand on Wednesday April 28, 2021 @10:05AM (#26969502)
Correct. And they can go one step further, they could just use the input they currently do not flag if they want to feed back into the model.

This is not a technical consideration, it is just PC at its worst.

Re:AI Dungeon will block certain words, review content flagg... by qayxc on Wednesday April 28, 2021 @03:14PM (#26972968)
Oh come on. This kind of thing is done everywhere in media.

You either deny that games are a medium or you're under the impression that whatever you read in news publications, books, see in films and TV is the result of the unfiltered creative outlet of the producers. News flash: it isn't.

There's heavy editing going on everywhere and just because it targets a specific computer model in this case instead of a more "traditional" product like a novel, film or stage play doesn't make an iota of difference.

In fact this demonstrates the maturity of the technology and its use if it gets the same treatment as every other public media.

Re:AI Dungeon will block certain words, review content flagg... by selfhoster11 on Wednesday April 28, 2021 @09:10AM (#26968894)
If OpenAI was truly open as their name says, there would be no dilemma. The technology would be democratised and anyone would be able to run their own instance with their own censorship.

Centralised systems is the only reason why you get these problems.

Re:AI Dungeon will block certain words, review content flagg... by ergot_vacation on Thursday April 29, 2021 @12:45PM (#26983443)
Tumblr was destroyed precisely because of Apples puritanical views toward sexual expression (along with Tumblr's extreme laziness in moderating ACTUAL child porn). Discord may well be as well.

"Private companies have the right" always fails to account for the fact that these "private companies" are monopolies, with immense control over wide swaths of people's digital lives. No company can compete with them. No alternative to them can be created, because they will immediately be squashed, or fall into obscurity. Participation in these services is "optional," but more and more only in the way that having electricity or running water is "optional."

As a result, these "private companies" effectively become governments, ruling over digital countries you are forced to live in, and affording you little to no rights. And yet, they do this in an ACTUAL country, which DOES afford its citizens rights. Thus a conflict exists.

Ordinary "Private Companies" can censor whatever they like. Monopolies cannot, because doing so threatens free society.

Re:AI Dungeon will block certain words, review content flagg... by winkeltripel on Wednesday April 28, 2021 @09:07AM (#26968871)
Perhaps you don't see how once a product contains a censorship feature, it begs to be expanded. Each step appears reasonable, but at the far side, the censor is very restrictive.
Re:AI Dungeon will block certain words, review content flagg... by gameswithgo on Wednesday April 28, 2021 @10:29AM (#26969771)
Take the products in the category of forums. Most of them do censorship, THIS forum does censorship. A few do not, like Parler. If you like, take a tour of the forum products that do not do any censorship, spend some time there, report back on what it is like.

Which forum products do you enjoy and which do you not?

Re:AI Dungeon will block certain words, review content flagg... by DyslexicAtheist on Wednesday April 28, 2021 @10:53AM (#26970067)
> All it takes is a handful of bored teenagers or middle-aged basement-dwellers to turn your product into something the vast majority of your clientele finds appalling.

the Internet's original sin!

the irony is that Tech amplifies it but Tech won't (IMHO) be able to fix it with a technological solution (it's not a technical problem).

Re:AI Dungeon will block certain words, review content flagg... by sixothree on Wednesday April 28, 2021 @12:12PM (#26971024)
You act as if the two are mutually exclusive. I don't think the points in the comment you replied to are negated by anything you argued.
Re:AI Dungeon will block certain words, review content flagg... by trash3 on Wednesday April 28, 2021 @08:46AM (#26968690)
Do both then.
Re:AI Dungeon will block certain words, review content flagg... by DyslexicAtheist on Wednesday April 28, 2021 @10:51AM (#26970037)
Intel just announced "Bleep" to eliminate bad language in gaming. I am still waiting for this to backfire. The interface alone is comedy gold: https://www.youtube.com/watch?v=W9f0h4nB6VM
Re:AI Dungeon will block certain words, review content flagg... by sbierwagen on Wednesday April 28, 2021 @08:07PM (#26975804)
>AIs that rat us out

Note that Google has a "celebrity detection" neural net, along with their child porn recognition net. https://cloud.google.com/vision/docs/celebrity-recognition

A paranoid person might look at this, and the trend of every manufacturer putting NN inference accelerators in their cell phones, and wonder if unauthorized photos of certain persons are being silently flagged for law enforcement review.

Re:AI Dungeon will block certain words, review content flagg... by dqx on Wednesday April 28, 2021 @10:05AM (#26969501)
Thoughtcrime? It's content they are preventing, not thoughts.
Re:AI Dungeon will block certain words, review content flagg... by eplanit on Wednesday April 28, 2021 @10:43AM (#26969923)
And those thoughts are conveyed through content, no?
Re:AI Dungeon will block certain words, review content flagg... by dqx on Wednesday April 28, 2021 @10:48AM (#26969991)
...and the content is being policed, not the thoughts. "All content is just thoughts, conveyed" is a meaningless slippery slope.
Re:AI Dungeon will block certain words, review content flagg... by TheDong on Wednesday April 28, 2021 @04:46PM (#26973980)
I do not think that's the slippery slope being made though.

The distinction between "thoughts" and "content" here could be made as "whether it is shared with others".

For example, if someone has a private journal they keep using notebook.exe + a local "journal.txt", and in that journal they write a fantasy work (with a header "this is a fantasy") about overthrowing the government, should they be arrested for that?

What about if they write it in google docs, but don't share it with anyone? What about if they write it as a facebook post? What about as a public facebook post without the "this is fantasy" disclaimer?

In the case of AI Dungeon, my understanding is the stories are by default private, and obviously have an understood "this is fantasy" header by their very nature.

If AI Dungeon were creating facebook posts and the limitation they imposed was "you cannot post this to facebook if our algorithm dislikes it", I don't think so many people would call that "thoughtcrime policing".

What they're doing, as I understand it, is closer to someone requiring that notebook.exe can't edit private files that are deemed "bad" in some way, regardless if anyone else will ever see them.

Re:AI Dungeon will block certain words, review content flagg... by ModernMech on Wednesday April 28, 2021 @05:14PM (#26974291)
This crucial difference here is which party is generating the content. In the case of a private journal, it's one way communication from the writer into the journal. In the case of Google docs, you communicate with Google's servers, but Google is never generating any content for your document it doesn't want to. It's still you generating your own thoughts. Google is not on the hook for anything you write in a Google doc.

In the case of AI Dungeon that distinction is lifted. By using AI Dungeon what you're effectively doing is engaging in a creative process with the AI. It's no longer a one way channel of your own thoughts. Now AI Dungeon is actually generating content, and that's where things become problematic.

It's also the reason this is not thought crime or thought policing. You are still free to think your own thoughts, and you're even free to write them down. You can write down your thoughts in AI dungeon if you wish. But AI Dungeon is under no obligation to engage with your thoughts if they contain subject matter to which they object. If you want to engage in a fantasy with the AI about abusing minors, you are free to attempt the engagement, and the AI is free to say no. That's not thought policing, that's freedom of association.

Re:AI Dungeon will block certain words, review content flagg... by notahacker on Wednesday April 28, 2021 @09:56AM (#26969405)
On an internet where websites have clumsily censored objectionable content like the town of S---thorpe for the past quarter century and where even liberal democratic states administered by people calling themselves free speech conservatives support global filters for "adult content", a private project backed by the marketing idea that AI is inherently unsafe deciding that people using its procedural text generation to roleplay their child rape fantasies might be bad for PR despite its otherwise unusually permissive content policy doesn't really sound like a step change...

If anything, peak faux outrage over "wokeness" is achieved when this is what people get angry about

Re:AI Dungeon will block certain words, review content flagg... by vsareto on Wednesday April 28, 2021 @08:55AM (#26968769)
>"thoughtcrime prevention"

You do know the idea of thoughtcrime was something enforced by the state and not by a company that's free to control what their product generates, right?

>"corporate self harm for the sake of wokeness"

"Oh no, we lost the business of people who like sexual depictions of minors, what a fucking shame"

Re:AI Dungeon will block certain words, review content flagg... by taneq on Wednesday April 28, 2021 @09:45AM (#26969277)
> You do know the idea of thoughtcrime was something enforced by the state and not by a company that's free to control what their product generates, right?

Straying away from the prior context a little, the term 'thoughtcrime' along with others such as 'free speech' were coined when it was incomprehensible that a single corporation would have more control over the global population than any government. I don't know whether the time has yet come to re-evaluate which entities these concepts ought to apply to, but it will come.

Re:AI Dungeon will block certain words, review content flagg... by Loughla on Wednesday April 28, 2021 @08:47AM (#26968704)
Make your own game, then. This is a company deciding what they value and what they do not tolerate. You are free to make your own 'free-speech' version of this game. If you believe there is a market for it, why not?
Re:AI Dungeon will block certain words, review content flagg... by valec on Thursday April 29, 2021 @07:45PM (#26988260)
that's the issue. latitude got a very, very good deal with openai and without it the economics just wouldn't work out.

people can't host their own models either as even their free-tier model needs tens of thousands of dollars worth of compute hardware and memory to run.

Re:AI Dungeon will block certain words, review content flagg... by Traubenfuchs on Wednesday April 28, 2021 @09:00AM (#26968813)
I would, if I had the skills, maybe call it "LI Dungeon" (Liberal Intelligence Dungeon). Regarding the market: I am surprised there even is a market for AI Dungeon. Any output I have seen is complete trash.
Re:AI Dungeon will block certain words, review content flagg... by DyingAdonis on Wednesday April 28, 2021 @10:24AM (#26969715)
Yikes. I think it's time to finally be done with Hacker News.
Re:AI Dungeon will block certain words, review content flagg... by jasonlotito on Wednesday April 28, 2021 @09:06AM (#26968860)
It’s nothing more than what happens here on HN. It has guidelines, rules, restrictions on what you can say. They will even stop conversations that start naturally.
[dead]
Re:AI Dungeon will block certain words, review content flagg... by dqx on Wednesday April 28, 2021 @10:49AM (#26970006)
Strongly worded? Trite clichés aren't very strong.