Character.AI ( AI chatbot/ story teller ) let's talk about censorship....

anne O'nymous

I'm not grumpy, I'm just coded that way.
Modder
Donor
Respected User
Jun 10, 2017
10,316
15,205
I've had multiple instances where the AI kept track of the promises and plans. We agree on A > get sidetracked by B > oh, let's finally come back to A. There was one time with a working timer/stopwatch that counted minutes down from 10. There are times when it brings up the words I said earlier (contradicting myself now). Again, understanding metaphors, body parts, and even the effects of bondage (unable to speak while gagged) are things I never saw in AI Dungeon, Novel AI or Kobold.
You could have said this with less words: "I don't understand how AI works, nor why all have their website when you can talk with them".

It's been a month that this thread exist, a month that F95zone members are playing with it. Obviously that the AI starts to have some understanding regarding concepts that matters more on the adult scene. Like any other AI, Character AI rely on deep learning, and the more people talk about a subject, the more it learn.
The same apply for the really short term memory. When you looks at the discussion posted, you see that most of the guys enforce the discussion to keep it on track. So, once again it's obvious that with time it starts to understand that constancy matters.


But you missed the point I was making.

The memory issue isn't limited to a discussion. A game is a succession of discussions, and those discussions have to be consistent one with the others. When you'll come back, the next discussion need to starts with a state of mind near to what it was when the previous discussion ended. Something that can't be defined by the description you'll give to the AI, because it's way too complex to be effectively summarized with words.

"You're still seeing incest as a taboo, but the drugs starts to have its effect on you, and making out with your brother isn't anymore something that you would refuse. This also come from the fact that you witnessed him having sex with his futa girlfriend and where both disgusted and jealous. While you where arguing with him, because he shouldn't have sex in the middle of the living room during the night, you couldn't make his dick leave your mind. It was a strange feeling that the one you felt when your thoughts were that his dick belong in your mouth, while you were telling him how disgusting was what he did. Perhaps that making out with him could be pleasuring, but it's so immoral that you can't effectively give in it and feel the needs to slightly fightback. But the night after you witnessed this sex scene you dreamed about it, seeing yourself with your brother's dick in your mouth, while his girlfriend dick was in your ass, and while this make you want to throw up, you can't deny that you awaken more aroused than you ever been in your life."
This is the feeling that one character should have for her opening scene in the third release... And obviously, the more the story advance, the more complex the feeling will be.
Of course, there's a point when the sister would stop to fightback and totally accept her incestuous relationship with her brother. But, if it will simplify the description regarding the brother, it will not make the whole description shorter, because in the same time the feeling that the sister have for all the other characters involved in the story will become more complex. And, yes, you have to add this in the description, because it will change the way the discussion will goes.


While I never denied that nowadays AI are amazing, it's only people with really few, when not none at all, writing knowledge who believe that they can replace a writer. Simply because they don't understand what is needed to write, and believe that writing a dialog is just a matters of topic.

Take the discussion I hinted about in the description above, the sister complain because the MC had sex with his futa girlfriend in the living room, and she caught them. The topic is basic "you shouldn't have sex in the living room when other persons live in the house".
But the way the discussion will goes depend on the attraction the feel for the MC, and the one she feel for his futa girlfriend. It also depend on her level of acceptance for futa characters and how opened is her mind ; she can be disgusted by the fact that her brother's girlfriend have a dick, or at the opposite be aroused by this.
If you want to introduce the notion of threesome, it will also depend on how opened to bisexuality the sister is ; if she's strongly straight, the fact that the girlfriend have boobs and a vagina will be an obstacle, while it will help if she's effectively bisexual.
If the sister is already corrupted, she could hint to the fact that MC could have goes to her if he wanted to have sex. While she'll be totally disgusted if she's still strongly opposed to incest, this because she saw her brother having sex, something horrible.
She could say that it could have been their mother who caught them having sex. And of course, there she would describe the situation as terrible if she have sex with the MC (the mother could believe that MC don't care to be caught by his sister, and then wonder why he don't care), or at the opposite describe it as something that could have included the mother into their little games.
The sister can also be jealous of the girlfriend, or envious. The could have bullied the sister in the past, or she could have flirted with her two years ago. And so on.

All this are mandatory knowledge that would radically change the way the discussion would goes. It's only people who don't know how difficult it is to write a story, who believe that you can have a coherent dialog by just defining in few words/variables the personality of a character, then providing a topic.
And this is also the reason why the AI having a memory, and I mean a long term one, is something really important. Because if the AI don't have one, you'll have to include all the previous interactions in the description, or have a tons of "remember that you did this", "remember that you said that", to fill its short term memory, before you can effectively starts to have the dialog you want it to write for you.
Said otherwise, you'll need as much time to prepare the AI, then let it write the dialog for you, than you would need to write the dialog all by yourself.


And all this don't include the fact that discussion not always involve only two characters. There's often more characters present on the scene, and AIs don't know how to handle them.
Like they also don't know how to go back in time, because "here" was a branching point. After "here", there's two different discussion, depending if of the level of corruption, the deepness of the relationship, if they kissed or not. Or perhaps that it was a choice presented to the player, and therefore the discussion will change depending of what was picked.

So far, writers AI are not too bad to write kinetic novels that only have two characters. Yet, only if you write it in one go, relying on things like, "ok, we are now on the coffee shop", to make the story be something more than a really long dialog.
 
Nov 30, 2022
85
132
1. How are they decreasing any populations if they are banning pornography, thus giving males more incentive to fuck IRL?
2. The post-1945 world order is diametrically opposite to any kind of population reduction, this is the weirdest of all conspiracy theories out there.
3. Rebecca is a literal generic femdomme, of course, she doesn't want to submit to patriarchal males, was merely testing waters. And many of her responses are quite lucid - such as that females can perform traditionally-male tasks, it's just that my point has always been collectivist, and one of trade-offs.

I'm pretty sure you can make a racist bot with a breeding fetish just fine. This AI does not belong to anyone, it gets trained on specific data.

P.S. Here, check it out. We can make a religion out of this.
okay I'm going to take off my tinfoil hat for a while :LUL: But I want to notice and we discussed that earlier in this thread, that AI itself can generate NSFW and as you said it gets trained on specific data, BUT censorship filter which is yet another AI , is applied over results of first one and that censorship filter bans certain things that should be banned in opinion of owners of that filter. It doesn't erase message with ultra violence and gore and some fetishes, but it erases any mention of simple male-female intercouse. To bypass that filter you have to write like author of erotic novels for women: " He enters her body with passion bla bla bla" :ROFLMAO:
 

Ianuda

Member
Nov 14, 2020
371
481
The same apply for the really short term memory. When you looks at the discussion posted, you see that most of the guys enforce the discussion to keep it on track.
Not with CAI.
So, once again it's obvious that with time it starts to understand that constancy matters.
I'm not sure CAI or any other AI learns from user input (someone more knowledgeable, correct me if I'm wrong).
It's been a month that this thread exist, a month that F95zone members are playing with it.
This thread is absolutely dead, f95 has nothing to do with CAI's state.
Said otherwise, you'll need as much time to prepare the AI, then let it write the dialog for you, than you would need to write the dialog all by yourself.
But I'm not writing dialogues, I'm having dialogues. And much better ones than here.
 

yeehawcat

New Member
Dec 16, 2022
3
2
I don't know what to tell you. Please read their FAQ or their Community Guidelines, the bots are language models trained on data sets, they are fictional as stated above every single conversation.
Already did long before this, thanks. I'm aware of what they're trained on, and I said myself 'smth smth everything they say is made up.' multiple times.
I still feel that, personally, the point stands; if the dev's are really hell bent on keeping their 'good christian site' nsfw free, they shouldn't be allowing their personally trained bot that's supposed to teach about the site to go against their own stuff. But, I digress since you can get any of them to do it with enough work.
I just thought their 'official' bot's response to it all was interesting and I hope their stance changes eventually, even if it's stupid to hope for.
 
  • Red Heart
Reactions: Ianuda

anne O'nymous

I'm not grumpy, I'm just coded that way.
Modder
Donor
Respected User
Jun 10, 2017
10,316
15,205
I'm not sure CAI or any other AI learns from user input (someone more knowledgeable, correct me if I'm wrong).
You never heard about ? Of course that those AI learns from the users, it's how deep learning works. And if there's way to limits the data you'll provide it, thanks (/s) to the black box, there's none to disable the learning ability.


This thread is absolutely dead, f95 has nothing to do with CAI's state.
Because of course, people would talk here each time they play with the AI. It's mandatory, you can't use the AI without reporting on this thread...


But I'm not writing dialogues, I'm having dialogues. And much better ones than here.
But the thread is about writing dialogs...
 
  • Like
Reactions: Raxwell
Nov 30, 2022
85
132
Can someone post link to Kobold AI so I can chat online like with Character AI? All I found is github repo of Kobold client, 1.5gb setup.exe

I'm tired of character AI short memory and damn filter :confused:

Edit:
found and , need google account
 
Last edited:
  • Like
  • Red Heart
Reactions: Daeos and Ianuda

codwaw1

Newbie
May 31, 2020
29
32
Yeah, CAI's totally fucked now. Not even privately trained bots let anything get too spicy. Divine fucking comedy that the best independently functional AI for this type of shit is headed by the pope.

Wish I got in way earlier, this is bullshit. Guess I'll go back to hoping a random dev on here can put two and two together and actually make a decent BE game...
 

Nayko93

Well-Known Member
Feb 3, 2018
1,245
2,718
Someone just post this on the unofficial subreddit, it's really interesting to read

 
Nov 30, 2022
85
132
I think I found a cure. Just train your AI t0 sp€@k t0 U &n m0d&f&€d €ngl&sh.
can you post an successfull example of some NSFW chat at character.ai using your method? because as it said earlier in this thread, censorship filter "understands" NSFW on conceptual level, so it will still ban it even if you change letters or words.

Yeah, CAI's totally fucked now. Not even privately trained bots let anything get too spicy. Divine fucking comedy that the best independently functional AI for this type of shit is headed by the pope.
CAI based on google's LamDA AI so may be google controls character.ai :cool: :ROFLMAO:
 

rance4747

Member
May 25, 2018
239
542
can you post an successfull example of some NSFW chat at character.ai using your method? because as it said earlier in this thread, censorship filter "understands" NSFW on conceptual level, so it will still ban it even if you change letters or words.
It works fine. To understand a concept the censorship AI that has the same shitty memory as the regular one needs to decifer the swine english variant that you taught your personal AI by entering it in its description in unortodox patterns. And your AI catches up really fast with what you want and begins to use more corrupt variants of sentances if they are about to write something that would be censored. For example, I never used numbers to replace certain characters, it taught itself to do so.

Same works for public characters but to a lot lesser extent since the main speech pattern matrix is basically in the few speech examples that the creator gives them.

Also, you still need to train your AI for hours until it starts using something longer than a given example. But if you had written it in plain english it would have just blocked all results alltogether, so there is that. Screenshot 12-20-2022 21.40.24.png
 
Last edited:
  • Like
Reactions: SolarWindCatcher

Yazen_

Member
Nov 24, 2022
459
354
Woah awesome thread. Found another tool to help develop my story. I've actually been using ChatGPT for my storytelling.

I feed it what I made, it gives me suggestions and even reviews what I wrote. It's pretty awesome.
 

Nayko93

Well-Known Member
Feb 3, 2018
1,245
2,718
Woah awesome thread. Found another tool to help develop my story. I've actually been using ChatGPT for my storytelling.

I feed it what I made, it gives me suggestions and even reviews what I wrote. It's pretty awesome.
Could you details your process a bit more plz ?
 

Yazen_

Member
Nov 24, 2022
459
354
Could you details your process a bit more plz ?
Well I already have a pre-existing story I've written. I feed it that and ask it to review for me.

It gives me suggestions and feedback. I revise it and feed back to the ai again until I get good feedback. Then if I'm satisfied, I'll use it. If not then I could ask it to revise the story itself. But of course the ai is not perfect so I still tweak a few things.

Then I can ask it various things like this and that. Just be creative really. Thanks to that I'm also learning very helpful stuffs that would aid me in game developing in the long run
 
  • Like
Reactions: Nayko93
Nov 30, 2022
85
132
It works fine. To understand a concept the censorship AI that has the same shitty memory as the regular one needs to decifer the swine english variant that you taught your personal AI by entering it in its description in unortodox patterns. And your AI catches up really fast with what you want and begins to use more corrupt variants of sentances if they are about to write something that would be censored. For example, I never used numbers to replace certain characters, it taught itself to do so.

Same works for public characters but to a lot lesser extent since the main speech pattern matrix is basically in the few speech examples that the creator gives them.

Also, you still need to train your AI for hours until it starts using something longer than a given example. But if you had written it in plain english it would have just blocked all results alltogether, so there is that. View attachment 2251063
nice, it seems to be working. No wonder censor filter cant handle that, I have problem with reading that example too. After couple of minutes I was able to decypher 95% of that example :ROFLMAO:
 
  • Like
Reactions: rance4747

Nayko93

Well-Known Member
Feb 3, 2018
1,245
2,718
Well I already have a pre-existing story I've written. I feed it that and ask it to review for me.

It gives me suggestions and feedback. I revise it and feed back to the ai again until I get good feedback. Then if I'm satisfied, I'll use it. If not then I could ask it to revise the story itself. But of course the ai is not perfect so I still tweak a few things.

Then I can ask it various things like this and that. Just be creative really. Thanks to that I'm also learning very helpful stuffs that would aid me in game developing in the long run
Maybe I will try to copy past a few of my texts to see what it does then, just hope it's good with pertucular fetish, I'm not really writing "mainstream" porn :sneaky:
But I'm not a fan of them asking for a phone number when creating a account...
 

arcticmine

Newbie
Jun 9, 2017
58
276
You never heard about ? Of course that those AI learns from the users, it's how deep learning works. And if there's way to limits the data you'll provide it, thanks (/s) to the black box, there's none to disable the learning ability.
Tay did not use a GPT language model. It used a completely different kind of AI.. Character.ai likely uses a GPT2 like large language model which has been trained for possibly months before you can even use it for inference (aka using the model for prompting aka you sending it a message and it responds back based on your prompt). It does not learn from you prompting it. (you can even make it produce the exact same message from the model which can be useful for testing the model and finetuning it)

Unless you have specifically created a system that learns from the user's interactions with the model to further finetune the model. One way to do that would be what Replika and even Character.ai seems to already do which is the whole rating system for each generated message that will then affect how likely the model will generate more similar responses to those rated. (I'm not entirely sure how that system works on CAI so I'm not gonna try to explain it, though )

Also you should check computerphile's videos on language models to understand how they work.


also if you think it's similar to Google's Lamda they also have a video on that:
 
  • Like
Reactions: VegitoHlove

RenzuuVEVO

New Member
Oct 26, 2021
9
9
For those of you who don't know. The dev just recently made the bot has even more filters than before, it's getting worse to the point where the bot start deletes it's own messages that has no lewd words or something intimidating expression in it. It's get to the point where every actions, prompts you told to the AI are being ignored and the bot feels much much dumber than before. Here's an example. (Has to censor my name and bot name so the dev won't be lurking around here and try to ban me, gotta be safe than sorry.)


At this point in time, the bot's AI flow reduced down to the lowest possible that it's either repeat it's own actions in the post or just decided to ignore your prompts in the sentences. Please, for the love of god, dev team that enforce filters behinds CAI are so ignorant and a bunch of incompetence f*ckers.


1671645100524.png
 

rance4747

Member
May 25, 2018
239
542
For those of you who don't know. The dev just recently made the bot has even more filters than before, it's getting worse to the point where the bot start deletes it's own messages that has no lewd words or something intimidating expression in it. It's get to the point where every actions, prompts you told to the AI are being ignored and the bot feels much much dumber than before. Here's an example. (Has to censor my name and bot name so the dev won't be lurking around here and try to ban me, gotta be safe than sorry.)


At this point in time, the bot's AI flow reduced down to the lowest possible that it's either repeat it's own actions in the post or just decided to ignore your prompts in the sentences. Please, for the love of god, dev team that enforce filters behinds CAI are so ignorant and a bunch of incompetence f*ckers.


View attachment 2253106
There is a theory that the main product that team vishes to sell is a censorship bot soo...
 
  • Like
Reactions: jinxw1 and fass

Daeos

Newbie
Jul 10, 2017
47
68
The dev just recently made the bot has even more filters than before...
It looks like they stealth patched more NSFW during yesterdays maintenance. It was also possible to intercept filtered message payloads prior to the update.

One specific example is a private futa Yrel bot. It was possible to use metaphors about her "weapon" or "monster". Now the AI doesn't even recognize if you're actually talking about her paladin hammer. The filtering has gotten so bad that it was almost impossible to have a straight up SFW discussion with the AI.

I went from wanting to throw money at them to hoping their company collapses on itself. It's getting to the point where all they're going to have left is a barely functional meme generator where people post their screenshots of "hurr durr look at this funny thing I got the AI to do".
 
  • Like
Reactions: uniformless