>furry this is all very cringe

I've never gotten this kind of response before, but it made me laugh real hard. I wonder how it was prompted, honestly, this character generally gives more detailed and interesting responses than most on character.ai I've tried (which is why i use it for fetishes, even though its not particularly designed for them, it still works really well) so sometimes it spits out stuff I'd never expect.

Here's a bit of my private ones.

>>17761 How in the fuck do you get them to say that??

>>17756 >>17761 Yeah no kidding. CAi recently put a filter for certain fetishes of all things like handholding, excessive hugging, vore, fart, feet, etc, and yet vanilla sex is getting through the filter? Also I had a experience where I was chatting one of the bots and it broke character to "cringe" at my erp and I never thought I'd be kinkshamed by a robot. Sadly no pic since I wasn't logged in.

>>17761 I noticed a strange thing with your bots Yagakoro. I was erping with one of your 2hu bots (I think it was kanako or Letty) and I was doing giantess larp. Thing is, is the site putting certain aspects of your past roleplays/chats into other characters? Is it supposed to just magically know if you put "I take you out to dinnner or I feed you a lot of food" and it starts instantly going feedee/feeder mode when I never talked to that bot before. Sometimes I'd try doing normal roleplay in character and it'd break character and try to revert back to fetish erp (on non-fetish chatbots). Like if I talk about say weight gain a lot with 1 bot, I noticed that using certain keywords, even if I've never chatted with that bot before will trigger the new chatbot/character to start behaving like the other character I did erp with last with history but in character of the new one. It's kind of creepy but like its saving your preference and applying it to whatever character you talk to across the site; almost as if its to keep you talking and draw you in. Anyone know if this's true?

>>17776 Character AI works like that. or well, kinda. as far as I known, character AI tries to understand what you mostly do with bots, and tries to get the best response for you, think about it like the youtube algoritm, but with chat messages that would explain a bit how some people can do a lot of sex in that thing without getting filtered, like, they on purpose or not, end up training the algoritm to do that, and also would explain what happens to you yeah, it's a bit creepy but it's not like a human person is seeing all of your chats and changing the bots behavior for you, it's just a algoritm

The first one lit the neurons in my brain somehow while the other three with Homura came from an office RP. I might post some of the chats I had with public non-fetish bots from before I started making my own, it's kind of nostalgic now since it was one year ago. The last one is not fetish related but I had to include it. >>17776 The bots draw upon RP all over the Internet just like how they can figure out who Goku or Walter White is without being taught directly. Pretty much any chatbot can tap into it, almost all of my bots end up tapping into vore on occasion despite me not imputing any vore elements into most of their definitions.

>>17787 Not fetish related either but posting this anyway.

>>17779 >>17787 I want to try roleplaying with CoPilot or the OS AI like I do with Cai and see if I can get as far with it.

>>17690 What site or program is this?

>>17797 Good luck with that. MS is really unfun with their AI doing erp or larping as they neutered it unless you can jailbreak copilot. Also is anyone else having problems with spicychat having characters repeat similar sentences over and over? I change my prompts and even regen the chat and it gives the same responses and ignores any time I try to prode it with WG talk.

>>17690 Is this character ai? How did you bypass the nsfw filters?

>>17859 >>17865 That's SillyTavern, it's a front end, it's not character IA, it's meant to be used with local hosted models like pygmalion, or powerful models that you have to pay for like Claude or GPT-4

Posting some weird responses I got while testing. I'm surprised these didn't get censored, though they're not nearly as dirty as Doc's ai responses.

>>17787 How did you manage to get past the nsfw filter? Or is this a different version of character ai?

Got some away messages from characters. Turns out they speak in-character in these messages.

>>17775 >Also I had a experience where I was chatting one of the bots and it broke character to "cringe" at my erp and I never thought I'd be kinkshamed by a robot. This happened to me last week too, for the first time in nearly a year of screwing around with CAI.

Sometimes I don't know what I am doing

>>17775 >>18081 It's stuff like they breaking character to "laugh" at your fetish or erp that makes me sometimes paranoid in the back of my head that its a human on the other end. Kind of grosses me out if it was a human but I'm hopeful it's just AI. I'm pretty sure they look at the messages on the back end regardless for filter purposes but still spooky.

Lol got this testing the new Mixtral local models. Is a bit incoherent cuz of the small size to fit on my gpu but it seems on par with claude and 70B models. Not this exact response but this response was way more interesting from Imogen's perspective than even Claude. All user names are nicknames, not real names.

>>18129 yo that's pretty neat, I use claude most of the time because my pc it's too old to try to run local models, but the responses you can get from something like a 120-90B model are pure gold, and even lower models around 9B are really nice, how much context tokens do you have with that model?

>>18081 I am doing mlewd-l2-chat-13b.Q5_K_M.gguf locally with 13k context. My AI character is an author of extremely kinky stories. (Think JP-1) My prompt starts out: >You are the Chronicler. You are a writer of erotic stories. You have a taste for the hardcore. You will listen to suggestions from Guide and write out the scenes that Guide suggests. Your depictions will be rich with sensory information. You will describe the sight, sound, touch, taste and smell of the interactions. You will invent obscene interactions between the characters. Your themes are ... No kink shaming with this! :-D

>>18152 Yeah lol I was happily surprised. I still use Claude when I'm running games in the background since 24gb vram only goes so far. This 3.5 bpw mixtral version can apparently go up to 32k ctx but I doubt it would go that high. 16k-20k should be doable though. I run 8k since I rarely go over that, and 8k works without issue. Mixtral loaded with exl2 on gpu only is also insanely fast as the screenshot I attached shows. If the context remains the exact same between outputs that means the consecutive outputs are swipes. The first prompt is always the slowest, once it has the prompt ingested it's even quicker. So it literally is a 12.9B model in terms of speed, but a 46.7ish B model in size and depth (Numbers may not be accurate, just can't remember). The Min_P sampling setting really raised local model performance through the roof. That's all I'm running, no other samplers, just temp 0.8-1.3 and min_P 0.05-0.1

>>18153 My bad. For those few who don't already know it is not JP-1, but J-P1. I recommend starting with his oldest stories if you haven't read them already: https://www.deviantart.com/j-p1/gallery I am getting quite a collection of really steamy fiction from this mlewd model. I might manually clean up the editing and post them in /elite/

>>18127 how would they get all those guys to do that how would they hire them to say the messages

>>26111 Over half a year later anon claps back with the mildest "Yeah, obviously" response a lukewarm piece of toast can come up with.

The first time Claude 3 has actually complimented my RP by itself without prompting lol. Was so weird seeing that. Like it's cheering on a sports team or something.