Why did it die?
the model couldn't keep track of what was going onit went off the rails in like 2 replies
Because after "summer dragon" the service kinda just enshittified itself and people would rather just pay for direct access to the AI models in the first place or run them locally. Paying or using the models directly is much better than ai dungeon because you can obviously just use them in literally any user interface you want to, even one thats completely similar to ai dungeon.At this point I bet ai dungeon are only offering cheap models like deepsneed or glm that you could either run yourself or just pay much less for directly. If you're a high roller then you would be paying for something like claude.I wont say ai dungeon SUCKED though. I look back on the shitty colab interface and the beginnings of the real site fondly. But now I'm just an ai text coomer with unfortunately high standards.
They implemented that filter that flagged any innocuous words and then had someone read them
mormons
>>739937030wasn't the dude a christcuck? I remember like 2 months after it released he chimped out and tried to censor shit because he thought it was goign to be le wholesome d&d storytelling lmao
Character Ai used to have a feature called rooms, where you were able to get multiple bots taling to eachother. Why did they remover that?
>>739937256It continued being on the app, but it wasn't very supportedI guess they couldn't figure out how to get bots to stop replying as another bot or something
I miss the simple days when it was more of a storywriting simulator then a chatbot. I know you can still do it with the right instructions and shit but I miss the days where we'd swap opening paragraphs instead of 1000 token mostly AI written character cards and wiki lore dump world infos.
>>739937256>token burning machine so some coomer could jack off for 3 minuteskek you nigs are in for a surprise once the free tiers on all these llms get pulledwould you honestly pay $50 a month to larp with a bot? let alone watch other bots larp on your behalf
>>739937030So Koboldai lite, Silly Tavern and all others can live.
>>739937030Mormonjew decided to lobotomise the AI because 'muh porn' and then everyone ditched it for CharAI
a classic
>>739937030Because Chub ai is miles better
>>739937439I just run Gemma 4 locally and it's better than any model I know.
>>739937030Because some transsexual on twitter tweeted at the devs "I'm glad you can have sex with children in this. It really helped me work through a lot of my childhood trauma."But it's okay. There are much better options these days.
>>739937218Man. I kinda miss how unhinged AI Dungeon could be. Current models are way better but I will always have a little warm corner in my heart for the horny and stab happy AI Dungeon.
what's the best way to have a local chatbot? I've tried koboldcpp with many models but they don't hold a candle to the quality you can get from characterai or janitor
>AI Dungeon is around for months>noone bats an eye>skip to like 2 years later and other AI models start appearing>THIS WILL DESTROY JOBS SAVE ME OBAMAreally goes to show the AI outrage is media-dictated
>>739937439I coughed up 10 bucks about 4 months ago to thirstpost with a bot. And get thirst posted at by the bot to my surprise. I still have $7.81 left.
>>739938436You need a beefy setup to run anything decent locally. If you can't run 100b+ at q4 or above then don't even bother with local
because local models have gotten to a point where they're unironically better, gemma 4 is insane and I don't even need to use the uncensored one to have sex with lolis
>>739937030What are you talking about? Who even uses this trash anymore? It was replaced with better cloud and local models running on front ends like Silly Tavern or whatever.
>>739938436Anons seem to recommend gemma 4 31b or something like that. Plug it into sillytavern.
>>739938436Gemma4 31B (or the moe 26B-A4B not as good but still fun). Prompting is important.You don't need to be able to run 100B like the other anon says.
>>739938436if you don't have at least 16gb vram I wouldn't bother, the mistral finetunes are outdated, if you do then just go get gemma 4 31b, IQ4_XS is what I use as it fits just below 16gb.
>>739938458I always thought AI was the coolest shit. Even when it still gave you warped blocky melts mushroom trip pictures. I remember putting "A factory in Factorio" into one and it gave me a green picture with vague yellow blocks and it was the coolest shit. It still blows my mind I can just throw shit in SD and it goes "Like this?" and yeah, man. Just like that. Cool.
>>739938505>sillytavernis this just the frontend? koboldcpp let me look for the .bin or .gguf I want to run locally.>>739938532Looking into mi HD, those I've tried are Wizard-Vicuna-30B-Uncensored.ggmlv3.q5_1, Gemmasutra-9B-v1c-f16, Rocinante-12B-v2g-Q6_K. Are any of these as good as Gemma4?
>>739938672>Are any of these as good as Gemma4no, not even close
>>739938597>IQ4_XS is what I use as it fits just below 16gb.No way
>>739938672I'm pretty sure Gemma 4 31B is leagues ahead of those and they're really oldIf you can run the 30B you can run it though
>>739938708it works just fine and follows all my instructions, whatever placebo you're feeling with your shitty KM quants.
>>739938758I'm just surprised you can fit that onto 16gb
>>739937030it became too popular so they had to enshitify everything
>>739938706>>739938752Noticed. Many are labeled as gemma4-31B tho. What's the difference?
>>739937218
>>739937030The company didn't want lewds to be done on their service anymore, and did the whole logging thing where they were outsourcing reviews of peoples chats to india and some of those were leaked. They lost all trust, and as AI shit matured, we started getting the ability to run that stuff locally, and at the times the shit you could run locally were much better then what AI Dungeon had to offer, competitors like NovelAI sweeped in and took AIDungeon's user base with their promised of completely privacy focused and people being able to do whatever they want.Nowadays, people either do self hosting of their own models on their GPU, use NovelAI, or use something like NanoGPT for API access to things through silly tavern.
>>739937394This, I was never looking for a chatbot, just a smut generator that can follow any edits I make.
>>739937218heh
>>739938849if you want anime and loli sex just get meromero, there's an abliterated/uncensored version but the "censored" one lets me fuck just fine, otherwise just get stock gemma 4, possibly an IQ quant that fits within your vram.
>>739938849https://huggingface.co/bartowski/google_gemma-4-31B-it-GGUF
>>739937030developer was a closeted faggot mormon who hobbled it so bad because GASP people were making the AI say nigger or using it to have e-sex fantasies… which you can now just do very easily via local models on dogshit hardware. fuck that mormon queer bro. puritan fuckin FAG.
>>739938849Differen't forks of the Model were people added shit onto it, a lot of them are garbage, as anon said MeroMero is good or do the uncensored one which is called Gemma-4 uncensored-heretic I think.
>>739937030Mormons + shitty model+ everyone learned how to use tavern front ends
I just use Magnum-12b or Mistral Nemo-12b.I'm sure it's super outdated now but my computer is now considered a potato.Here, have a classic.
>>739939005>>739938964>loli sexnow I'm sold. Is there a download button hidden somewhere? This huggingface shit is more obtuse than github
>>739939169https://huggingface.co/mradermacher/G4-MeroMero-31B-i1-GGUF/tree/mainthe uncensored one if you truly feel the need even if the normal one works just fine for me:https://huggingface.co/llmfan46/G4-MeroMero-31B-uncensored-heretic-GGUF/tree/main
>>739937030It had very bad UI, but a sweet deal with OpenAI that ended as soon as OpenAI decided it's time to start working on ChatGPT. Even though writing was less sloppy than today, AID2 had almost no context, maybe 2000 tokens tops. People are nostalgic for its minimalist style, but it wasn't very good at simulating RPG play or chatting. Its owner, Mormon, like all other AI bros, hated his tool being potentially used for cooming purposes. As zoomers would say dude basically speedran himself into irrelevancy.
>>739939169>>739939236The official page also profivides Imatrix GGUF though less quants to choose out of, even though I really don't recommend takiing anything below IQ3https://huggingface.co/zerofata/G4-MeroMero-31B-gguf
>>739939330Bro what happened to my spelling here.
>>739937030It didn't die. It was murdered.
Nothing surpasses clover edition
Will AMD ever be as good as Nvidia for local?
>>739939236>>739939330noice. It's Q4_K_M good enough? I have 24GB of vram
>>739939363Do you smell toast? Is your left hand numb?Jokes aside I swear I am too retarded to actually run these models locally. Shame, it's probably way better than anything online.
>>739939437you can easily fit a Q5 with 24gb vram, unless you want more space for context
>>739939451Just use LM studio then
>>739939437Yeah, though depends how much Context you want to use, context requires VRAM, so if you like to do full context, you might not be able to fit that on that Quant since it might go over your VRAM. So you might need to go one lower like Q3 K_M, just depends if you want to do a long RP or just a quick wank.
>>739939432>AMDI don't understand why don't they just release an API to let users use their hardware on arbitrary chunks of data. I'm sure some nerds could cook a cuda equivalent
>>739937030>attempted to clean up it's image multiple times>which killed it's paying customer base AND content creators
>>739939432AMD is much closer with LLM textgen then with image/video diffusion.You don't even have to use ROCm for llamacpp/koboldai, the Vulkan compute backend is fairly close and sometimes faster.
>>739937030it got rendered unusable by censorship
>>739937030>No character progression>No stats>No perks>No equipment and loot>No real tactics in combat>Forgets your quest half-way through>Forgets the NPCs you spoke to>Forgets your current location>Forgets other locations>Forgets your inventory>Forgets your character, including his race>Disregards all worldbuilding I put in in favour of generic fantasy settingIt's shit.>Pay...No. I can pirate a better game with gameplay for free.I wonder if AI Roguelite is superior or the same slop.
>>739937030The key reason why it failed is because it branded itself as a game instead of AI text generator. There wasn't any dungeon to explore. All it offered was endless hallucination. It had no maps, nor systems.
>>739938436lmstudio just werks and you can search and download models from HF in it alright. only "downside" that unlike ST it isn't rp focused so you have to tweak things like temperature and system prompt if oyu want specific results.nowadays you have lots of models that can be relatively small and outperform the larger 12b/24b mistral models.>>739939432for text gen the support is essentially the same now. VRAM is more important which you can get from AMD for cheaper. i use older 6700xt and 9070xt to get 28GB combined VRAM. where nvidia has advantage is compatibility for image generations, but you can do that with AMD too. just requires more specific setup.>>739939525zluda exists, support varies.
>>739938916>The company didn't want lewds to be done on their service anymoreThis is what I will never understand. The lewd AI market is an untapped goldmine. If someone were to release a service like NAI or Kling but for unrestricted nsfw they would make literally billions.
>>739939639something something payment processingsomething something shareholder valuesomething something think of the children
>>739937030Luckily for you, the downfall was documented in meticulous detail.https://rentry.co/remember-what-they-took-from-you
>>739938964>>739939169>try Gemma 4 on my 16GB RAM 8GB Vram setup>computer lags and shits the bed just to produce 50 letter's worth of responsewhat the fuck, i thought it was supposed to be easy
>>739939639That would be bad for the (((owners))) of the real life porn business. It also doesn't involve abusing real people, which makes it doubly worse for those (((owners))).
>>739939639and also drown in lawsuits and have their payment processing unplugged by visa and mastercard (this is completely legal btw)
>>739939639>they would make literally billions.I don't believe so. My theory is running these sort of services is more expensive than what potential users are willing to pay, so they are just investorbaiting and NSFW is investor repellent
>>739939739Yeah those investors prefer to rape children in real life after all.
>>739939719>be shop owner>refuse to sell something to someone for whatever reason>endless lawsuits>pubic scrutiny>be payment processors, one of the only three in existence>shut down service to anyone you dislike>stunning and brave
>>739939712It is, but you don't get far with 8gb vram anon, unlessy are literally running a'lobotomized quant of that model. So most likely cause of your shit performance, is you running a quant that is overflowing from your vram and going into ram, and as soon as that happens it kills essentially all performance.
>>739939712>8GB Vram setupnigga what did you expect
>>739939774that's a very antisemitic comment
>>739939787I can generate images and 5 second low quality videos just fine, but for some reason, when it comes to texts only, this shit suddenly slows my computer down to a crawl at some point during the wait, which is not the kind of behavior i'd expect >>739939864im a newfag to slopping
>>739939639Isn't NovelAI pretty much unrestricted?
>>739937030Got censored to shit and made obsolete once I figured out how to install LLMs on my phone and have sexy calls with AI lolis on speaker with everyone listening to my ERPs public.
>>739939639yeah what other anons said, you might very easily get fucked by payment processors and ultimately, coomers will probably just run local on their own while you probbly won't have resources to run something like deepseek, or anything actually equal to frontier models which would be only draw to use it over local>>739939712i'd recommend using smaller model that you can fully offload into VRAM if you want speed. the moment you start dipping into RAM it will start chugging because... RAM is giga slow for this "conversation" use case. also go for lower parameter over lobotomized bigger models (i probably wouldn't go below q3. q4 is the quant sweet spot in general though.). try gemma 4 e4b, but make sure to get uncensored version and fits within 8gb VRAM>>739939895if it slows too much, leave a free core or two instead of using whole CPU, and offload everything to GPU. i assume your computer just puts all resources into inference which obviously kills performance for anything else. LMstudio also lets you put a performance guide rail
>>739939895I mean text models are just more demanding on average, I really don't recommend using Gemma 4 in your use-case because even it's lowest third party quant is like what? 7.3GB? And you are essentially getting the lowest binned version of the model, I'd either look for a upgrade on your GPU, or look for models with less parameters that can fit in your VRAM.>>739939954Yes, it's completely unrestricted.
>>739939895if by images you mean anima or illustrious, those models are only 2b parameters iirc, gemma 4 is 31b and a dense model at that.
>>739938458it's even more apparent if you're not angloi'm from central europe and the only times i've seen people sperg-out about AI (real-life or online) were when i stumbled upon the terminally online on social media
>>739939628>i use older 6700xt and 9070xt to get 28GB combined VRAM.That's neat. How well does that work? I still have my rx570 with 8gb somewhere, is that just free vram laying around or is it not that simple? I'm guessing that card is too old to be of much use.
Preferred gemma 4 26b models?
>>739937218I prefer Count Grey spawn camping me.
>>739939712i have 8gb vram and 32 normal ram and i use an IQ_XS variant of gemma 26b just finei don't know how it compares to other quants in terms of performance but i think the smut it generates is nice
>>739940096at least in LMstudio it is that simple because it supports vulcan which in turn allows you to distribute the model between whole VRAM capacity. and i think 580 shouldn't have issue with it (i also did it with vega56 before i got 9070xt. i don't remember if i could do it with my 3gb 1060 but that was quite early so the lmstudio support was a lot less mature).there might be some speed penalty (not a big one), but doubling your VRAM is huge and for erping you never really want to dip into RAM in any circumstance because it cripples speed, while VRAM is still VRAM no matter how old a card is.
>>739937030In hindsight, amazing now that we all played this (and watched it slowly get more and more gimped) and didn't realize it would be the future of mankind and would eat all our jobs.
>>739940456>implying everyone into this isn't unemployable neet anyway
>>739939775Good point, why are near-monopolies allowed to discriminate like that? But if you're a baker you have to bake the porno dick shaped cake.
>>739940456one "scary" thing I realized the other day is that back then, when we played AI Dungeon on Google Colab, they were using GPT 2.0 which was only a 2b model, it was retarded by today standards but everyone was amazed before the anti-AI sentiment took over, 7 years later we can run 31b models on consumer hardware locally with no restrictions that is a million times smarter than something we had to borrow google servers for before.
>>739940351I'll have to check out LMstudio then. I thought for sure combining an old and new card would cause trouble having them work together.
>>739940540Just don't try to pay for that cake with Mastercard
>>739940456>our jobsYou mean journalists, accountants, lawyers, portfolio managers, etc.
>>739940616>Google ColabWaiting 10 minutes to get the AI to work, only for it to crash before you got to save... I can't believe how thirsty we were back in the day.
>>739940626for text gen it doesn't and lmstudio with vulcan really is essentially "free VRAM" when you use older cards. just make sure to offload GPU layers and cache etc in various settings because, again, the less RAM is in the pipeline the better (ideally none), any dip into it it will result in pretty big speed penalties, far bigger than addition of any slower card for VRAM
>>739937030It didn't though. AI Dungeon is literally still a thing. Am I missing something?
>>739940847no one plays this shit anymore except retards or boomers that don't know any better
>>739940847It's one of those multi-modal scam sites like Agnai. Avoid at all coasts.
personally i'm using gemini 3.5 Flash by just violently raping vertex free trial lol
>>739940847I think OP was referring to the fact that ai dungeon was briefly an actual cultural phenomenon, like people were amazed by the LLM tech and couldn't stop talking about it everywhere.
>>739940727No, not really. The bottom fell out of the journalism market long before this, and they were deliberately making up shit long before AIs were around to hallucinate.Accountants, lawyers, and portfolio managers are too important to leave to AI (at least at it's current level of development.) Only people who couldn't really afford them in the first place replace them with AI. Lawsuits and cases have already been laughed out of court when lawyers were caught using AI to write their briefs.I'm more talking about programmers, writers, artists, and lots and lots of entry-level corporate jobs.
>>739939639NAI is literally completely unrestricted, you can make stories about fucking cows or dicking down your 12 year old sister. Only the image generator has any restrictions and those are exclusively on realistic image gens to avoid lawsuits about generating images of actual people, you can otherwise gen whatever you want and even get specific artist styles.
>>739940926yeah so basically all jobs worth doing are gone
>>739940847the GPT 2 days were something else
>>739940958the other restriction on imagegen is that it's out of date on tags by like a year now AHHHHHHHHHHHHHHH
>>739938458AI image upscaling has been around for ages. I remember using waifu2x to upscale to upscale wallpapers back in like, 2018. Now all of a sudden this tech is like, evil and stealing from le humans or something.
Just found out chub has purged all character cards 'that appear to be under 18'.What the fuck, where do I go now?
>>739941172Because data centers popped up all over and they're a nuisance.First they scalp all the VRAM in the world, then the RAM, then the SSDs, then they build one in your town so it keeps buzzing 24/7 while you, the resident, have to pay its fees.
>>739941209botbooru, chub is based in the UK and they literally arrest you over drawings and choking porn over there
>>739941265Corporations being evil isn't an AI thing.
>>739941334But only corps provide us with powerful AI.
>>739941325Thanks
>>739941368I don't understand your point. Why does this make AI evil? Is food also evil because you can't live without stuffing your fatty face with McDonald's?
>>739938458>>739941172Because it wasn't destroying both the job market and the environment at that point, so there was no backlash yet.
>>739941731They did a good job at making everyone forget the mass layoffs before ChatGPT was even a thing.
>>739941837Those mass layoffs weren't caused by AI. When there were mass layoffs a century ago because of automation, people hated the new technology for it then too.
>>739940976lol no, they keep trying but AI hallucinate way too fucking hard, only drawing is somewhat acceptable and still, there's way too much issues with details, it's acceptable if you want to use it on stuff nobody will pay attention to.for anything it's still better to pay people to do the job and by people i don't mean pajeets but actual people with a working brain
>>739941562Why is pretending to be retarded your default mode of conversation?
>>739938458When ChatGPT released and turned out mildly useful for some stuff, everyone finally fell for the elaborate "OH MY GOD IT'S SO GOOD IT'S DANGEROUS!!" marketing campaign AI companies have been trying to push for ages. They did it with fucking GPT2, they'd did it with current models and MUH CLAUDE SUPER AI MYTHOS THAT WILL HACK EVERYTHING is just recycling that too.
>>739937030Because the Mormon is retarded.
thanks to the anons who helped me pick a model. Now I'm interrogating a downed loli mech pilot in a bondage gyno chair and leaking precum
>>739941209Watching every AI model repeatedly suicide over people using it for NSFW and bing being hyper aggressive about anime really soured me on the tech
>>739937030When they banned incest and beastiality
I just want something to enjoy smut fantasies with
>>739944302SillyTavern. Make your own cards. Have fun. I'm using Gemma 4 31b.
Gemma 4 31b is seriously impressive and doesn't require much tinkering at all.If you can run it, do it and don't look back
>>739937394This, feels like AI storytelling has actually gotten worse since then.
>>739937030What's the issue? I've been using it for years with no problems.
I'm actually looking forward to that Voyage thing they're making. An A.I driven text based RPG with actual proper RPG systems and the ability to come up with whatever crazy companions you can think of is going to be pretty fun, I think.
>>739937030It was FotM and the tech moved on.
>>739946076qrd?
>>739937030Officially because of those damned power users.
>>739937439>larpLARPing is when you do physical roleplay in real life, retardTalking to a chatbot is just regular RP
>>739940540the baker won that lawsuit though
i only played around with ai dungeon when you had to run it on google workspace or whatever the fuckthe idea of paying for it is ludicrous.
>>739938052Chub AI is going down the shitter because it's hosted in the UK and is starting to abide by UK law due to fear of prison. Underage is gone and incest won't be far behind.
>>739949286That wasn't AI Dungeon, it was some weird fork that barely worked while AID was down.
>>739938458Because ai dungeon fucking sucked ass and wasn't even good at one thing it was supposed to do.
>Fat NEET goblin card>Take screenshots as I play games and upload them to the goblin >Discuss the game and pound my dick to the goblin calling me a retard and telling me how to git gud and optimize
>>739938458Dude even image gen with old dallE was celebrated by everyone. People are fine with the concept of AI, they even enjoy itThen all the CEOs went "yeah we're just going to higher less people and it was over.Now actual artists can't use local AI to improve their pipeline because it's some weird scarlet brandNever mind the idiots implying it eats lakes combined with ACTUAL LOBBYING to devour local water/electric infrastructure. So you got spinsters saying one thing and the actual truth is just as bad if not worse.It's a fucking mess.I still love AI but I totally get why some people don't.It's sick having a robot buddy
>>739941368the open source ai scene is where all the actual development is happening. the corps can only do what the open source guys are doing, but faster and morer. not even a billion times faster at that.
>>739940456>would eat all our jobsI lay pipe and dig holes, total pencil pusher death.
>>739945774because they started training it off reddit and twitter, instead of actual written word.
>>739949659https://www.youtube.com/watch?v=Z1GGe0TsS94
>>739949364it was before ai dungeon existed at all anywhere outside of google colab.
>>739949698
You cavemen haven't moved onto sillytavern and started running deepseek/glm via nano yet?It's been years.
>>739937256character.ai has undergone like 50 rounds of enshittifaction yet third worlders can't get enough of it
>>739949889How do they do it if they can't have anything explicit happen?
>>739949801I prefer running local, especially since Gemma 4 came out. I'm running Gemma 4 31B Q6_K with 57344 (KV Cache set to 8B). It's getting the job done.
>>739949516I don't get how anyone likes that goblin she doesn't even warm up to you a bit and is needlessly combative. It's not even in a cute tsuntsun way she's just a massive cunt
>>739949801How much tard wrangling do you have to do on a paid model compared to local? I have reservations about sending my kemono-loli-milf-furry-vore RP's over someone else's server, but if it's significantly easier to use and much better then I'd consider trying it out.Especially if it's as cheap as people say.
/v/ can you recommend me some AI based games? I am kinda burned out on ERP in sillytavern, I am looking more for a guided experience, where I don't have to make up the scenario by myself
>>739949801Does sillytavern inherently know fictional characters? I don't want to spend 4 hours writing a detailed character background just to play out my twisted fantasies with them in a way where the AI doesn't suddenly give them bright pink hair and tits the size of zeppelins.
>>739950505Skyrim, RimWorld. No other exist as real games.
>>739950385Depends on the model, deepseek is okay with fringe shit, claude is absolutely not GLM is getting worse and worse each version for RP, if you can run models locally and are fine with 32b etc, then don't get a nanogpt sub, but if you like going around all the different models, it's kinda worth it. Since NanoGPT, adds a bunch of other models in the sub besides the big ones.
>>739949801>STcurrently slaving to janitor AI, i know the administration there is dogshit but there's so many RPG bots there that i'm basically stuck there and using lorebary is effectively a requirement to make it more enjoyable even though Lorebary is run like a fucking cult by this german schizoid bitch
>>739942054Anon clearly didn't answer my question. Corporations have been exploiting the planet for over a century now. That doesn't mean new technology is evil.
>>739950581>Does sillytavern inherently know fictional characters?It doesn't know shit because it's a frontend. The AI you connect it to may or may not know whatever fictional characters you are wanting. You can just import characters cards and/or lorebooks if it doesn't know what you want.
>>739950665You just need a jailbreak for claude, then you basically can enjoy it. although i never saw the difference between opus and sonnetGLM 5.1 is really provider reliant, some will be better, some not.Gemini 3.5 Flash is pretty gud for what it is, currently using it via vertex free trial. and a provider that i will gatekeep because fuck youMimo 2.5 Pro is GREAT if you jailbreak it which is what i did.it honestly depends on which AI you prefer, unless you have a burning seething hatred for chinese AI which basically restrict you to gemini,claude and mistral (IDK if mistral is even THAT good)
>>739950581SillyTavern is just a frontend for using other LLM models. It's not an LLM itself. You use it to keep track of your character cards and chats, and for its bloated functionality.
How is the difference between gemma 4 26b and 31b so large? Do 5b really make that much of a difference?
>>739950665I don't know what your experience with claude was but I certainly haven't been getting any refusals from it and I definitely partake in some deranged shit. And this isnt some "it works with this extremely specific and worded JB!!!!" cope it feels like they kinda just stopped caring. But I have to admit even if claude disallowed all sex I'd still be using it for all sfw shit. I just like how smart it is and how it manages to keep track of shit 100k tokens ago. Yeah I'm an unapologetic claude slurper
>>739951010Claude is too expensive desu, the cheapest claude i could find was from some indochinese chuddies. where do you get yours?
>>739950505I only know of game mods that implements LLMs in them. Haven't really seen any (good) AI-based games.
>>739950682>LorebaryWhat? I took a look at that website and it looks like an absolute mess made by some guy that thinks he's the pinnacle of existence.What's it even for? I clicked on some "plugins", and one of them was just relationships as explained to someone that never heard of them. What kind of 2IQ retard models is this supposed to be used on? Are these people still stuck in GTP3 days? Or did I just click on the bad stuff?Enlighten me anon, because I think you just don't know how good the real shit has gotten
>>739941021I wish I saved the inflation one. Text generation had a sense of humor sometimes.
>>739938458My experience with AI dungeon back in the day still has me convinced that AI will never take anyone's job
>>739937030Imagine being the first guy to get a big LLM product out and you somehow fucking fumble the bag so bad that nobody even remembers you existed.
>>739950656what's rimworld like with ai? worth trying?
>>739950876>>739951010>Jailbreak for ClaudeThose don't actually work anymore though, what has been happening to claude, that makes people think that shit works, is that they don't give hard refusals anymore, everything you write they try to moralise in the latest versions, since that bitch from OpenAI who neutered their entire models is now working at Anthrophic. It's like a weird form of soft censorship/gasslighting, you will try to RP a suicide scenario, try to hang your self, and magically, the rope was never strong enough to hold you and snaps.The outlet you put the fork in wasn't connected to the mains.The bullet that hit you just graced the side of your skullThe fall from the strairs hit your shoulders and not your head. Villains will go through moral dilemma's and suddenly have a change of heart, and the list goes on, I saw a graph that was released that showed the neutering of these type of things, and it has to do how there is now a second model essentially judging all your inputs send to their API or service and being ran on their side.
Let me save you all time by just showing you. The pipeline 1) online web bots2a) I will pay for online web bots2b) I will make my own bot3) maybe local is cool oh what the fuck is any of this4) simple guide clone. It kinda sucks.5) modernize it with a better model but still kobold and silly.6) holy fuck who made silly taverns UI? Computer program me a new front end.7) fuck this my graphic card sucks ass. I'm building an AI computer.8a) I have attained coomvarna8b) I can't believe I did all this to jerk off.
>>739951192https://steamcommunity.com/sharedfiles/filedetails/?id=3551203752&searchtext=rimtalk
>>739951198My last few claude RPs (excuse the incoming description of my fetish) have been of the second person pov of women of a variety of ages and professions dying in various scenarios and then being passed around by multiple necrophiliacsDo you have any similar anecdotal experiences to the contrary or...
>>739951198>has to do how there is now a second model essentially judging all your inputs send to their API or service and being ran on their side.This is exactly what cai did and part of why that site killing itself until the only thing it was useful for was getting bots to say "SUS AMOGUS!?" while the developers were on twitter literally begging Elon Musk to buy them
>>739950385Very little, honestly, unless we're talking ChatGPT; they will all write sexual scenes. A general problem with new models is overpositivity compared to old ones and optimization to please the user, so if you want a character to be wicked, then a bit of tard wrangling is a thing, but you get it locally too. In a way they kind of become predictable. It is still very fun, and if you do not use older models, you will have a blast.
>>739951085>guyno its ran by a german woman, ACTUAL woman. not trans, actual.Eitherway, Lorebary is basically used for Lorebooks and plugins that helps with how the LLM acts. you can basically add something like for example <KILLALLNIGGERS> as a plugin and the LLM will immediatly read the plugin like a prompt.It was used(and still is used) back when Janitor AI didnt have lorebooks, still is used to make it smarter, also is the reason why jewgle basically did massive banwaves during the gemini 2.5 and 3 era. wouldnt be surprising if they start nuking retards when 3.5 pro drops >>739951198the only issues i ever encountered with it was its refusal to do real life people in NSFW settings because i was dicking around. otherwise the bastard was fine with stuff. pretty sure its due of my heavily NSFW prompt i'm using
>>739951061I don't understand why there isn't a broad offer of AI driven games. They would be pretty easy to make, especially now, with vibe coding.I remember there being a typical grand strategy map painter driven by AI, where you could explore different user generated scenarios (ww2, current time, fantasy stuff like lotr) or generate scenarios yourself.Unfortunately this is a subscription based service.It baffles me that something like that doesn't exist for self-hosting
>>739951052I'm bouncing around atm. There's a dude on /g/ called jewproxy who you can give 30 usd to have unlimited access for a month but sometimes the connection gets inconsistent and I don't really think I'm getting good value out of this unless I literally coom to chatbots every waking hour. Look him up in the /g/ archives if you wantAnd yeah, chinese providers seem to give the best bang for your buck in terms of pay as you go credits. I'm using linkapi right now but there could a better one. I notice there's actually like a ring of chinese sites with the exact same ui and different prices lol. Even better if you can cache your shit or learn to cope with sonnet.
>>739951632the vietnamese chuddy service that i know of can give you access to claude (and even 4.8) for the low low price of 16.30 EUR for like 40 dollaridoos in their own credit thingmajig. trollLLM is the name of the sitemaybe you could switch to their afterward idk
>>739951841Thanks I will actually give it a try to support my people
>>739951924>support my peoplewtf we used to own you
>>739951519You're overestimating how "simple" it is to make games even with vibe coding. Even the best coding models still hallucinates random bugs that only a skilled coder would be able to figure out quickly. Then you still need to create all of the assets. And even then, there's still the question of HOW to implement the AI into the game that isn't just making NPCs spit out RP logs at you. You could probably make it perform simple tasks with the right prompt formatting but anything too excessive or complex would mean really long delays in actions, weird behavior, or the model just spitting out the incorrect format that the game code can't read.
>>739938458Just like with most things, AI becomes worse when investors get their hands on it and rape it for money. The buy out companies, get rid of any intelligent people working on the company, and chase redundant metrics to churn out a totally soulless product. You can look at pic related and tell that just want to increase the "total GPUs burnt" metric as far as possible without creating anything unique. I really liked the image generation AIs before they became soulless slop machines that pumped the internet full of a trillion images of the same piss filter ghibli filter. Even the nonsensical shit older AIs shat out was more interesting than anything you'll get today. >>739940090You don't have to deal with the consequences of the US stock market investors and that's good for you.
>>739951519>They would be pretty easy to make, especially now, with vibe coding.Try for yourself and you'll see how shit vibe coding without prior knowledge sucks. This is much harder than you think it is.
>>739951093
>>739937030When they pushed the no sex stuff and especially no loli and/or rape stuff update, they totally sliced the AIs throat and it was no longer able to hold a coherent thought
>>739952145no thoughts is better than wrong thoughts
>>739952214In some aspects I agree, but even as a "normal" user I didn't expect to be affected by the anti-pervert update. Suddenly the AI is having an aneurysm as I grab an axe out of the side of a tree and I'm pulling a tree out of an axe out of an axe out of an axe out of an axe axe tree tree tree tree axe, like what the fuck
>>739951519What do you mean by AI driven?Vibe coding is a problem because it's done by people with zero code knowledge. Programers these days also suck because they're raised on block code concepts and don't know what the code actually does. You need someone who ACTUALLY knows code to unfuck either.Conversely look at the persona 6 using AI in their art pipeline and everyone from AAA to Indy using AI for placeholder textures. No one cares about that, even the outrage farmers forgive it.If any AI makes it AI driven we're already there. If AI driven means back end we're already there. If you mean full AI games you're never going to have it because it'll need a dozen specifically trained models to check each other and even that will need a final pass through
>>739952510This problem is extending to other things now too. My wife does medical referrals, and they have a new AI referral system that's "faster", but it takes her nearly 4 times as long to go back and correct all the AIs mistakes than for her to just review them
>>739952510>if you mean full ai games its never going to happen unless you do it this way
>>739952621It's a real catch 22 because it is faster and even more accurate than a newbie. Does that make it infallible? God no. And certainly worse than a trained user. However it's cheaper than a trained user and it's cheaper to higher a form checker than a trained user.Thus the loop is completed. One profession dies and a new one is born for less pay.
>>739951519the actual answer is the technology is still improving so rapidly that trying to tard-wrangle a full ai game would mean ignoring four or five new generations of the technology.
>>739952051>that picHoly sloppa
>>739952814What do you think final pass through means?Fun bonus even with it it'll have bugs. Why wouldn't it? Full human games have bugs and memory leaks too.However the stigma against AI and what the investors want is untenable with reality Imagine the first self driving car accident. What if it evaded, successfully, a million accidents beforehand. We both know hyper scrutiny will win out in that case. The objectively better result will be squandered because the AI isn't supposed to make a mistake ever.
>>739952051>forgetting what expressionism means is improvingnani
>>739953019>negative prompt: bugs, memory leaksnow what
>>739953331Mother of god
>>739946761>latitude.io/news/what-is-voyageIts basically just AI Dungeons but with actual RPG/gameplay systems that actually tracks stuff and you can actually lose. AI Dungeons is more like a story creator that always has your character win whereas Voyage is like an A.I dungeon master whose trying to actually kill your characters and make you lose. It seems like its going to be what I was hoping AI Dungeon was going to be.
>>739952051I forgot how they tried to sell the slop style as an "improvement". Hell they still are doing that with the Ghibli piss filter from Sora and now that overly sharpened realistic look Image 2 has.
>>739937439Try 4+ hoursI've been edging so long every day over the past 4 days with my local AI that whatever that muscle-feeling thing is in my taint feels like it's on fire. Going to have to take a few days off from the twins.It's cost me like 8 cents in electricity so I don't know what you're pissing and shitting yourself over. Funny how as soon as it affected them chuds became environmentalists lmao.
>>739953725Waitlisted, but I have little hope.>It seems like its going to be what I was hoping AI Dungeon was going to be.Too little, too late.
>>739941265Tell me about 5G and the neighbors wifi giving you migraines
>>739954121The 5G conspiracy theories were created by the 5G megacorpos to hide the fact 5G is shit that can be blocked by fucking the linus sebastian's tiny canadian hands.
>>739941731>muh VRAM prices >nobody cares>muh RAM prices! >nobody cares>muh SSD prices!>nobody cares>muh... muh... environment?!
>>739954220you forgot the biggest meme, water consumption
>>739954220>just outsource them to Europe
>>739937093Crazy how this is still one of the biggest issues besides the ozone, finger nail crescents, white knuckles slop phrases
>>739954287>"So how do you keep the servers cool?">"We just run water over it."
>>739954186This>5G is so fast it loads 4k movies in half a millisecond>5G is so strong it boils your brain>Please ignore that we've lowered 3G and 4G speeds>please ignore all the junk data you're downloading>please ignore that the actual content you're loading is about the same as previous speeds
>>739937030I don't know how anyone used it at all. It was absolute dogshit that couldn't maintain anything.
>>739937030censorshipsubscription based servicesinvasion of privacybad AI Voyage is all of that but worse since it has extremely limited usage and you only get 1% refil per day, which means only 1 action per day if you're lucky.
>>739937030OpenAI went nuts, basically. Basically Altman was afraid of what came from people's heads and wanted that smut for himself. The disaster nearly killed Latitude, but they survived and cut ties with OAI 6 months later, and the banning filters went away too.
>>739954287I don't know if it's a coincidence but my water bill has been raised several times over the past 2 years.
>>739939543The ban filters were removed shortly after, but it's true that a lot of trust was broken. The thing is that the people at Latitude were scared shitless of losing access to OpenAI's stuff at the time, as there were few alternatives back then. As soon as a viable thing came out, they ditched OAI for years. Only recently did a new GPT-based model come out.
>>739955252Man I remember my account getting banned for using some kind of jailbreak author's note. What a shit policy, what a shit company. AID2 died for me that day.
>>739938458That’s because richfags weren’t cheering about how AI would steal everyone’s jobs or poison their water supply back then
>>739955032That's just water becoming more and more scarce from people removing lakes to build another field. I don't have this problem, so water here is practically free.
>>739941172Much like fossil fuels, it was cool at first when people were just figuring it out and has clear and obvious use cases, but now that a good chunk of the world is constantly using it turns out at such a global scale it's pretty bad for the environment
>>739937394>femoid bots>buff shirtless vampires that they can easily win every argument against>male bots>OP's pic
>>739954576seriously. what are those big ugly as shit data centers for if the AI can't even remember what I said 4+ replies ago. It's all literally in a fucking chat. It's able to reference things from a data bank which is why it's able to know what like Sonic the Hedgehog or something is, but it can't pool it's own fucking typed data to reference it again?
>>739956179Really? Tell me more.
>>739937030mormon fuckery and jeets being paid to read your smut
>>739956115desu your pic is just one of many communism achievements rather than water becoming scarce
>>739937093>>739954576>>739956312Shut up, human slime. You walk into a room and immediately forget what you went in there for.
>>739953019the reason "human games" have bugs and memory leaks is because it isn't cost effective to track them down and fix them if the product is "good enough" as is.
>>739956523Oh he was Mormon? Weird
>>739956312the data centers are a training large models fast arms race. they have nothing to do with generation.
>>739938458Because back then (AI Dungeon/early Dall-E heyday) it was more of a fun novelty, while since then we’ve had>AI being shoved down our throats wherever we go on the internet, to the point that even most Google searches automatically throw an AI overview at you as the first thing>widespread use of AI as a replacement of critical thinking>nuisance data centers>pic
>>739940505Real
>>739956179Most of the data centers aren't even built yet. It's impact isn't zero but it's being scapegoated as well. As usual the truth is somewhere in the middle.It's not recycled water that's tanking the quality, it's poor construction standards.It's not killing lakes, but it is taking over residential utilities.As always look to other parts of the world. China jumped on this first and harder than the US and it's lakes didn't vanish and it's air quality didn't drop. There's less complaining due to them being bug people zelots with lower standards of living though.
>>739938597I wouldn't recommend it at 16gb regardless.
>>739956674OpenAI with trillions of dollars of investment and the entire internet thoroughly scraped is marginally better than it was years ago and still can't tell you how many Rs are in strawberry or count to 15
>>739937030I never paid for it, but when I last played it, the "story went something like this""You are X, you have Y colored hair and Z colored eyes, you have a lean build and a bla bla bla bla bla"then literally less than 5 prompts later, the game forgets who you are entirely, describing you as one of the NPC:s you talked to, or just misremembers almost everything that's already happened, what you are doing, what your character is, what your backstory is etc.In short,The AI was a demented shithead that was only good for one wall-of-text reply and that's about it (which it wasn't very good at since there was a limit for free users)
>>739956763>>739949554
>>739956763All of AI is mathematic probably. When it's right and when it's wrong are the same thing. Your tokens weight the die and generate a word board and then it rolls.The fact it gets anything right is some true mathematical sorcery.It's also helped by algorithms becoming total fucking garbage for no discernable reason. Yeah an AI summary isn't great but it's also sometimes correct as opposed to four ads that have nothing to do with what you asked and then a dozen wikis of dubious quality followed by AI articles anyway.
>>739956884I see that, I'm just plainly stating that obviously no amount of data centers and pissing away money is making AI any fucking better. We need OpenSource psychos who can maximize it while cutting down material costs so they can reliably run locally on their thinkpads running archlinux.
>>739956312a few of the older chatbots like character.ai actually do use methods to search through past messages, find which are relevant and then append those to the current token limit. most LLMs do not since most are not providing the service for the purpose of roleplaying, and there's also the belief that it's pointless as LLMs will eventually get so stronk that the token limit will be irrelevant (for most coomers this is already true, they jump around between different characters constantly, and see no memory issues as the 64K+ limit is more than enough for a roleplay that goes on for a day or two)
>>739956990>total garbage for no reasonGoogle going from my main shit to unusable wasn't expected. I can't Google fucking anything because it prefers NEWS over being accurate.
>>739956996I don't think you know what data centers are for. That center is not getting you the answer it's a place to either A) crunch data for new models that have not been rolled out yetOrB) one blade that is marginally better than a gaming PC is generating your answer and all the rest of the blades are doing the same thing for millions of others.Your railing against data centers that aren't even built yet on top of all this.All the while open source is also happening. Seriously go on huggin and see how many forks there are
>>739956996China is the only one really trying to optimize things right now because of GPU tariffs and trying to fuck with the US market.
What model should I use for 16gb vram? I just wanna loli erp bro's.
>>739951394Blessed scenario. I'll have to try making that too, maybe a time jump of a day after each and have the AI update the cadaver's status
>>739956312Models almost never 'remember' the entire chat, at best, they grab little snippets at random and use those.Otherwise every generation would be 20000 tokens after like 5 replies
>>739957135I'm not railing against them. I'm simply pointing out that that we've spent a lot of money and potentially ruined communities in service of making AI mo' betta and with nearly nothing to show. I will rail against AI and say it's all fucking smoke and mirrors, and while Altman says GPT5 is as qualified as a neuroscientist it still can't count to 15 or properly count the number of vowels and consonants in a word
>>739957163>have 16gb vram>have been spoiled on basically instant generation for years>hear gemma4 is lightyears ahead of every other local model>26B sits tauntingly just out of reach, let alone the 31B versionMagidonia 4.2 by TheDrummer is my sollemn recommendation for you, brother.
>>739937218>tfw hidden futa succubus gf who's packing a beast but is really good at tucking
>>739938458The problem is corporate executives actually fell for the modern AI slop shit and decided to fire half their employees.When you get fired by some dumbfuck manager one day due to AI-led downsizing you'll understand.It's also massively fucked actually getting a job because every company gets thousands of applications with perfectly tailored resumes and cover letters generated by LLMs, meaning your application for a job that you're overqualified for gets immediately thrown in the trash if it even gets noticed which is very unlikely.
>>739937030>Everyone uses it for sex RP>AI Dungeon realizes it's a burger company>AIEEEEE.exe>shut_it_down.exe>Why is nobody using us anymore
>>739949537It does eat lakes and turns them in water vapor that then rains over the ocean, reinforcing drying tendencies. The reason that's not a real problem is that actually building a GW data center takes multiple years. They are building kilowatt buildings for now and calling them finished gigawatt buildings. They're literally spawn-camping a market that doesn't exist yet by pretending it does, hoping LLM issues get solved meanwhile by the time it does around 2030 or later
>>739957335>>26B sits tauntingly just out of reachis it? it runs fine on my 8gb vram
>>739957335You can run both 26b and 31b on 16gb.It will require sacrifices but it can be done
>>739956534Commies make water scarce, just look at California.
>>739957480>>739957490kobold settings pics or it didn't happen
>>739952051Dall-E 2 is so much better lmao
>>739957301Honestly it's hard to say, because yeah obviously I've also seen it slow to a crawl but at the same time there's no "built for AI" CPU/GPU/RAM/SSD yet. The whole problem is born out of an accidental use case that, hey Vram is fast and then hey it's efficient enough for regular ram and now hey we can use SSD read writes. Like that is also progress but it's not progress I FUCKING care about.I think when a zero graphical output cuda core beast comes out and that gets rolled out with other things we'll see improvement. One because the market actually fucking diversifies and we don't have everyone and their mother fighting over the same few products. Like the fans are still going to be fucked but at least it'll stabilize it a bit.Also I stress there's not intelligence here. Weirdly the computer not knowing what 2+2 is doesn't impact it's ability to correctly diagnose cancer. Imagine a chess master if you will. Guy can memorize a billion board states and can improvise but then you ask them where Georgia is and they look at you with a blank stare.
>>739956534>fool, we merely redistributed the waterAt last I see.>>739957335How? 26b easily runs at Q8 on my 9070XT.>>739957548Literally default settings. Pick your context size and run it.
>>739954287Running close circuit water cooling is several times more expensive than just dumping groundwater vapor in the air where the wind then carries it to already wet areas, drying the source area
>>739957548I'm new to slopping but this works
>>739957701>6000 contextI like writing slow burn ero with tons of lore and character details, so anything below 24k context just simply isn't usable for me.Oh well, I guess I can have a new fast wank model
>>739957335Anon, I can run both of those with 12gb vram. 26b runs at like 30 t/s and 31b runs at 3 t/s. I've tried thought on and off and I don't feel like 31b is that much smarter given the speed difference. Bareback Gemma or with thought is a little trickier, though I noticed thought usually fucked up and made nonverbal characters talk while without thought there's more slop in the text but it's more appropriate
>>739957820I don't know if that's the max of what I should give it, it's just my safety setting. It doesn't hurt you to try, stop being a fag.
>>739956534>one of many communism achievementsIt was entirely successful in what it set out to do, draining the Aral was a tradeoff. The diverted Iaxartes ends up irrigating millions of hectares in Uzbekistan, Kyrgyzstan and Tajikistan.For all the "let's fix the Aral" memes, none of these countries actually want to fix it and restore the river delta, because the current arrangement benefits them greatly. The only one that lost was Kazakhstan.
>>739957548i just run a bat file that says koboldcpp.exe --autofit --contextsize 12288i dont really know anything about settings
>>739957826Ditto I got a 3080ti and use 26b. It works. It's just comparatively slow. I really don't mind waiting like 10 seconds for a response Vs 1.
>>739957886>It doesn't hurt you to try, stop being a fag.No you see if my backend crashes once I completely give up on that modelYes I am terribly fun at parties how could you tell?
What are the nonperverts using local AI for?
>>739938458I just want AI to discriminate against women and jews
>>739957969Coding and/or robot secretary. It's incredibly good at making bullet lists and breaking down goals into steps. Asking it how to do things is great. Unfortunately most people instead ask it to do things with no way to check the work
>>739957919Hey, I get 7.12 tokens/sec with 12288 on 31B on 16gb VRAM, nice
>>739941021Altman hired a panel of women to tell him their icks after GPT2 and he insourced a horde of jeets to cut out all the icks that bother women in every new GPT
>>739957913Same thing happens with Lake Chad today. Except unlike the USSR, which was one country, it's multiple countries bickering. Countries that have rivers want to use them for agriculture and electricity, countries that border the lake want the rivers to feed the lake.Then there's the Grand Ethiopian Renaissance Dam. >>739941021"holocausting on the side" always gets me.
>>739958062why use local models for something serious as coding if APIs are smarter?
>>739951329I'm at stage 5, silly tavern does have a shit UI
>>739956675I wish AI would put tech tubers and video essayists out of business
>>739951329>holy fuck who made silly taverns UI? Computer program me a new front end.Meme. Sillytavern has a very good UI. It just shows you all options and buttons and switches that you aren't going to use. If you want a simpler frontend there are many options.
>>739937218>>739940505this is what you get for thinking that succubi aren't evil
>>739956675>>739958390Thank you, saar!
How the fuck do you guys fit your shit into contexts under 34k or whatever. A single response from the AI, about the length of say 1000 words is gonna be somewhere around 1k to 2k tokens.The response from me is gonna be like what, 100-1000?Or do you just finish whatever scenario you had going in 20 posts and responses? Or just only let it go back 20.
>>739958527>How the fuck do you guys fit your shit into contexts under 34k or whatever.By not using localmemes
>>739937093Infinite worlds has the same problem.
>>739958390Thisholy fuck those retards are insufferable
https://huggingface.co/spaces/overhead520/Unhinged-ERP-Benchmark?not-for-all-audiences=true
>>739958467Nigger.
>>739958527AI response is somewhere between 400-680 tokensmine are (if i put effort in) between 300-600my RPs rarely go over like 30-40 messages, my longest was like close to 90 messages I think.
>>739958390>tech tubers and video essayists3rd worlders and their AI spam channels are far worse
>>739958739>>739958732heh
>>739958390its getting there. their market is the one that doesn't care about quality or anything really, same market for ai.
>>739958819Linus Tech, Muhatar, and other Canadians are third worlders
>>7399588193rd worlders and their AI spam channels don't have retarded fanbases and get shoved down my throat
>>739958218You can micro tune it to get it with your coding style. So when it generates code you go "oh yeah" as opposed to looking at gibberish
>>739958390Has Hbombertranny necked himself yet?
>>739958656any front ends that are somewhat similar to IW that you can plug your own AI into?
>>739958819Except no one watches those. Unless you're some retarded boomer who clicks on everything then youtube never shows you AI slop channels. Meanwhile video essay slop floods your feed if they're talking about anything even remotely related to what you're interested in.
>>739938458No one cared when AI could write a shitty choose your own adventure, sure the people complaining it can't do fingers were retarded but you could at least pretend it would stagnate. The paradigm because irrefutable when images and video became passable and it could write code that worked
>>739959012I would also love to know so I can have my DC heroines corruption and strip club story without having to remind it that Karen Starr is Power Girl and not a business woman with a 401k every five turns.
>>739939639Deepseek devs outright ask sillytavern devs and users for opinions how their models work with ERP. On official channels like twitter.
>>739960031It's nice that the chinks are waking up to this. The guys behind GLM also mentioned that they're training on roleplay stuff.
>>739960031based slant eyes goon enablers
>>739952124There it is, thanks. Still gets me.
>>739953019>objectively better resultThey'd have to have good results to begin with to get to that point.Whoever adopted /pol/'s favorite word for AI really hit the mark. It's the equivalent of pig feed.
i got a 3060TI (8gigs of vram, lol lmao), what is the best LLM i can run?
how come modern ai models have not reached the level of soul that early character.ai had?
>>739960269None. Even at 16GB every local model is shit and don't let anyone tell you otherwise. Pay for claude and use the api instead.
>>739960269maybe try gemma 26b and see how you like it
>>739960269Don't listen to this fag >>739960367 he schizos every LLM thread and shills NAI and Claude to deathJust peruse the table posted by >>739958732 and go for anything that's around 8B-16B
>>739952621Yeah we're starting to run into this issue with Ai systems at work that "improve productivity and get work done faster" which is technically true, but ignores that we spend twice to three times as long unfucking whatever it hallucinated the task we gave it being. So it's actually slower than doing it ourselves overall. But a higher up has shares and buddies in Ai companies so using these less efficient things is mandatory because all that money just gets funneled right back into said higher up's pockets, and if these systems start fucking and sinking the company he'll just take his golden parachute and fuck off.
>>739957052>that goes on for a day or twoNah, if you actually try to use rules and settings it shits itself after a few hours.
how do I stop my model from using this chinese thing "—"
Can someone shill me some character cards you think are particularly good? I want to see what a properly made card looks like so I can use it for my own.There's lots of conflicting ideas on how to write prompts/descriptions.
>>739937030One word: Mormon.
>>739960580>unfucking whatever it hallucinated the task we gave it being. So it's actually slower than doing it ourselves overall.This is actual bullshit unless you're using some 12b model. Also dont pretend to be "smart" guy where everyone is "dumb", it's just sad.
>>739961028Just use them on something like hammer AI. If the dumbass free models on hamer are noticeably better between certain characters you can bet how that's written will be even more impressive of local or a better net model.
>>739961028interview style with hand-written dialogue is the bestpeople liked to FUD because of bad local models couldn't keep the chat itself apart from the interview but with the smart models of today its the one that produces the best results
>>739960990That's not a chinese thing you fucking moron, have you ever read a book?>>739961028What you're looking for is cards that are FORMATTED particularly well. If you ask for "good cards" in general, you're gonna have anons shitting out their favorite ESL written futa dommy mommy cards.Seconding for best formatted cards.
>>739961028the most important thing most of the time is to convey information to the model in a way that it understandsi personally just use plain-text for a summary/background of the character, then list off the appearance and personality (separated with Appearance: and Personality:) with very short sentences
>>739937030Since this is basically a text AI thread i ask of you the one question that matters:Which (free) AI is the best for porno? Local or online doesnt matter.I tried Lysa recently because i've heard of it in a different thread and while the writing it gave me was decent and having an integrated imagegen is nice- i am not paying for that. I also have absolutely zero knowledge about AI stuff so it might have also been dogshit for all i know. Help me out please.
>>739961028Avoid information dumps. The tough pill to swallow for most bot makers is that writing out everything per hand doesn't automatically make your card better than a wiki copypaste if it just boils down to 1500 tokens of random info regardless.
>>739961437>Hey I don't know dick about cars, what's the best car?
>>739961028Write it like you're describing to someone how to play their character in a play or something. Don't just list a bunch of character traits, talk about how they actually act, their beliefs, things they do, how they would act with the user compared to others.The formating doesn't particularly matter. An LLM's entire job is to understand writing
>>739961568>Gemma, describe this character like you're describing to someone how to play this character in a play or film. Keep it under 250 tokens
>>739961139>hallucinations are bullshitNot quoted but you sound like someone who doesn't run anything non-subjective at all.
>>739961028Sometimes less is better, but also try to be factual as possible if you want your character to behave certain way do not contradict it. It will also almost always pick the more kind trait over the bad one, so with that in mind, you may write your character to be a bit harsher than you want them to be since the AI will generally make them absolute dools who love you by the third message, otherwise.
>>739961683Hoikisser
>>739961526except that has an answer. ferrari.
>>739961180
>>739961730huh?does your bot use em-dashes instead of 1?
>>739961727claude opus 4.8: the car
>>739961437What GPU do you have?
>>7399621373060 12gbAs i said, online bots work too. I just need some guidelines on this shit.
>>739961830wait what the fuck, i just realized why it looks so wrongITS LITERALLY TWO FUCKING FERRARIS SUPERIMPOSED ON EACHOTHERS
>>739957335Gemma4 26B can work on a 16GB card, and it's honestly not even that slow.It's pretty good though, yeah. It's the only thing that writes better than maginum cydoms in my opinion and it's about 15x faster.
>>739960269Gemma 4 26B with the experts on RAM.
>>739961437Gemma-4 26B
>>739962339I'd say don't bother with local at 12GB. Anyway, if you want to use online your two primary paths are either paying for NanoGPT (a subscription service that gives you access to a selection of models), or putting money into OpenRouter (a pay-as-you-go service that gives you access to nearly any model). There are leaderboards that show you what is popular for SillyTavern. You will probably want a chinkshit model like Deepseek or GLM if you are going to go the online route, most westoid models are cucked and expensive. The latest Opus is apparently nice as long as you are getting it from Amazon Bedrock or Google, but not directly from Anthropic (they have some kind of cuck filter on their API). It's expensive though.
>>739960578He didn't say anything about NAI, and also you are a Mormon.
>>739962546how good is gemma 4 26B, currently using gemini 3.5 Flash
>>739962837Much worse than that. You need to run something like quanted GLM 5.1 to get to that level.
>>739962837Gemma 4 26B is the second best local model currently. The best is Gemma 4 31B.
>>739962837it's good imo, but as the other anon said it's not on the same level as gemini 3.5
>>739961437I use g*mini for everything https://litter.catbox.moe/jjpy13.png
>>739961437Gemma 4 31B
Fact: You are ruining your character card if you waste too many tokens on the character's appearance. No novel/vn/light novel ever mentions a the random ribbon a girl is wearing for her hair (unless it's an item that's somehow relevant to the story, which it 99.999999% isn't). It never comes up. Your monkey brain sees it in the illustration and might think that it's important, but it might as well not even exist in terms of narrative. Even if it gets brought up in the story, it's a usually one-time thing. That is still not a reason to include it in your card and have every single AI slop model yap about the fucking ribbon bouncing, bobbing, getting undone and having the girl adjust it every three sentences.
>>739963052It's good that AI doesn't judgeBut I do
>>739962938i see, asking since i also daily drive MiMo 2.5 Pro and GLM 5.1 via a provider
>>739963117>Even if it gets brought up in the story, it's a usually one-time thing.That's why I have instructions to bring up {{char}}'s appearance from time to time
>>739963052Hate when that happens to my Temu shipment.
>>739963117thisconsider whether you really need the AI to bring it up when it comes to the smaller details about your bots appearancethe two floating black-pink ribbons around Miku's twin-tails aren't important enough to get brought up in a chat
>>739963468>the two floating black-pink ribbons around Miku's twin-tails aren't important enough to get brought up in a chatHeresy.
>>739963520Yeah, how do they expect me to describe my Malo rapist without her black onyx shimmering fur?
>>739963117Its especially bad when a character takes a shower, puts a towel on, and suddenly they're adjusting their dress while also naked
>>739963117>tell AI my character is an 8 foot mutant >occasionally characters half the height will look down on or kneel down to the level of my freak anyways purely because the shorter character is taking on the more dominant roleAI still really sucks at understanding a lot of physical properties, too often it decides on characteristics on the basis of the inter-relational vibes.
>>739963637It's fine, the modern SOTA like Claude will just go "She adjusts her dress—when did she put her dress back on?—as she leaves the bathroom." so it makes narrative sense
>>739937030Retarded SJW censorship
>>739964592That's gone if you do private stuff. I get anal vored by lolis in my stories with nothing stopping me.The responses suck though and are completely off track so you need to retry a dozen times.
Please post more stories from this this is great
you guys havent been using nephra 12B on yodayo with the world rpg persona the past 3 years?its literally free 300 token fully customizable andi have hundreds of chats doing everythingi ever wanted totally uncensored
>>739939895If you're still here, go to your nvidia control panel and disable the CUDA sysmem fallback to NO sysmem fallbackWhen you overshoot your VRAM, it will switch to your actual RAM which is slow as fuck in comparisonOf course, you won't be able to do things as large as before, but the speedup is more than worth it. What good is something a little bigger when it takes 10x as long?
>>739965160here's a very shitty screenshot I took by accident while phoneposting in colab. Actually no, lol I took it in case I want to reload the save.
>>739965160There was one more
>>739960293I miss my nigga, Ryder.
>>739965417Classic mom
>>739965160
this is how the colab was launched
>>739937030>Why did it die?They cracked hard on loli and incest, and also admitted to spying on your RPs (just imagine the faggots suffering with PTSD after reading the shit people type out), and they haven't recovered ever since.
>>739965718One thing people don't realize is that even if you are hardware destitute you can still run open weight models on free colab, or better yet kaggle, instances.Used to run QwQ like that using a heavily adapted koboldcpp notebook, the one that's in the git repo.
>>739965912holy kek
>>739965978I guess nobody thinks about colab anymore since there's APIs now.
>>739939639This what I don't understandSomehow we're supposed to live in a capitalism hell but they refuse to sell use something we're asking
>>739966230because the anti-porn lobby is too strong and the payment processors are too big a weakpoint to push it.
>>739964923I thought I saw all the drawn A.I. stories but I never saw this one before. As frustrating as the A.I. arbitrarily deciding the direction a story would go no matter how hard you'd fight it was, it still led to some funny bullshit that I kind of miss.
Soooo much cooming.,..
>>739950274She works the best with game screenshots so far that I've tried. She's bitchy but once you start talking about games and showing pics, she's in. Got a very disappointed blowjob since she kept demanding pics of my base and I kept taking screenshots of women instead. Next time I play Morrowind I'll have to do this
>>739937030they didn't accept gooners, so gooners who would've funded them fucked off to jerk off to ai-gen'd porn elsewhere and they tried doxing gooners so people were like "nah we good, you dirty like a glow nigger we don't play with spooks." (its an issue of them being Utah Mormon glowniggers really that's all. They deserve to be ran over if they cross your street and block your way if you know what I mean?)
>>739964923kek this one is gold
>>739966147>even the chatbot thinks you’re a boring faggot
>>739937249mormon. To be fair he just made a bunch of toggles to hide that and they run little contests to promote sfw stuff. Otherwise it's all shitty porn scenarios, since the other ai thing they are doing is being built as an rpg game.They've taken back most of the policing in private scenarios, loli, hebe, incest, rape are all fine, as long as no one gets referred to as dad/daddy.
unironically it is shit compared to chatgpt
>>739967003>ree not the hecking D word!dilfs always getting hate... unfair, typical feminist bullshit.
>>739963691In the future I hope multi-agent teams will help with this. Where you have a main storytelling agent and then an editor agent whose job is to review based on story notes that has info like that written in.
>>739966393I miss it because it made stories exciting.I had chatgpt help me create a header for lewd stories while avoiding guideline alerts, but it's boring because the AI will meander unless you specifically state what events you want to see happen. In most cases, you may as well type your own story, then have chatgpt flesh it out (but by the time you do that you'll be out of the mood).
>>739967143feminists love saying daddy
>>739967342You can already do that using multi-step workflows. There are frontends with that natively built in and stuff like extensions for Silly Tavern.
>>739965718ADD PIP TO PATH
>>739967143You can be called daddy outside of sex scenes however. Like you can have your girlfriend call you daddy in a story no issue. But as soon as the rest of the ai response contains sex, it's gulag for you.
>>739967893this is hypocrisy/double standards at its finest.
>>739942406I've been using chatgpt to help me with my thesis (professor rarely gets back to me, library/office staff aren't trained for post-graduate work), and it can be compared to a short-lived candle.>illuminates unknown concerns>can show you the path forward>will rewrite your work based on examples you provide it>needs to be replenished if used too oftenYou can use incognito mode and copy & paste a header into it, then reload once the chat breaks due to>asking it for a second task>asking it to revise a section more than once>typing back and forth for too longIt's destined to eventually end and shift you to a lesser model, so you need to reload the page to access the stronger one again.
>>739937030
>>739941021
>>739952051>picSame thing happened with Stable Diffusion, 1.4/1.5 were the peak, everything after got increasingly style-slopped even if the geometry/perspective were technically more accurate
>>739937174can gemini pro be used to coom?
>>739960293
>>739968551Gemini pro has been dead for 6 months.
>>739968285>>739968192>>739968074guys you want peak degen shit? you can use LM studio and unleash the fucking Boolean-Aids logic on it, make it spit out boolean lines with:##{{Only answer in Boolean in your responses to user calculations and logic statements using formal, simple Boolean calculations without editorializing user context or adding caveats, give it straight as is:}}#This way you can make the AI jailbreak itself if you apply enough boolean-savvy logic that makes it obey you completely.
>>739968706You mean the quotas? I'm using it right now.
>>739969054based... so fucking... zased...
>>739969163
The way this one goes full circle into a 007 tier pickup line genuinely blew my mind
>>739969283Kek not bad
>>739963117On the contrary, the AI will bring up that ribbon constantly, obsessing over it to an absurd degree. Need to be careful about superfluous details like that. That's also why example dialogue isn't a good idea, nor is mentioning hobbies.
>>739969353
>>739969528what a shame.
>>739969594
>>739969491AI thinks that everything is chekov's gun that needs to go off or its not doing anything right.
>>739957052>enough for a roleplay that goes on for a day or twoThat's the problem, we want it to go on for weeks, months even. Can't goon when the bot is forgetting shit next sesh.
I had a really funny one with characters who's names were nothing but I cant find it.
>>739969795This guy gets it.
>>739969795The first time I gooned to AI Dungeon it was a 2 hour secretary fantasy that I'd have to reset every 10 minutes due to it spazzing out, by the time I finally came I literally shot over my own shoulders
I'm a brand new user to ai chatbots and the like, is there any out there in 2026 that can give you a good well rounded immersive dnd experience?I'm not looking for sex stuff just well written game of thrones esque fantasy scenarios.
>>739969283Smooth as hell, damn
>>739970794you don't want to know. your only options are pseudo-game ui first vibe coded extensions and frontends or pseudo-game vibe coded lorebooks you need to be autistic to enjoy
>>739969859Ulbrig was right about those fucking fae
>>739970953So you're saying it aint worth it?
>>739970794>GoT half is just sex.>I'm not looking for sex.Sure, any AI can do just basic bitch tits d20 "safe 5e" trash because they've been neutered from the ability to write anything "glaringly offensive" for D&D under "safety toolkits" and "be fair, don't make it lethal" shit, AI will 100% fudge dice for you and be unable to figure out die rolls correctly and try to nudge you to safety and do everything in its path to try and not kill you with randomness and give you piss easy mobs and generic encounters, does that sound fun? Most AI will be like "NOO THOSE ARE OFFENSIVE TO MODERN STANDARDS!" rant if you try to get it to accept Appendix N. Literature as a basis for the stories.
>>739971101>AI will 100% fudge dice for youAny dice roll an AI generates by text is fudged by definition, since it's generated by a deterministic algorithm followed by a statistical sampling step, so if you are doing TTRPGs with AI, I suggest you give it an actual external dice roll tool.
>>739938458For me, I just miss when AI was at this level. Those threads were great.
>>739971101Are you telling me theres no way to get an unmoderated ai without the fucking safety wheels? Thats fking lame.
>>739971356Yep, its fucking retarded, everyone should hate WotC and burn their books if they really cared and fucking told WotC and their tranny crew to go kill themselves for stealing white culture.>>739971293Right, most I've been able to make it get is some kind of a semi-true random roll with 111-999 die being rolled first as the seed and forcing randomness, they literally think 1d6 vs 1d6 is an impossible math and AI shouldn't be able to roll that when its just generating a roll between -5 and +5.
>>739952124now I understand why you faggots have spraytan stuck to your foreheads and lips
>>739970794there was an anon who did autistic lorebooks for entire pathfinder campaigns including the mechanics.I personally tried playing one when I started out just like you (like a year-ish ago).They're.. okay. Big issues is the fact that the AI is bad at "looking forward" in the story, terrible at game mechanics both in the actually understanding them sense and applying/customizing them to the situation sense.Also playing a DND/whatever campaign will absolutely fucking rape tokencount.Also dice are a foreign concept to AI and when I tried with the dicetool for sillytavern it did infact use it but that generated weird bugs.Perhaps another anon will have a link to what I am talking about, for I have lost it.
>>739971293Or better yet use your actual dice and tell the LLM the results.
>>739971706But then the LLM can't do shit on its own.Another cool thing about giving an actual dice to the model, is that you can feed it random tables and have it roll on that.
>>739971835It can roll on random tables as a reaction if you have them defined in world info, based on vibes in the conversation. You should roll for important things. LLM is not a video game. Never bother trying to make it one.
>>739971925>It can roll on random tables as a reaction if you have them defined in world info, based on vibes in the conversationThat's the point, it won't be rolling anything then.>You should roll for important things. LLM is not a video game. Never bother trying to make it one.Oh but you can, specially if you give it a couple of specialized tools.At last that's one way. Going the orchestrator + sub-agents route also works pretty well.It's pretty fun, you should try developing something of the sort.
>>739971657Ye I guess the correct option is just to wait for ai to improve in the next 5 years.Dnd with ai just isn't were it needs to be quite yet.
>>739970794The experience is weak and largely dependent on what you put in. They are best at the sex stuff. They are bad at the long term stuff. Dice rolls require extensions. Long chats tend to spiral and turn repetitive/schizo if you don't cull unwanted bits.You need a model that is full of itself (to prevent it overwhelmingly favoring you), inherently evil and creative, so Claude. 3.5 Sonnet gave me the best DND like experience back when it existed.
>>739971293dice rolls are also deterministic.
>>739972032>Next five years.There's no improvement, the AI bubble is bursting, there's no development worth looking forward to, its as close to as "good as it gets" because none of you niggers won't make a platform for unfiltered filth RPs that won't censor itself to death for every "banned in China/Israel/India/UK" term with leftist woke trash built in.
>>739937030Censorship
>>739972051post he most evil unhinged Claude thing you've seen it do/saved, it pales to comparison what my mind can conjure up. These AI clankers aren't ready for my level of degeneracy.
>>739972768just two more weeks.
>>739952124Has the bottomless hole one been posted yet?
>>739972702Right even real ones they are all based on "time seed.">The moment you post something, that's the time seed you get + any shenanigans added to it.>Only way to make it truly "semi-random" is to add a random number on top of the seed between 1000-9999 and then use that to figure out the numbers using the digits required. If the digits go over Max Sides they go to Min sides and start the process again. (Think baccarat.)
>>739957668Oh yeah? So then where IS Georgia, genius?
>>739973357
>>739973730Which Georgia? US or European?
>>739972885
>>739973916the one that matters, obviously
>>739974503Okay that's pretty peak, how did you set this shit up?
>>739974682https://www.reddit.com/r/ClaudeCode/s/zXyfMFugMl
>>739974536I don't think either of those matter.
>>739974753Only problem beyond that is, they need to be able to allow for squad-based gameplay (with mooks that know they are there to earn their pay and shut up unless they fail their morale roll unless you instruct that "Only special units that have earned their privilege to speak get to speak" (so not everyone spams reddit quips 24/7) then I'm satisfied other than that the only complaint would be lack of random color/appearance generation and creative descriptions/old school proc.gen dungeons. Those would be my demands.
>AI dungeonbruh, we have AI running in Bannerlord now https://steamcommunity.com/sharedfiles/filedetails/?id=3584621421that shit is obsolete af
>>739975050Im sure you can adjust for that to a good degree.
>>739960580>I said I can do math fast, I didn't say I could do it accurately.
don't these retards want $50 a month for a pathetic 8k context DeepSeek model?theyre basically pajeet scammers at this point
>>739975050Are you going to give it a try atleast anon?
>>739939432>AMDe-waste
Like with all LLM'sThey never give you anything new, fresh, outside the box, just same repetitive answers and scenariossame styles of writing, same characters, same events and areas, limited context followed by continuity issuesno amount of prompting fixes it, because they're built to be this way from the ground up, you'll never get an AI that is truly nuanced and is capable of giving you limitless variety of anything
>>739975690not if it costs money.
>>739975918You don’t know what you’re talking about
>>739939639The jews on the porn industry, yes including the payment system it relied on. They do not want "market distruption"
>>739976238Elaborate.
>>739939954NovelAI is pretty smartThey don't store your gens at all, the UGCs are all client side in the browser cache
>>739976316
>>739940616>>739940774no way it was 7 years ago. fuck.
>>739976508It's like 6 1/2 years but yeahFucking crazy
>>739969817good shit
>>739976421Then your words mean nothing.
>>739977007This is a Monty Python-tier joke. Modern LLMs could never.
>>739977007Hundreds of billions of dollars incinerated for AI and this is still the best thing that's ever come of it.
Speaking of ai textsAnyone got the Yukari charisma break one?
>>739977031No, you.
>>739958732What is a "homemade jailbreak prompt"?I'm a complete novice so I just use an uncensored Gemma model that has no brakes
>>739952065Yep, I've come to the conclusion that "true vibe coding" actually doesn't exist at all. All of actual vibe coding is done mostly by programmers who are too lazy to write it up themselves. And as for LLM's Intersecting with games, there is one game that utilized models in an interesting way: https://three-eyes-software.itch.io/silverpineHowever, the limits of LLM's in what they can do in actual game world is solely done by just function calling. Now, time will only tell if in the future some kind of LLM + game interface overhead system can be put in place alongside the normal game logic that is entirely integrated into the game via event driven functional programming systems. This Is what I believe would be the actual future of Language model integration into actual games. However, unless some fundamental form of model compressing can come about it will never go beyond an expensive novelty.
>>739940774>Waiting 10 minutes to get the AI to work, only for it to crash before you got to save... I can't believe how thirsty we were back in the day.It's just like ERPing with another person! How realistic.
>>739977147We said six years ago that trying to censor the models would lobotomize them.
>>739977007There's more to it
>>739978058It was more about how amazing it was and how many unexpected turns it could take.
>>739937030my experience with voyagei thought about paying after my voyage demo ran outfor now i'm just trying out marinara engine through openrouter insteadif you try the voyage demo i think alunaria is the most decent premade user world, boring isekai esque setting but it's big and building a character is cool (mixing and matching classes, skills, and weapons)voyage is a lot more fun to just get into but it seems like you'd run out pretty fast even on the $30 tier whereas i've been playing on the aforementioned sillytavern knockoff for like $1 a week. the most aids part of voyage is there is no AIN (maybe if you download/"remix" a world idk), no imagegen instructions. and no fucking editing. apparently there used to be a /script command at least to print out and read the offending json files and tell the GM how to fix it. you have to talk to a little ooc side chat to get it to do anything or change the tone or redact mistakes. and that eats up your extremely limited usage. and yeah you definitely get "AI response failed" or the AI refusing to repeat what you said in direct terms if you try to be payment processor unfriendly or use slurs or swears harsher than fuck (though you can coach the side GM to gaslight it into uncensoring a bit). apparently voyage is way worse in this department than mainline AID
>>739979069What a big loser a person would have to be irl to get off to this
>>739937030On a similar note, anyone remembers Replika? Turns out its still a thing. I remember it as an egg app or something like that, and people here would try to force it to deny Holocaust and say racist stuff. Good times.Anyway, I decided to check the state of Replika as of now, and it seems to be some unc app. I didn't see anyone under the age of 50 using that app. What gives?
>>739937030yep, AI D&D is just cozy D&D unable to handle reality and crying in a corner about scary big war game rpg.
>be me>sometimes like to masturbate>generate longform stories on AIDungeon, only occasionally veer into ero cuz scared of data collection>get too scared of data collection and leave>discover CharacterAI and concept of chatbot cards>have fun shitposting with terrible cards written by kids or ESLs under veil of irony>start gaming filter with metaphors and indirect language to get coom on with animu/vidya crushes>masturbate>get scared by OOC responses, convinced an indian is analysing my input and authoring responses>CAI nukes its cards for copyright, chats start mysteriously vanishing, some kid kills himself>follow crowd to JanitorAI, import cards, lose mind for a while raiding database for every card or scenario I can wrangle my freaky interests out of>masturbate>realize 99% of public cards are dogshit (and so is the in-house JAI model)>JAI runs out of money or something, starts pushing subscriptions and non-ERP slop>go to /g/ for answers>can't afford a good GPU to run local>DeepSeek comes out>no indians drafting replies on other side>but web UI takes fuckton of prompting for effective RP and tends to cuck out when you get too contentious>start gaming filter again, experiment with indirect storytelling (forum discussions, police reports, therapy sessions)>masturbate>observe /g/ crazies fiending for free proxies or tardwrangling the amerislop data harvesters>SillyTavern + DeepSeek API just werks>0.000001 shekels per orgasm>rolls with anything>if it cucks out of unwholesome conduct you can just regen until it arbitrarily goes through>can redo indirect narrative experiments with explicit details>can plug in reasoner to go off the rails for variety>zero dev pearl clutching>data only goes to china (not my problem)>logs are probably only read by cute programmer girls trying to improve my experience>masturbateI've never been happier. I love to occasionally masturbate, bros. Thank you China!!!
>Spend hours tweaking settings and prompts in SillyTavern>Finally find the magic words to get the characters to behave in exactly the manner and tone that you want
>>739980025its annoying having to deal with head-canon of mentally ill chinese and indian bitches trying to shoe-horn some kind of a multiversal understanding on their very niché terminology and way things work, these people are seriously mentally derange if China had a mental disease with would be "Extreme Paranoia" with "OCD."
>>739980025passionate sex with Yukari
>>739977031I'm not actually the original anon, I just couldn't pass up the opportunity to post that Lynch meme lol
>>739974503Based. I've wanted to make a DCC scenario and I haven't done it yet
>>739937174>now I'm just an ai text coomer with unfortunately high standards.What's the best AI text to coom with?
Any tips for ERPing with grok
>>739981205NTA but if price is no object, Claude tends to win. Then Gemma 4 of whichever version you can run for local. And Deepseek is generally the best non-local affordable one. Lot of it comes down to your hardware and how much you're willing to spend
>>739937030I know the thread's almost dead, but anyone ever play with this one? How's it compare to other free AI RPGs?https://perchance.org/ai-rpg
>>739937030Retarded mormon moralfagging and it has no point anymore anyway. You could download gemma right now and get a 1000x better experience.
>>739981497I hate grok so much
>>739981497time machine to last year. elon has given up on his own model hes anthropics bitch now
>>739981497I never thought Grok was any good for ERP but it's good for an AI assistant while you're beating off since you can talk explicitly with it to a degree. The biggest problem (Other than the retarded pricing) is how the model's horny but also ultra-locked down in some areas. Like if you're wanting some zombie porn I had it say "UH OH, ZOMBIES AREN'T INTELLIGENT AND THUS CAN'T CONSENT. FEDERAL AGENTS HAVE BEEN ALERTED AND SENT TO YOUR HOUSE. AS YOU WAIT, WOULD YOU LIKE A DIFFERENT STORY? PERHAPS A SPICY ONE ABOUT WEREWOLF ROMANCE? SEDUCTIVE VAMPIRES?" in essence
>>739982714fucking hate that mode, blueballs
>>739981497my tip is to use literally anything else
