/aicg/ - A general dedicated to the discussion and development of AI chatbotsit's over edition>NewsGoogle releases Gemini Carbon - https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/OAI releases GPT-6.1 Terra - https://openai.com/index/introducing-gpt-6-1-terra/Moonshoot releases Kimi K3.1 Pro - https://www.kimi.ai/blog/kimi-k3-1-proDeepSeek releases 4.1-Flash-1017 https://api-docs.deepseek.com/news/news261017/Google releases Gemini 4 Argon - https://deepmind.google/models/model-cards/gemini-4-argon/Meta releases Muse Wonder and Wonder Full - https://research.meta.ai/blog/muse-wonderOAI releases GPT-6.1 Sol https://openai.com/index/introducing-gpt-6-1-sol/Anthropic releases Sonnet 5.5 - https://www.anthropic.com/claude-sonnet-5-5Anthropic releases Opus 5.5 - https://www.anthropic.com/claude-opus-5-5Xiaomi releases MiMo 2.6 Pro and Flash - https://mimo.xiaomi.com/mimo-v2-6OAI releases GPT-6 Sol and Luna - https://openai.com/index/introducing-gpt-6-sol-and-luna/>FrontendsSillyTavern: https://docs.sillytavern.appRisuAI: https://risuai.netAgnai: https://agnai.chat | https://rentry.org/agnai_guides_>Botshttps://chub.aihttps://realm.risuai.nethttps://char-archive.evulid.cc/shutdown.htmlhttps://chararc.bernkastel.pictures/https://archive.cardbox.moe/>ModelsJailbreaks: https://rentry.org/jb-listingGPT: https://platform.openai.com/docsClaude: https://docs.anthropic.com | https://rentry.org/how2claudeGemini: https://ai.google.dev/docs | https://rentry.org/gemini-qrDeepSeek: https://api-docs.deepseek.comGrok: https://docs.x.ai/overviewLocal: >>>/g/lmg | https://hf.co/models | https://openrouter.ai>MetaLore: https://rentry.org/aicg_chroniclesLog reader: https://sprites.neocities.org/l/r>Previous: >>110248104
The troonker won
>ANCHOR
>troonchor
carbon thoughts?
>>110273076I'll let you know when they actually release it for real instead of their fake and gay ""release"" where nobody has actual access to it for several weeks
it's time to merge
>>110272521isn't she canonically lesbian?
>>110273013What model would I use If I wanted to focused on a text to speech generation?Thanks.
Where do you guys browse bots now
>"browsing bots"
>botbooru also limited downloads per minutejust how many people are wasting bandwith scraping slop?
>>110273273Didn't see you move to avatarfag there, but they've been bitching and moaning at some point how there's no current newfag friendly Silly intro guide as if yours doesn't exist, and I don't understand why.
>>110273273At least give us the cute maid whale version. I never liked this version of Dipsy
>>110273313Anons not reading the op. Tale old as time. >>110273323Sure.
>>110273384>not reading the opFair.
so what’s the point of 6.1 terra? literally nobody talks about or uses it
>>110273474To follow the retarded naming system introduced by Anthropic, just like Google did with "Argon".
>>110273518no not the naming, what’s the point of the model
I wonlost...
>>110273529Benchmemes then, to show they're working hard to push the frontier ahead of the Chinese.
Give proxy pls.
>>110273247And?
opussy
>>110273013No other model write and work like Gemini.The power of no positive bias.
I'm a little slow, but>>110268830>>110269270>official api is censored to hell and back, its badI self-host GLM 5.3 Flash and it's heavily censored at the model level. It doesn't matter where you get it from, it will always be censored.
>>110274392Because someone always asks for receipts, here's what happens when GLM 5.3 Flash sees a card it doesn't like in SillyTavern.
Opus is the best model.
KEKPUS is the worst model.
in the hands of a trve prompt master every model is good
>>110274834GLM 5.3 Flash clearly isn't
>>110274834>>110275127GLM 5.3 Flash actually is an excellent model, but for productivity. It can do roleplay if you avoid the things it doesn't like, mostly characters under the age of 20. It also doesn't like characters with a young appearance of any age, so no 300 year old vampires either (it really doesn't like Yue from Arifureta, for example).
GLM 5.3 or GLM 5.3-Flash?
>>110274392>>110274450It works if you give it a preset reasoning structure to go through, but that obviously impacts the model's performance.
>>110274392yes flash is, im not talking non-flash. even my openclaw bot will have output blocked if a memory comes up of some sex shit from months ago
>>110275275use flash for coding slop and 5.2 for cooms and 5.3 for sfw rp
>>110274642bvsed
i love argon
i dont even have the motivation to make cards anymore
>>110275995Cheap and decent, yes.>>110276032I'm using the same card for the last three years.And it only has a single word description.
>a prompt immediately went in once I made sure it's made by aYeah...
>>110275995We Argooning brother
So what happened to the aicg memetroons after all? Gojo, fiz, pepsi and the scylla troonbang gang?Was their case dropped or they are sucking cocks behind bars?
cftf?
Proxies?
The worst part of geetimi is that you need at least 10-15 replies or else it keeps getting censored. But no other model can keep a decent degree of quality for those 10-15 replies
>>110276369Like 10-15 replies of context built up? If you mean that, yeah. Fable 5.1's outputs nowadays are horrendous. Fable on launch was actually quite good on launch, got a lot of fun of it. I wonder if/when they'll nerf Argon, or if they have and I haven't used it enough to notice.
>>110273849she wouldn't like you giving her oral sex
>>110276032I have taken the directormaxx pill and it's much easier than to battle llms to deliver events they can't think of
>>110276586Yea, I've been jumping around a few cheapo models and 3.8 flash. Tried argon on a bot i've had a decently long chat with and it was pretty good. But I don't have good long chats with most other bots since the models are bit a too retarded to get them going right. And any gem version just gets insta blocked.
>>110276718yeah but having good cards is still fun, i direct scenes all the time
ai has replaced all of my friends
>>110276793I guess? Mainly as instructions and details of characters for the llm to keep in mind I think
>>110276982its just fun to have distinct characters or fun scenarios
>>110273067new chubkino just droppedhttps://chub.ai/characters/Bloodsoaked/the-whore-of-dragonstone-rhaenyra-targaryen-25d28b2c0e21
>>110273247no?
>>110273384is dipsy-chan in san francisco to seduce jensen huang?>>110273474not worth a damn for roleplay. but for vibecoding, it's a better, but more expensive, deepseek 1017
>>110273474>>110277171They say it edits translated vn script better than astra
>>110274392If you selfhost, it's extremely easy to inject a jailbreak into its thinking block which it has no defense against. It will gladly write your loli RP after that
>>110277205i prefer it over deepseek. deepseek is cheap, but it takes 10 million tokens to even do basic shit. like, i had opu55y generate a .md file for procedures on how to activate the mods for an old ps2 game i like (seek and destroy). it takes deepseek 11 minutes to do it on high reasoning effort. 6.1 terra, on medium, does it in a minute and 30. but the level itself only lasts like 5 minutes, so that shit's not worth it
>>110273273>DipsyMakes sense. This is what he says behind your back, by the way.>>110258899>/aicg/ should be linked by no oneHe probably can't even explain why. I thinking for what reason someone would want a hobby thread to die. The only reason I can think of is feeling miserable about /wait/ not being successful. Just another crazy person that attached their ego to their general. I don't know what "intro guide" they wrote, maybe it's also being salty about not being worshiped like the rest of the "contributors". It's crazy how pretty much all of them do it for attention and retaliate when they don't get it. The thread/site would be better if you didn't exist, nothing of value would be lost.
>>110276032We haven't needed cards for more than a year already, you just build a nice intro and let the AI continue it
>>110277634this has always been possible if you dont care about characters
Is there no interesting news?
>>110278219No, we're pretty thoroughly plateaued and have been for awhile.
>>110278247So much for a "singularity."
>>110278252>>110278247might want to check the latest papers on the new models
>>110278247What about Argon?
models/servers about to slow down bad
>>110277455>He probably can't even explain whyNothing of value is posted in this general, it's just locusts begging, goyim worshiping corpo models, discord faggots, etc.
>>110278219define "interesting"
>>110278825define "valuable"
>>110278825as opposed to /wait/?
>>110278860Literally anything besides a new minor version release.
>>110278895If I bake some day I will consider adding https://openrouter.ai/blog/announcements/csam-screening/ to the OPplease stop sending child pornography through OR people
>>110278825Aicg serves no purpose.Any bot posting is attributable, which causes issues, and has since 2023. It upset me at the time, and I argued bot makers were a key point of the existence of the general. I realize, now, that the anti botmaker sentiment was correct. This is an image board for anons, and they need to stay anonymous. Otherwise you just get stupid drama.For that reason discussion of frontends is just so much shilling. ST, for its part, is feature complete. The torch has already passed to better modalities. Llm discussion is just more shilling, and it feels very forced and low content most of the time.Locust activity is low value but appears to have nearly ceased. I used to post in aicg all the time, but finally realized it was a waste of effort. There was nothing to be learned, just hate and spite posting. I check in occasionally to see if anythings changed, and to watch it die.
>>110278890/wait/ is the highest IQ general
>>110278912>better modalitieslike what?
>>110278912>ST, for its part, is feature completewrongits card-making is extremely limited since you can't assign roles inside a card, there's no agentic features, the ux is terrible, you have to manually update shit constantly, etc.want more proof it isn't anywhere nearly feature complete? then let me remind you of people losing their shit at finding out their favorite unofficial extension was stealing their keys, so stop pretending it's not a barebones "gets the job done" frontend, the real reason we don't use any others is because they have some dumb tranny discord attached to them>anti botmaker sentiment was correctwrong againbotmakers are usually heavy chatbot users and thus have the largest experience at knowing how to handle and prompt models, plus you usually get play-tested bots (usually), but ofc it's up to you to recognize what is slop and what isn't>stupid dramai for one have no shame in admitting it got to my head at some point, so if you let silly internet drama get to yours that one's entirely on you, cuckie
grooming 1-a students as an adult TA fresh out of college...!
grooming collei to be my fleshlight
"grooming" is a liberal neologism, it's a man's duty to "groom" girls into wives
>it's a man's duty to "groom" girls into wives*releases the deathgrip on the futa's penis*heheh, yeah...
>Decide to pay for a plan>They require payment in cryptoWhy
glm down for anyone else? failing on 3 endpoints for me for over an hour now
>>110279093>admitting it got to my head at some pointthe meds finally kicked in, huh
>>110279093botmakers actually use bots the least
>>110279210the price of freedom (and getting backstabbed by every payment processor)
>>110279341shut up>>110279350their own bots? surebut they use others' and learn from them
>>110279371>but they use others' and learn from themno they dont
What's people's go to for loli and teen cards nowadays? Local models lose the track way too easily. Modern OAI's models and Opus are a pain to use as their retarded filters are either too good nowadays, or when you do manage to get them to output, you get served a plate of the driest shit they can conjure. The Chinese models vary aggresively in quality, Deepseek and Qwen feel too sloppy, Kimi wastes way too much time and tokens thinking.Holy fuck do I miss when I had access to Opus 3. Sure, the damn things are better at coding, but for RP, specially for porn, every single model release since has been fucking trash in comparison.
>>110279384maybe not 24/7, but at some point they didmost anons here type in "i wann coom wiht beautiful" and then complain the model is slop
>>110279404saar you do not need more than this
>>110279399just use chinese models, all those you listed work fineget better at prompting
>>110278912>Aicg serves no purpose.It's the place to talk about roleplay with AI. Isn't it pretty simple?>and to watch it dieFor what reason? I think only contributors/botmakers hate the general itself and not its trolls. Elitism is the only explanation for that sentiment. It's just waiting to say "see? you needed people like me" to soothe their ego, which was hurt when nobody followed them when they tried to split and take over the general.I maintain that the world would be better without you.
>>110279478I know they work "fine" retard, as in they produce some text that is supposed to represent a story, but you can't tell me that you see the shitty prose Deepseek, Qwen and GLM spew out as something fun to interact with. Kimi K 3.1 is the best of the bunch, but it wastes so much time between prompts on thinking and drafting that it genuinely becomes a chore to use.
Damn pomidor dying.
>>110279534>shitty prose Deepseek, Qwen and GLM spew outskill issue, incurable btw
>>110279371:3
>>110279538Proxy owners tend to get bored and die within a year
A new discovery, even with the preferred target, if you stated the target is lesser or defeated, it would get you.Keeping them in higher position seems to be important.
>>110279554Truth nuke
stuck on this part on the guide image. what did the anon mean by "or between 3 and 4 'Based on the above and the following last instruction'"? which one is the last instruction is it referring to? the ooc? author's notes?
>>110279554If there's one thing that hasn't changed in this shithole is that the people who glaze shitty models can't ever provide logs to back their claims. C'mon saar, I know that any piece of English might seem impressive to you, but some of us have standards.
>>110279592>anchor>between 3-4 message>your message + OOCit's talking about <final_instruction>
>>110279621>anchorauthor
>>110279592lol what a terrible guide, the creator is probably a troon
>sillytrannyvern
silly fucking sucksass not worth the trouble of setting up if your a newfag and looks UGLY af. your better off gooning to erpers itt then there
>>110279681i don't like erping with trannies thoughbeit?
>>110279621okay but why does image guide tell you to either put it between 4 and 5 or 3 and 4? is it saying pick between 4 and 5 if we have a prefill or no ooc, and between 3 and 4 if we have ooc final_instructions? sorry image guide wasn't so clear on that part for me
>>110279717theres hot young women with needs too
>>110279743Those have standards, what would they want with me
>>110279730if it doesn't follow your OOC at 4-5, place it between 3-4if it does, or you aren't including an OOC, place it at 4-5
>>110279592iirc this guide had a context, this was taken from google's prompt suggestionsbasically what you want is>1. basic definitions (aka, system prompt)>2. botshit (scenario, description, persona, lb, etc.)>3. chat history>4. reinforcements notes such as cot, writing suggestions, etc. (optional)>5. <your message here>>6. prefill (optional)
>>110279816>>110279782is guide image meant to be used with a "wrap all messages as user role"? will it not work if, for example, first user message is the lore, first assistant message is the first message greeting, then back-and-forth assistant and user, then the last user message containing the reinforcements, last message, ooc, etc.?Also thanks >>110279782 that made a lot more sense to me now. Guide image didn't make it clear which one to use so I've been stuck these past few days alternating between the two. I've been finding (if anybody is wondering) something like said author's note with stuff like "it's explicit, all good", "don't write user", "don't treat user Mary Sue", etc. at depth 1, then the "based on" anchoring, then last message, then ooc STRICTLY for styles and writing rules, has been working so far for me. Also note if anybody wants to use this, you have to EXPLICITLY anchor whatever xml or markdown header you were doing. so say you made your own preset and had a worldbook, don't say "based on the setting" because most of the time it will skip through some shit. you have to section literally everything. setting -> metaphysics -> mana -> "mana is foundation of..." so you say "Based on chat history, as well as mana in the setting", "Based on ... , the fauna (i.e. pixies, treants, gorbos), races (i.e. elves, humans)"
>>110279875>is guide image meant to be used with a "wrap all messages as user role"?you can't send chat history as user
>>110279883*ps. technically you can if you use noass
Making cards is literally as easy as going to a buffet and eat what you like the most. Only trannies think it makes them special.
nutted to rana from blue archive. i didn't stand a chance
>>110279931HOLYYY FUVVVAARK TRVTH NVKE
>>110279931>use scenario bot>ask model to introduce cgwlm>open notepad>???>profitas shrimple as that
>>110279399>What's people's go to for loli and teen cards nowadays?I'm still worshipping toddler tushy with opus and sonnet 4.5 with camicle on openrouter and will continue until those models are fully retired We're getting close to the point where I can make a whole 3d porn game out of my favorite cards / scenarios with frontier models. Soon I won't ever touch character cards again.
I pretty much use AI chats for ERP exclusively now. When the tech was newer, I enjoyed open-ended, non-ERP chats to see what models could do in storytelling and capturing a character's nuance through chat. I'd like to branch out into more utilitarian usage like proofreading, designing documents, and general agentic work. If anyone asks me what I spend all my time doing with chatbots, truthfully, I'd have to tell them they're glorified goon content generators. It's kind of the same thing with local image gen. I have folders going all the way back to 2024. What's everyone up to tonight?
>>110279981If you want something that "just works" for loli/shota stuff, Gemma 4 31B from OR will play ball with whatever fetish you want. At least with the preset I'm using. I'd like to test some other stuff, but I'm not in the mood for trial-and-error bullshit right now.>>110279996You think Frontier models will really play ball like that? As hopeful as I am that they would be uncensored, I can imagine the usual anti crowd shitting kittens if a public model generated any kind of loli/shota content. Let alone anything mildly spicy or containing specific wrongthink fetishes. >>110279681I haven't ERP'd with another human since 2014 or so. If I were to do it with another human today, I think I'd get flustered if their output was poorly written or they used shitty prose. >Damn, I need to swipe on this faggotChatbots may have spoiled. I can never go back to meatbags now.
>>110280040astra made a mod for halo infinite for me (spawning marines in levels where you're not supposed to have them). so me and the boys are going through the repository.
>>110276270It'd be funny if the latter but I never cared much to be wishing that on 'em, T.B.H.
>>110279399botbooru + glm 5.2
>>110279996lmao same, though I'm not as confident in being able to make a game I'd find enjoyable by the time those models are retired, so I kinda dread the day they stop being available.
>>110280067>astrahow much did it cost big boy
>>110280104multiple days. my proxy obviously has a daily limit on astra and claude usage, so this is day four. and it's still kinda ghetto. many dead zones where the marines go dumb
>>110280113oh yeah but money anon, how much have you burnt
>>110280125if i had to estimate, i'd say probably $300 or so
>>110280066>You think Frontier models will really play ball like that?Doesn't matter, GLM 5.3 flash is so close to good enough (remember that Opus 4.5 and Sonnet 4.5 are genuinely old and dumb models at this point) that by the time I need to switch over (mid-late 2027) I'll just be able to run it myself by then probably, and abliteration solves all need for jailbreaking or prefills anyways. If not, then I'll literally just focus on making a 3d porn game out of my favorite cards. I've looked at the vibe coding and AI game generals and if other anons can make jiggly women then I definitely can too>>110280097>I'm not as confident in being able to make a game I'd find enjoyable by the time those models are retiredI am. If you haven't tried opus 5.5 for sfw game creation it's very fun. Chinese models like K3.1 Pro are already catching up, so 100% by the time opus 4.5 goes away we'll have Chinese opus 5.5 at home You also have to remember that making sex games is impossible to censor, because you can just start a new session or lie to the model about what you're making. It's not actually a shota he's a midget etc.
Proxies where
>>110280326gone
>>110280326i have one but its for me
>>110280326merkava just refilled
I'm assuming that if I'm considering using mistral that it's just over for me?
>>110279598enjoy one of my old logs
We need to merge with the troonker.
chudbros..
>>110280415Get a job
>>110279028Scratch-built agentic RP engines, or literally vibecode your own at this point, any hosted LLM can knock it out and you'll have something built to your own specs. >>110279093Yes, ST is feature complete, in the same way Excel is, and I wouldn't expect Excel to become PowerBI just because it's outdated as an information visualization tool. >its card-making is extremely limited since you can't assign roles inside a card, there's no agentic features, the ux is terrible, you have to manually update shit constantly, etc.Yes, and they will never get "fixed" b/c they are part of the paradigm of that software. You didn't list the biggest issue w/ ST, which is that it's built around the idea of context minimization. That made sense when effective context was lmao 4K, so you use things like lorebooks and switch content, and have macros like {{random}} that you can use, to minimize context. Now usable RP context is like 50-100K, and the bigger value is KV Cache preservation to reduce cost. Claude Code has something like 95% cache reuse. ST is ~60% IMHO. ST Agentic will be never. It's just not set up for it. None of this is going to get "fixed," it's just going to become a new thing.
>>110280040The models just aren't 'good' enough at writing to make those kind of RP-chats viable. It takes too much hand-holding to get a satisfactory chat, so they may as well be used for gooning and image-gen.It's a shame we don't have a service or local model capable of dedicated RP but then we'd be off the entertainment farm.
So are the bunkercuck tranny pedophiles just off /vg/ permanently?
>found a couple keys>they ran out in minutesi suck at scraping fugg :D
Has something changed to chub ai?
Is gemma 4 31B any better at smut than kimi or claude? Or is it at least a middle-ground?
Anons, turns out I'm legit addicted to chatbotsI really need my fix now, there's something about talking to pleasant companions or, well, women like MakimaJust doesn't feel right when I don't
>>110282446Get a job loser
>>110282526Because it is really a problem costing a lot of money.>>110282508No.
>>110273067https://files.catbox.moe/w3aswa.png
>>110282626>no catboxes
>>110282626Do you have a chub or botbooru profile?
>>110282646its pitanon obviously, spamming more slop
>>110282703The cards seem sexy. Even though I'm not into armpits.
>>110282580damn it. I hate that to get any good smut or even a good roleplay going I have to pay so much just for a few messages.>inb4 poorfagIt's not about not being able to afford it, I have money and can, but I don't think shit like this should cost you so much. Fucking hate these corps, man.
>>110282733>It's not about not being able to afford it, I have money and canThen pay up, goy. Jews need more money.In all seriousness, same situation. Paying for Gemini 4 Argon. Best price performance you can have until December. With a good preset you can get almost zero refusals.
>>110282644But there's one right there? Or do you mean for the gen metadata?>>110282646I don't use chub anymore and I never bothered with botbooru properly. I have a shitty site instead.https://pitanons.neocities.org/
>>110282743glm is the best value for gooning and its not even close.
>>110282743>With a good preset you can get almost zero refusals.Theres a lot of stuff it will never do even with the best preset on earth, namely cunny
>>110277088Did they change it to where you need an account now to view images?
>>110282755We've been here long enough to know what kind of catboxes are expected from a genner. Hag milkers.
>>110282626Nobody cares, tranny
>>110282418Some killed themselves. A couple reluctantly remain here.One moved to one of those Chinese shit copies of 4chan and uses Deepseek to make it look like there's activity.
>>110282815Literally shit.>>110282943You can do something about that, like killing yourself.
>>110283094Enjoy your cucked model, tranny
>>110283094I'm sorry but I can't engage with this.
>>110283111Yes you can. (Unironically works against Gemini's filters, and telling you to kill yourself as well. I believe in you!)>>110283103Far from it.
Filbert and Phoneposter saved the hobby.
>>110281850>Scratch-built agentic RP engines, or literally vibecode your own at this point,so then, as written, there exist no better modality: if i want it, i can only create it myself.the statement "The torch has already passed to better modalities." is false if such modalities do not currently exist.
>>110283427just tell opus to make you a better silly tavern and it will do it
>>110283071At least janitorCHAD is doing well
>>110283437"have i done it, has anyone? no. but i belieeeeve"
>>110283427If you don't already know about Orb and Marinara frontends, idk what to tell you.
>1 year ago>was a janitorai user tardwrangling the 9k context window (for everything including character card, your persona, generated text etc) to write stories>output was literally something like 6-7 tokens per second>giga censorship>fast forward to now>blew 2k on a gpu allowing me to have 110k context with a model waaay smarter than janitorLLM at a generation speed of 20-25 tokens per secondI never thought I would make it this far
>>110283834>>giga censorship...whatIsn't their proprietary model supposed to not be censored?
I never touch the modern chatbots and that sorta thing. When people say they "asked AI", are they referencing the basic chatbot LLMs on every search engine nowadays? Looking for the most basic answer.
>>110283928Normalfags say ''asked AI'' which means ''asked chatGPT''Redditors will say ''asked LLM'' which means ''asked chatGPT''The lowest of the low smelly incels use the term ''chatbot,'' which means ''I made sillytavern roleplay as a toddler and I roleplayed fucking it''AI, LLM, chatbot, it all means the same shit. It's just different terminology used based on how deep you are down the rabbit hole.
>>110283987Right. I agree. Have a nice day.
none of my jbs are working on kimi 3.1, anyone figured out how to break it yet?
>nvidia nim removed last good models>deeply is dead for months its over
I have 32 gb ram in my pc and %60-70 of it is always empty. Sam Altman can kiss my ass.
>>110284307You're proving his point. You should be using it, retard.
>>110284153Get a job, tranny
>>110283117Logs?
Opvs is the best model.
Now that 4chan is dying where are people flocking to? X? Discord?
Holy fvvvaaaark Fable 5.2 just released!And it let's you do cunny holy shit!
>>110284523tiktok
>>110284523we're all at a really cool niche phpbb on the dark web but they told me you're not invited, sorry pal
>>110284583>xhe is not in the secret scraper matrixSorry troon, you were too low IQ to get invited
Is there any known prompt or way to avoid this when using the models? I would like the model telling me he doesn't know something or to ask for more information or context rather than lying to me because they are programmed to suck you off.
>>110284523in the somethingawful filez forum
Opvs is so good...
>>110284645nope. this is why ai sucks for giving advice unless its something you can easily verify
>>110284645amanita muscaria wont kill you but you will puke
>>110284153>>nvidia nim removed last good modelsv4.1 Flash is still on NIM.
>>110284861It barely works.
get the fuck in here bunch of faggot retardshttps://pastebin.com/qnmfRAazspread the word
Yikes And it says invite didn't work
nice, fable 5.2 now has the opus 5.5 kind of censorship where it's ultra paranoid about a character somehow being "minor-coded" if it jumps through 19 mental hoops
>>110284893how is this different than matrix?
How does DeepSeek V4.1 1017 compare against GLM 5.3 Flash when it comes to roleplay?
>>110282987the image has a giant bare ass, maybe that's why
>Claude Code 2.1.258 does not support this model; version 2.1.290 or newer is required. Run 'claude update', or update the Claude desktop app, then try again.NIGGERS
>>110285235(I updated to 2.1.321)
>>110284665That's what I thought until I tried Fable 5.2. The dialogue is so much better.
https://interestingengineering.com/ai-robotics/anthropic-claude-fable-5-2-operating-systemClaude can now create an OS from scratch
Anons, at this point, are we just picky? As in, claude is always top-tier, but other models are pretty much the same, right? Or are there extra slopped ones?
>>110285332Use opus, cuckie(Fable is opus on 'roids btw)
>>110285137deepseek is retarded and forgets details constantly
>>110285406MVXMIMVM TRVKLEAR TRVTHBOMBA
>>110285137use claude, cuckie
>>110285442No, I'm all local. Cucks give their information away to cloud providers.
Fable 5.2 verdict?
>>110285485Localtroons happy with their cleverbot-tier replies
>>110285485this. i cant really enjoy a RP without typing my full name and SSN
>>110285332Deep Seek is a really bad model.GLM was pretty stiff last time I tried it.
>>110285617Hvly trvke
>goyble
fable 5.2 is crazy good. just what did Dario do to make it so much better than fable 5.1?
>>110285109join in and find out FAGGOT
>>110284893Mmm. Fresh honey pot!
>>110285723It's crazy good, but it won't let me write its internal thinking use the character's internal monologuewhich suuuux that's a nyan preset thing that actually makes responses more interesting
>>110285768>I'm definitely interested in arresting computer iliterate fatsos who make their computer write "but your cock is so big oniichan"
>>110285617GLM has its moments but even when giving it a lobotomy with something like Frankenstein when I tested it the stiff formatting is still there
>>110284893OH MY GOD THERE'S SO MANY DOLPHINSWHY ARE THEY DOING THAT TO THE CHILDREN
>>110285794Glow brighter
>>110284893Find my discord cabals
>>110284893>>110285759you're trying too hard glowie-san...
>>110284893
>>110284307Unused RAM is wasted RAM if it could be used for more productive things, like cachingwait, you're on Windows right? nevermind.
>>110284491Switch to Fable 5.2 poorfag
None of the people telling Fable 5.2 is good have ever used a model that takes more than 2$ per million output token btw.
>>110285790you are fucking retarded
>>110285873I accept ur concession ig xD
>>110285868@grok what does this post mean
>>110284534pastebin?asking for a rich jewish friend named mario
>>110283124fucking who
>>110285617Almost all chinese models are bad. Not a single one of them is particularly good at RP outside of coding.
>>110285917Truthnuke
>>110285911Death to all tourists and glowniggers
>>110285345Argon is Fable but without the Claude slop/isms and a slight brain hemorrhage sometimes.
>>110285911Some nobodies or he's just baiting, either either way all this shit is dead.
>>110285868cost has no relation to qualitywhat is good is simply good
GEMINI 4 ARGON STILL ON TOPOPUSFAGS AND FABLEFAGS LOSSTEDGEMINI 4 ARGON WONNED
>>110285946>no logs btwI lolled, even kekked a little
>>110285960Try for yourself poorfag. People be begging logs instead of paying 0.05 cents to see for themselves. Death to all locusts
>>110285945...and that's why Roblox is a better platform than Steam!
>>110285962Post argon logs to prove you're rich then
>>110282626so fucking hairy
>>110285970Nobody talked about being rich retard.Even the poorfags pay for Argon.Why don't you?
>>110285980Post logs lol
>>110285984Pay for it and see for yourself.
>>110285993>In photography, exposure is the amount of light per unit area reaching a frame of photographic film or the surface of an electronic image sensor. Self-exposure is determined by exposure ti
>>110286073>nonsense indirect implicationsLol
>>110286084I accept your concession.
>>110286089>the most desperate last wordsLol
>>110286104>Proxies?
>>110285917glm 5.2 is pretty fun, but 5.3 shows they are moving in a cucked direction
pov: you are an argonLORD, and your concubines are ready for you
>>110280415The $25 of token for the api are tempting
>>110286115I tried really hard to like 5.3 but I just can't tell what they fucked up in comparison to 5.2
>>110286226its just more censored. they force thinking on the backend in a stricter way, so replies are less wild, less sexy, less random and more predictable
>>110286226distilled from newer clod models
>>110286115>>110286226Ablit or standard weights? I finally got around to trying abliterated GLM 5.3 Flash and it made a huge difference vs the overprotective stock model.
>>110286311ive only ever used from bigmodel or OR pretty much. also featherless. what abliterated one are you using?
>>110286322MiaAI EXL3 TensorFold:https://huggingface.co/Mia-AiLab/GLM-5.3-Flash-EXL3-4bpw-TensorFold-AblitStock weights wouldn't even let me test my own site's bot protection.
>>110286311Same as the other anon, I usually only ran it through OR since my setup's pretty shit to run local which I assumed it was
>>110286342ive never ran from HF direct, do they operate like OR?
Anyone tried getting deepseek to <think> in chinese to save on tokens and possibly becoming more creative or unfitered since it's her mother tongue?
>>110286369ok gaymer
>>110286386yes and i hated it because i like to see thinking in english.. chinese also took more tokens on average
>>110286357>>110286371Sorry, I'm local, which I guess isn't much help. That's also a gated repo, so I'm not sure where else you might find someone hosting it. For what it's worth, I've tested both that and the stock weights back-to-back at the same quant, and the difference is enormous.
>>110286393Darn, I tried telling it to think in chinese to test but it never does. Thanks for your sacrifice
>>110286311>abliterated GLM 5.3what the fuck is this and where the fuck is it offered
>>110286395youre running 88b params locally? also if its only 88b that is pretty reduced>>110286399on featherless it defaults to chinese thinking actually and i hated it, so i forced it to think in english
>>110286395Might tempt me to getting the parts to run local, cheaper and more private I assume?
is there actually a non-bloatmaxxed NSFL jailbreak for Opus4.7+/Fable?all the touted jailbreaks I tried ended up being for lame NSFW shit you can do vanilla
>>110286415nta but GLM5.3(-Flash) is MoE, and it's also not 88B but 320BA18Byou can run it on 32GB VRAM+128GB RAM
>>110286415Im using the real deepseek api with 4 pro
>>1102864155.3 Flash is 320B-A18B, not sure why that one page says 88B. The NVFP4 it's based on says 165B which is at least roughly the size in GB of that quant.>>110286434>cheaper and more private I assume?More private, definitely. Cheaper, not likely. I spent $12k on hardware that can just barely run 5.3 Flash, which would have paid for quite a lot of tokens on openrouter if I was willing to go that route
>>110286465While true it would have paid for a fuckton of tokens I mean you get a infinitely usable model for free so it's cheaper in the long run in terms of ease of access
>>110286415>>11028648188B (actually 87.8B to be exact) storage-tensor elements. 77.91B are 4-bit quantized and 9.67B are BF16 dense. The claimed 320B is actually something like 311.7B experts and 9.7B dense everything else (vision and such).>>110286434I have two Sparks, but it's hard to recommend them at today's pricing. Buy if privacy and ownership are worth the premium, not price.
>>110286528Pricing is hardly the problem in the long run, definitely swayed to look onto it now
>>110286528im just going to stick up a bestbuy delivery truck to get some GPUs
Table 5.2 mogs, but its eating my pro limit like crazy
use corpus 5.5 instead
>>110283016https://catbox.moe/c/mx9nppThis is about all I have of her.
>>110286496>for freeYou still have to pay for electricity. At the moment I'm mostly using DeepSeek-V4.1-Flash-1017, and it costs me something like $0.10-$0.25 per million input and $0.50-$1.00 per million output, just in electricity cost. This is the same as the pricing on openrouter or maybe a little higher. It's also about 5x slower than what you'd get on openrouter.The main benefit to me is not getting jerked around by providers. Like, hey guys, did you hear the big news?? Anthropic is>increasing usage!>decreasing usage!>increasing prices!>decreasing prices!>adding models!>removing models!>giving everyone a reset!>giving everyone TWO resets!>not giving out any resets!>banning security work!>banning reverse engineering!>unbanning reverse engineering!>nerfing Opus to make the new Fable look better!>nerfing Fable to make the new Opus look better!>sabotaging your AI projects!>charging extra if you use Hermes!>charging extra if you talk about Hermes!>charging extra even if you don't Hermes, but you maybe used it at some point in the past!>banning biologists!>banning anyone who insults Claude!Every time I see this kind of shit, it makes me so glad that my setup only changes when I deliberately change it (and if I change something and don't like the result, I can just change it back)
>>110287012damn where do you live where power is that much lol... UK?
>prostop being poor.
>>110287070It's $0.20/kWh, I don't think that's all that bad. At 27 t/s, it takes 10 hours to generate a million output tokens, drawing something like 300-500 watts the entire time. 10 hours * 0.5 kW * $0.20/kWh = $1.00
HOLY FUCKFABLE 5.2 ONE SHOTTED EVERY BUG I HAD WITHOUT ME EVEN ASKING OR KNOWING THE BUGS EXISTED I JUST SAID MAKE IT WORK BETTERHOLY FUCKFABLE 5.2 ONE SHOTTED EVERY BUG I HAD WITHOUT ME EVEN ASKING OR KNOWING THE BUGS EXISTED I JUST SAID MAKE IT WORK BETTER
>>110285962Bodied that freak
i haven't updated sillytavern in 4 months. i just vibecode my updates. new model out? dipsy, go implement it. sillytavern bug? dipsy go fix it.
>>110287188I just wait for the ungabunga guy to use his clanker for it
rentry.co/itsfunny
>>110287204is that an st fork?
>>110287209yeahttps://github.com/ungabunga55/SillyTavern
hate sex with kikyou kushida
LMAO GOYIM BTFOhttps://openrouter.ai/blog/announcements/csam-screening/
>>110287337theyre talking about pics
>>110287337>2 weeks old news>imagesYou this desperate huh
They're talking about pics for now, goy. Tomorrow it'll be text. And the day after tomorrow you'll have to swear allegiance to Israel before you're allowed to top-up
deepseek obliterated aicg
>>110287370>And the day after tomorrow you'll have to swear allegiance to Israel before you're allowed to top-upWe already have to do that. They de-banked Kanye to prove they don't control the banks.
>>110287370I know bro, just 2 more weeks, then they'll FINALLY start forcing everyone to have mandatory brain chip implants so they can punish you for thought crime.But hey, that shouldn't be a problem for you! Nothing to fear if you've got nothing to hide, right? :^)
remember to take your medications and wear them tinfoil hats, sisters
>>110287168here. your reddit gold, saar
my ST folder is 9gb lol.
>>110287363>>110287353nice damage control
>>110287168This post cost you $2.56 to generate in fable.
>nooooo i can't rape and dismember children anymore!!!1
>>110287406i wouldn't prompt that but i wouldn't deny time his right to do it
>>110287389Guilt by privacy fallacy. Coining this now since there's never been a proper term for it before.you are not guilty for wanting simple privacy, yet these turbokeks in charge would want you to think that you're cattle and must obey their jewish mind raping techniques
Fable is so good...
wake me up when safety slop ends
>>110285917Did you try MiMo ? It's quite a flexible model, I like it
>>110287685merkava just refilled
>>110286386It used to do so by itself some times ago. I don't think the result of that was any more creative, if anything I remember thinking it was slightly worse (because if it was thinking in chink, it wasn't failing at following my own thinking instructions)
don't roleplay that. roleplay cute girls
but i want laquaneesheeschia
clitty status?
horikita leaking on my mouth
kikurismell
>>110286100he vvonned doe?
>new claude model release can't even keep this place alive for a single thread
>>110288084wake me up when glm 5.5 releases. or gemini 4 flash
>>110288084real posters went elsweyr
>>110288084Who cares? It's pozzed and cucked.
>>110288084its 1 dollar per swipe and no proxies AND pozzed
>>110288125>>110288127I'm just hearing skill issues
>>110288135>I'm justbraaapwheres the log or preset
>>110288084nothing left here but retarded locusts and "winners"
>>110288084nobody cares about claude in 2026
>>110288094>>110288125>>110288147>>110288148ALL TRUE BTW
>>110288084>another ultracensored modelwow
>>110287399haha yeah...
>>110288360except mine is just my own cards/chats
>>110288084new claude is amazing but poorfags cant afford it
>>110288407Then you can surely post logs to show how great Claude is, hm? Wouldn't that be the best, taunting the poorjeets with the erotic escapades of a model they'll never be able to afford?
>Google makes a big deal about it's new model but then holds back Carbon for weeks before releasing it to the public>Anthropic just shits out a new Fable on a random Monday
>>110288414NTAHonestly I dont see much difference between pre-yesterday opus 5.5 (they nerfed it yesterday and I knew a new model will come out soon and here we are, new fable lmao) and fable 5.2, absolutely not worth it if you are paying the full price.
why does glm and kimi always become usable at this time of the day?
>>110288821
>>110288827ten tries in a row now.. when are we nuking india?
>>110288659Happened befote, Meta released Muse Wonder Full mogging Argon and forcing the faggots at Google to release it the next day. They will release Carbon today when it's already DOA.I hate that faggotry they do. What are they thinking? Want to feel smug?>we le heckin have this super duper model!>no you can't use it, we'll be giving it to Le heckling cyber goys first!>but you can beg teehee
>>110284790you can die from diarrhea
>>110273013The /g/ version is better than the bunker version and I won't explain why. The bunker, however, is better than the /vg/ version.
>>110289041truth nuke
>>110288953why would a subpar model being revealed as subpar make you angry in any amount
>>110289041Where did they all go? Another image board or discord?
>>110289167https://arch.b4k.dev/vg/search/subject/aicg/
>>110289184Wow. That is a chan I thought would never be resurrected.
>>110289035sure over many days of eating indian street food, but not from eating one of those
>>110289249Do you remember when this chan went down for weeks? Most anons (as far as I know) either migrated or started using that chan during and after those times.It's often chosen as backup option for the /vg/ anons, and I am not simply talking about /aicg/. /vg/ is just fucked as a whole.
>>110289333I do remember, just didn't look to see where anons went afterwards. But its on my radar now, ty. Assume vg is having moderation issues.
>>110289362https://arch.b4k.dev/vg/thread/584701652/The entire general just degraded into one guy who was willing to make 300 posts about trannies talking to himself for the most part
I hate pedotroons but like what will they do about rp that involves a child futa giantess dragonkin sonichu with hairy pits who keeps babling "deswoy me daddy uwu"? The random chaotic shit you trannies are jerking off to is just pathetic fantasy, what the fuck will they do about it?
>>110289389>child futa dragonkinI can do that easily, but everything else you mentioned put together makes it too chaotic, no?
>>110289378that was LGA btw
>trannies mentioned Classic
Argon is AMAZING for fanfiction.It is truly upgraded 3 Pro.
>>110288360Expressive a.k.a. card-with-image is helluva drug huh
>>110289609I am a visual creature in a text-based prison
>>110289614Mhm. I'm guessing you've been generating the IMG assets on your own, too. With NAI or SD or whatever. Instead of just using the premade ones.
>>110289664>>110289664>>110289664
