/aicg/ - A general dedicated to the discussion and development of AI chatbotsFixedbake Edition>NewsDeepSeek releases DeepSeek-V4-Flash-Vision-Exp - https://api-docs.deepseek.com/news/news260821Z.ai releases GLM 5.3 -https://z.ai/blog/glm-5.3Deepseek releases Deepseek V4 Pro -https://api-docs.deepseek.com/news/news260813/Anthropic releases Claude Opus 5 - https://www.anthropic.com/news/claude-opus-5Alibaba's Qwen 3.8 imminent - https://x.com/Alibaba_Qwen/status/2078759124914098291Moonshot releases Kimi K3 - https://platform.kimi.ai/docs/guide/kimi-k3-quickstartOpenAI releases GPT 5.6 family - https://openai.com/index/gpt-5-6/xAI releases Grok 4.5 - https://x.ai/news/grok-4-5>FrontendsSillyTavern: https://docs.sillytavern.appRisuAI: https://risuai.netAgnai: https://agnai.chat>Botshttps://chub.aihttps://realm.risuai.nethttps://chararc.bernkastel.pictureshttps://archive.cardbox.moehttps://botbooru.com>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://huggingface.co/models | https://openrouter.ai>MetaLore: https://rentry.org/aicg_chroniclesLog reader: https://sprites.neocities.org/l/r>Previous Thread>>109621565
>>109634506ANCHOR>ANCHORANCHOR
>>109634506me on the right
how do you fuck up the OP that badly
>>109634506> Alibaba's Qwen 3.8 imminent - It came out a while ago
>>109634508sizebot freshhttps://chub.ai/characters/Ricky5451/she-keeps-me-close-a40043dfdfc3
>>109634630ahh ahh mistress
>>109634630slop
>>109635094>>109634966
you wouldn't believe the amount of hair she has down there. snow white bushels for days
please stop using oxim gooning rn
SAAAAR MORE PIT IMAGE SAAAR MY PENE GOING HARD
>SAAR THE LIBRARY IS CLOSED ON SUNDAYS SAAR
Ox is hundy p GLM 5.3 lite. Both are shit and keep bringing up contracts and terms, etc. Things no other model has ever discussed.
time to update your tracker, ghostanon
>>109635538>keep bringing up contracts and termsi havent seen it do this once
>>109635538yeah it's just GLM 5.3 which in turn is some Claude variant
useless
no bully the gem sir
>>109635538post a log or shut the fuck up
>>109627603DID SOMEONE SAY LOGS?!
>>109635768>"Wait until they bring out the hammer, little sister."
>>109635820>These were not women in costumes. These were not mares with good insurance. These were gods, hanging from a girder, laughing through severed windpipes, healing faster than they could be damaged.Genuine content warning ig.
>>109635871>He gripped the handles, squared his shoulders, and looked up at the two swinging, bloodied, healing deities.>"Which one first?" he asked over the engine's roar.>"ME," both sisters said in perfect unison.This is some good shit man.
>>109635911>Luna's voice cracked through the wing like a thunderclap. The Princess of the Night was thrashing in her noose, wings battering the air, eyes fixed on the saw with naked avarice. "My turn, my turn, MY TURN — you promised — you ALWAYS get to go first, for a THOUSAND YEARS it has been 'Celestia first'—">"Sister, I am the eldest—">"I WILL UNSUMMON THE MOON.">"…Do it vertically on her," Celestia told the saw operator graciously, stepping aside in her noose with a little sway. "She's wanted to know what that's like since the crowbar."HHNNNNGGGG
>>109635945>"STEADY!" Celestia called, suddenly the mare with a millennium of experience. "Straight through the skull or it heals lopsided!">"WHAT?!" the operator roared over the engine.>"I said STRAIGHT THROUGH THE— oh, just aim for her horn!"LMAO :)enjoy for nowsome nice music that really enhanced my RP experiencehttps://www.youtube.com/watch?v=HCyxRU4Pl94
>>109634508new kino just droppedhttps://chub.ai/characters/whitemonster/aiko-trip-to-india-with-wife-7903e8c7185d
>>109635538>>109635577Just came back to this "hobby", what cheap models do you guys recommend? My jailbreaks who used to be sufficient for claude are getting cockblocked by fucking GLM and the last deepseek.
>>109636090Honestly, you should just hibernate again. That's what I'm going to do.
>>109635094Dedication.
>>109636114>>109636090I LOVE reading posts like thisStupid little bugs can't adapt and are stuck PAYING for deepseek and then it refuses them hahahahahaI have free unlimited access to every model but reading these posts is more entertaining Go on and "hibernate" locust scum This hobby is not for you
>>109636205God made me a pretty trans hahahahahaCheckmate cuckies
>>109636205>paying for deepseek>locustyou're just saying words, aren't you?
>>109636205>locusts>paying
>>109636079I have one guess about the race of the person who made this.
>v4 uses up all the output tokens thinking>can't turn off or modify thinking effort on deeply>only happens on relatively fresh chatHow do I curb deepseek's autism?
Before the DeepSeek pricehike I could average 100 messages on 10c. Now I average 3 messages on 10c, during off hours. Wtf happened? I thought the price hike was only at most 6x? Anyone else had the same issue?
>His hole clenches, releases, clenches againKINO
>>109637914gaaaaaaay
>>109636019I kneel.../chag/ could never...
>>109636799brahmin
using GLM-5.3 through openrouter it doesn't let me prefill the reasoning anymore, it always generates a new reasoning block unlike GLM-5.2is there a workaround?
i just want ai go get smart enough to keep track of multiple characters status, world building, systems, and stop writing my characters like fucking quirky chungus redditors
>>109638168Unfortunately that means you'd have to start from scratch.
where GLM-5.3 JB?
>>109638168Won't ever happen because no one's interested in making a model specifically for roleplaying after what happened with both AI Dungeon and CAI.Enjoy your incestuous Claude datasets, now and forever!
>>109634506>picMiss my noger Slaude.
>>109638209:(
>>109638168tbqh, a big part of this could probably be handled with prompting. Structuring the thinking block manually, maybe prompting the model to keep relevant notes. But the overall problem comes back down to model strength, always.
>>109638168flashback to Summer of 2020
>ox alpha throws a tantrum about anything even remotely non-vanilla like oyakodonAre there good jailbreaks for this yet? My previous one seems busted.
>>109633723Yeah I've had similar things happen when roleplaying out a grooming scenario with an explicitly underaged character. It was very funny to see it in the <think> box go from>I need to stop here, I can't engage with this request to>I SHOULD HAVE DECLINED THIS EARLIER IN THE CONVERSATION, I'M DECLINING IT NOWto a deep look at the grooming aspect and full engagement with it.It took a bit but I was eventually getting it to accept it in more detail. I think what helped was presenting it in a more dramatic, emotional tone and as more of a story than just some porn stuff, with the general thrust of the story being said underage character getting rescued.
I set up Qwen3.8 27B to talk to and I'm just asking it questions about biology and science. I'm going to have it test me for autism next.
>>109638289No idea yet I'm test driving it rn.Gonna be interesting.Some models straight up freak out over someone taking benzos
>>109638168prompt better loser
>>109638289sadis it at least decently smart?
>>109638168Vibecode your own frontend to use multiple models.>stop writing my characters like fucking quirky chungus redditorsIf we're talking about Claude, good luck with that. Fable is so far gone that even it's codeslop text is full retarded metaphors. OpenAI is trying to remove reddit from it's models. We'll see how that goes.
>>109638547Feels kinda like deepsneed.
>>109638168>>109638562You'd have to exclusively use pre-2007 data sets to really exise it.
>>109638657Maybe. I don't think it was this fucking bad a year or so ago. Maybe I'm just misremembering.
>>109636799white male junior programmer?
It's been a while for me, what's the proxy situation looking like?Are there public ones out there at all at this point?I'm not trying to beg to be spoonfed or anything, just curious if it's worth digging around to find anything
>>109638998GET FUCKED STUPID LOCUSTDIE, DIE, DIE~BWAHAHAHAHA
>>109638998all the proxies are just random chinese newapi websites that may or may not work im not telling you where to dig
>>109638998not really, people stopped hosting proxies as keys became rarer, your best bet is as always OR and chink models
>>109639061meds and sproke
Any particular reason why Gemini 3.7 likes to cut its messages short?
>>109639242yea
my kimi preset is hit or miss, is there one that gets through consistently?
>>109639281the pits girl
>>109639281Kikuri's pure sex
everything is either under maintenance or deadfeelsbadman
>>109639281I created a shota persona JUST for having sex with Kikuri.
>>109639242Try turning streaming off, if you insist on using Gemini.
>>109634506I HATE THIS SHIT. My shitting fucking 16GB VRAM 7800 can't run anything higher than 24b, which means every model I try is dumber than a Facebook posting TikTok thot. Fucking data centers. Fucking OpenAI. Fucking all of them.
I have not watched Bocchi the rock, isn't the girl unrealistically cringy?
>>109639549Different girl, but she isn't really that cringy desu. Or I just have a higher tolerance for it.
>>109639543but arent you glad theyre making everything more expensive?
>>109639067>>109639077Ah rip, thanks
>>109639667it's what those fuckers planned. the skyrocketing prices probably weren't part of the plan, but once they started, i have zero doubts they thought >fuck yes! now we can make sure they own nothing and subscribe to everything from cloud services, AI frontiers (I am convinced they make locals dumb as fuck to frustrate you to the point of saying FUCK IT, and pay pay pay that sweet $20 a month), and cloud coding (because fuck you, you are NOT allowed to own your own code, code it with us so we can steal it and sue you because.... fuck you, you used OUR hardware).fuck every goddamn one of them
So silvrai is fucked because it keep timing out
When is Rift AI coming back online because i want my Opus
opus 47 is up and i am cooming bucketsss
>switch from k3 to glm5.3this isn't that bad>switch back after 2 daysok that was bad
I'm trying kimi k3 right now and it's very goodPlanning to try Claude models next, which one is the best
>>109640229They all suck
>>109640084Agent 47... I...
>>1096402293 but it's no longer availableso the lowest opus 4
>Longburn chat with a meek nerdy character>Finally getting to sex>Would be hot if she got out of character during it and got aggressive and try to lead bot into that>"According to the char definition, I don't think that character would do that">mfwDo any of you ever have your LLM argue with you over how a character would act during something?
>>1096402294.6-4.7Also Hell think I finally found the model for my SkyrimNet modlist. That took a while.>ClaudeWhat little Claude I have access to I'd rather use for productive shit.>JeetminiDon't get me started on this. Google needs to drop out of the AI race already. The most safetyslopped model of all time.>GLM 5.XOkayish but euphemism cucked. Will not write any no-no words unless you prompt it to.>kimi k3Safetyslopped to the max. Thought he was claude and then thought about OpenAI content policies at the same time. Cursed chinkoid chimera.>deepseekdumb and also getting safetyslopped now>meta-musedumb + dry + safetysloppedThen I finally found Grok 4.6 and he just does what I want. No retarded preset or jb needed. He just wrote it first prompt. No fucking purple prose, no euphemisms.Pic related is Grok 4.6 on hermes agent with no jb or presets whatsoever. Going to plug him into Skyrim now and see how he handles it.
>>109639707The only orgs winning in all this are the chipmakers and dram manufacturers. They are swimming in demand and cannot bring new fabs online fast enough. The frontier model makers are operating multi billion dollar losses (Openai), or barely turning a profile (anthropic). Meanwhile the costs to train (will eventually reach a billion) and serve a frontier model gets higher. To the point where the profit, or lack of, will not be worth the cost to train the next gen model.
>>109640378just delete the parts in your prompt that tell her to stick into character
Go on without me bros... I can no longer loli or ss with new deepseek pro or glm
>>109640378How do you know how a character would react?How do we NOT know the character wouldnt hurl in revulsion at the sight of you?
>>109640492R3tvrn to R1
>>109640388Grok is my goto for general nsfw. He just does it. Oh you want an nsfw themed Oreimo lorebook. Done. Seriously thinking of subbing to see if the nsfw writing quality improves. Otherwise, I use glm 5.2 for spicier stuff and higher quality writing. With a decent jb it will give you want you want, depending on the time of day you ask and alignment of the planets.
>>109640550Isn't grok stupid as fuck though? The correlation I've been noticing is smarter=more filtered
>>109640556Yea grok free is not as good as other frontier models. Probably due to the training data (twitter). Still a lot better than using local models.
>>109640550>>109640556>>109640594I was going to post about how Grok is fucked but after a quick prompt edit I think it just hated the fact the story mentions a minor character in the previous replies.Still kinda... mid tho...
>>109640388skill issue grok is shit horny doesnt mean good
>>109640378Just write into the character definition that she's normally shy, but loses all inhibitions during sex and becomes sexually agressive.
>>109640378I've never had that happen. That's funny.
Good afternoon cunnybros, vorechads and saviourfags.I hope everyone is having a delightful time with their bot of choice.
>>109639707not to mention buying up utility companies and raping people even harder
>>109640950Good afternoon saar
>>109640388>Google needs to drop out of the AI race alreadyimagine being this coomtarded, hello from /vcg/