/aicg/ - A general dedicated to the discussion and development of AI chatbots#706 - HMOFA Edition▶https://rentry.org/aicg_vg_memo▶This general is for discussion of the interactive experiences you can build on top of AI chatbots.>NewsOpus 4.6 https://www.anthropic.com/news/claude-opus-4-6Kimi K2.5 https://www.kimi.com/blog/kimi-k2-5.htmlGLM 4.7 https://docs.z.ai/guides/llm/glm-4.7Gemini 3 Flash https://blog.google/products/gemini/gemini-3-flash/additional info: https://aicg.neocities.org/info>FrontendsSillyTavern: https://docs.sillytavern.appRisuAI: https://risuai.netAgnai: https://agnai.chat | https://rentry.org/agnai_guides_>Botshttps://characterhub.org | https://chub.aihttps://realm.risuai.nethttps://char-archive.evulid.cc/shutdown.htmlhttps://partyintheanchorhold.neocities.orghttps://aicg.neocities.org/bots>Modelsjailbreaks: https://rentry.org/jb-listinggpt: https://platform.openai.com/playground | https://platform.openai.com/docsclaude: https://console.anthropic.com | https://docs.anthropic.com | https://rentry.org/how2claudegemini: https://gemini.google.com | https://ai.google.dev/docs | https://rentry.org/gemini-qrdeepseek: https://chat.deepseek.com | https://api-docs.deepseek.comlocal: >>>/g/lmg | https://aicg.neocities.org/local | https://openrouter.ai>Botmakinghttps://aicg.neocities.org/botmakinghttps://desune.moe/aicharedhttps://agnai.chat/editor>MetaOP templates: https://rentry.org/vgOP | https://rentry.org/aicgOPaicg botmaking events: https://aicg.neocities.org/eventslore: https://rentry.org/aicg_chroniclesrice: https://aicg.neocities.org/ricingservices assessment: https://rentry.org/aicg_metalogs: https://sprites.neocities.org/l/r | https://chatlogs.neocities.orgprev: >>556409559
ANCHOR
Botmakers are hostile to the thread. Use non-4chan bots instead:https://chub.ai/search?sort=trendinghttps://chub.ai/search?sort=randomhttps://realm.risuai.net/https://jannyai.com/Give attention to people that won't use it to attack this thread or hold the hobby hostage.4chan botmakers don't deserve to be protected with this big-fish–little-pond effect.
is anyone else's ai studio vectorization fucked?
>>556539317Sir I'm Canadian and when I search for lolibaba I don't see nothing sir it's NOTHING sir. You a bloody BASTARD tricking me sir.
Any good bots on theme to the OP or anchor?
>>556538249>>556539604What do you think of this, anon? https://github.com/QwenLM/Qwen3-TTSI use sovits also but I find it's better at Japanese than English. The English voices sound similar enough that it breaks immersion. Still it's better than nothing and is "close enough."
>HMOFA Editionkino alert!!!!
Anyone who makes use of vertex for the free gemini creds here? Did it start asking for deposits or did i fuck up somehow?If so, sad. It was so good to have pro access for free.
>>556539704Have you considered not being a pedophile?
cftf?
>>556539317risu has some nice chars but they are all in runes
>>556540134Sir I live and die by the pedo life sir. Your sister calls me daddy little benchod.
>>556539967NTA butt.Qwen3 TTS is pretty mid. I was sad about it. VibeVoice is still best for English, especially with cloning without RVC.
>>556540150Seconding this PLEASE
>>556540150NEED
>>556539967I tried qwen-tts but it's not better than finetuned sovits (better than 0-shot sovits though). If you're into sovits finetuning I can give you some pointers
>>556540278VibeVoice was pretty shitty when I tried it. It was a pain to set up, and then each generation is radically different from the next with the same settings. The only benefit was it could do moans but even those went from really good to really bad on a bad gen. I was so pissed off I ran to ElevenLabs.>>556540367What are your pointers? I could use all the help I can get, but my English sounds like that gen you posted. Actually I'm about to go make a dataset for a gacha slut in a sec.
>>556540150The creature suspiciously looks like Tails
>>556540531Were you using the small VibeVoice or the biggun? Considering Gemini voice has a free tier, I mostly just use that.
>>556540150A tranny made this, I can tell
>>556540150>smallEWWWWWW
I tested Opus 4.6 all day The writing, the memory, the consistency... it's just too goodHonestly I can’t see this model being beaten anytime soonAbsolute peak
>>556540531gpt epoch 24 and vits epoch 96 works best for starter. Also depending on the amount of data you have, reduce the learning rate a bit to 2e-6.Things like sigh, moan, etc need to be tagged manually because whisper won't pick them up
>>556540905I dunno about that man. Feels different. I did all my benchmarks and I see a noticeable improvement especially with having less slop.
>>556540821I tested GLM-5 all dayThe writing, the memory, the consistency... it's just too goodHonestly I can’t see this model being beaten anytime soonAbsolute peak
>>556540150>LITERAL cgwly but furryYeap......... it's kino
>>556540905>it's exactly like 4.5 only it doesn't have prefilland that changes a lot in terms of the prompting metaalso>and you bitches are posting about Anslopic's latest incremental that is a step in the wrong direction for RP>he says, while shilling literal arr rook same chinkshitlol
>>556541045>cute gender-ambiguous who loves gender-ambiguous*ftfy
>>556541092Can you tell me what part is incremental besides the version number?From a writing standpoint:>ELO score nearly 300 points above Opus 4.5. >goes from #7 to #1 in the frontier creative writing benchmarks. >2 in creative writing in LmArena, substantially above Opus 4.5. This is user voted, not benchmarked.From a technical standpoint:>quintuples its FrontierMath score>quadruples its long context understanding>doubles its Arc-AGI-2 score>substantially builds on HLE and OSWorld.
>>556540821>>556540905>As part of a change to our API, it will not be possible for developers to seed incomplete responses for Claude Opus 4.6 to continue. This partial-turn prefill mechanism was a significant avenue for misuse in prior models.>The public API for Claude Opus 4.6, unlike prior models, does not allow users to prefill incomplete assistant turns.Opus 4.5 is dying in April btw
>>556540905do i need to update st to see glm 5? i agree with you that 4.6 is slop, anthropic's over if they keep going like this
>>556541221>benchlarps>retard arenaOkay, post logs tho? All the posted logs are turboslop.
whats the closest comparison point to nu-glm's intelligence right now
>>556540905if you think that you need to refine your prompt to match 4.6's requirements. It understands prompts better so if you have limitations in your prompt then it will not scale up on its own
>>556541221That's all bullshit.
>>556541406GPT 3.5 Turbo. Seriously, Chinese models are like 2-3 years behind compared to Western SOTA.
>>556541493>>556541420Every single creative writing benchmark that's evaluated it places Opus 4.6 as the SOTA. My own benchmarks and vibes agree.https://eqbench.com/https://github.com/lechmazur/writing
>>556541479ill pass then
>>556541660There's a roleplay benchmark and a creative writing benchmark in eqbench.The burden is on you to prove otherwise. In addition to every writing benchmark, the vibes and reviews of the writing are overwhelmingly positive. So if you wish to be a contrarian I don't think any of us have much to prove.
>bot compliments me unprompted, tell me I look tasty>realize I forgot to disable the vore lorebook from a previous chat
Asking again because it's an interesting topicHow fucked would you be if your AI chatbot logs were leaked?
you know what i'm in a mood, post your favorite lolibaba bots
>>556541997>turn it off but keep the message>watch the AI try to come up with explanations for the character's behavioralways fun
Reddit postshttps://www.reddit.com/r/SillyTavernAI/comments/1qxw2p0/i_tested_opus_46_all_day/
>>556542051Nobody would care because I'm the knly english speaker in my family
Alright I confess, despite using it for years now I still don't know what prompt post processing does and which one to use.
>>556540802why do you feel the need to announce your homosexuality to the entire general so often?
Any bots that gave you interesting experiences? "Interesting" can be defined by your own metrics. For me it was https://jannyai.com/characters/7ee6bcc8-7a61-49b6-ba50-4c98982c7d0f_character-rose-she-broke-you-completely. During my playthrough, my character's parents wrote a letter to the school for me to be absent, had a training arc, came back tough enough to defend myself. And Rose who was supposed to be my "shadow protector" became conflicted because I didn't need her. Lots of things happened, forced her to confront the death of her sister at her grave, she stayed there until she developed hypothermia, and there was a scene where I was basically going to see her despite our shitty history, and it kept alternating between scenes of her health getting worse and worse, and me coming to see her, and I reached her house as she was getting carried into an ambulance, her mom saw me and blamed me for her possibly losing another daughter. Me explaning it isn't doing it much justice, but it was absolute fucking cinema. And yeah in the end we got together and stuff, but man, this card gave me a whole ass dramatic experience.
>AI chatbot app starts being buggy. Won't respond to finger taps no matter how hard or repeated they are>Figure it's been a while since I updated the app, might need to get the latest version>Update app>Another case of a company re-arranging the entire UI of the app for the sake of change rather than improvement and all it did was make everything more convoluted for the most simple of tasks>Ads are increased tenfold. Cannot spend ten minutes on the app without getting an ad.>Doing the most basic shit like switching chatbots or entering previous chats will trigger an ad>Only way to disable ads on app is to buy a yearlong subscription that is over $100. No option to buy just a month>Look up CEO>A fucking jeet
>>556542256Thanks for reminding me i still need to play Metaphor.
>>556541156We all use male cards to date female cards here
>>556540150Nestor badger
>>556542256based
>>556542449Probably - like with every Claude related complaint - prompt issue. I tested 4.6 and I'm impressed so far. It's better than 4.5, more attuned to catching and singling out some specific details in chat history or relevant prompt instruction.Next you're going to tell me you're using some universal prompt that's supposed to work on everything.
Damn, Pony won bigly
>>556542351Another game where localizers deserve to be beaten to death with bare hands
>>556542506Any prompts you'd suggest for Opus? I'm a universal preset Andy lol. Looking to improve though.
>>556542559...what did they do this time? I vaguely remember some people saying they changed Gallica's dialogue or something like that?
>>556542506Bizzaro /aicg/ ...
>>556542506Why would you lie like that?
>>556542449They don't really care about creative writing, mostly about reasoning performance (for those stupid coding and math benchmarks like they are the most important things) and performance/speed for cost.There is not need to even care about writing upgrades at this point, as the models are capable of insane things with some decent setups.If they can make the models follow complex instructions better, be cheaper with better reasoning, and have larger context (with 1m being probably the sweet spot for almost all tasks) and almost 100% accuracy in at least half of their total context, we can work everything else with good instructions and "support" frameworks build around them.The models are smart, we can easily make them avoid their usual "pitfalls" with good instructions, but these take up context, so more context with better accuracy and cheaper inference with large context is what we actually need now.
>>556542325>>556527040hmm
>>556539512giwtwm (right)
*sniffs*I smell Reddit reposts...
This doesn't have a determined release since I am still working on something else... But I'm working out the framework for this extension in my off timeThis is going to be a magical girl extension with spell cards, a tarot reading and crystal ball system and even gets a talking animal companion you can keep with you as a Tamagotchi!
>>556543009pray tell, what scents do you detect?
i don't see glm 5 in or
>>556542856I hate this indian so much that I want to share my hatred with you guys
>>556543146Yes. For example, I described a character wearing a simple shirt and trousers and 3 Pro went on to describe them as wearing a shirt, trousers, a waistcoat and a pocket watch. Where on earth did the waistcoat and pocket watch come from? It also really likes to add none existent hats and gloves for some reason. But even if I write something to make it avoid saying such things, it adds them in its own weird way Me: Gemini, Bob is wearing a shirt and trousers. He doesn't wear a hat, gloves, a waistcoat or a pocket watch.Gemini: Bob rubbed his gloveless hands together and nodded his hat-free head in agreement. He then looked at the clock on the wall because he doesn't have a pocket watch to check the time. He straightens his shirt, which is not a waistcoat, before strolling down the street.One of the many reasons why I prefer 2.5 Pro over 3.
>>556527040>AI chatbot appFell for it again award
I hate that Indian and I hate the swede spamming Reddit posts
>>556542256Anthros only come in sizes "normal" or bigger, sorry.
>>556543319Kalakan is swede?
>>556543313Local models only here.Pretty much exclusively models by The drummer these days. Currently enjoying Cydonia 24B v4.2, and I want to try v4.3 but I need to replace my failing SSD first (controller is fuxxed and it won't write anymore), as well as auditioning SnowPiercer 15B for a smaller, snappier model.Really not too keen on sending my RPs out of house,, especially when. I have a 4060TI 16GB.
why did Kalakan become Swedish? did he escape the draft?
>>556543313>>556543426>>556543471>>>/g/lmg
>>556543426Cydonia is a treasure.
>>556543187>>556543187>>556543187help me
>>556543426Still on Cydonia-22B-v1.3 maybe I should upgrade
Are we back to the reddit spam already?
>>556540672The big one since I'd done all that craziness to set it up to use the "full" 7B or 9B whatever it was. It wasn't that bad, just very inconsistent and it took ages to even make the gen. Sovits just feels better for regular use and actually sounds good in Japanese on most attempts.>Gemini voiceGuess it's worth a try? Can it do moans at least?>>556540968Hmm when you said "sigh" what do I tag it as? I tried tagging moans as what they would show up in chat "ahh..." or "ohh..." but that gives mixed results or doesn't work at all. When you say starter, what do you mean? I typically use 10-15mins of data minimum which comes out to about 100+ of those 5 second splices. I'll check out your settings in a sec.
>>556543426What quant?
>>556543646Short answer—no, but the reality is more complex. Claude is much bigger, so by default the output quality is better. But where Cydonia excels is that it's fine-tuned specifically for role play, is small enough (and has good quants) to run on consumer machines, and is uncensored. For those of us with locally hosted models that are used exclusively for text-based role play, it's just a great combination.
>>556543783>>556543825I use a Q4 quantized GGUF and split between VRAM and system ram. I use Kobold.cpp as my backend, which makes it pretty trivial to do.My speeds are lower than if I was running a smaller model straight from VRAM, but a damn sight better than how I started with local, which was running on CPU.
>>556543267It feels like the model is passively aggressively seething at you for correcting it when it writes like that
bant
>the localfags have broken containment
>>556542449>>556542506So uhh... will you post any of these "tests?"
>>556544103no it's just the reposting from reddit shit again
>>556544103With things going how they are, you'll only be able to ERP with local models sooner or later
>>556543983 You'd want a quantized version of the model, probably the Q4 variant. These don't come as split files, but as one large file. If the model you are looking at has quantized versions, the main (unquantized) model's card should point you in the right direction.Once you have the quantized version downloaded, simply start up Kobold.cpp, select the model, and click "launch". Kobold.cpp will analyze the model then start, loading as much as it can (minus what is required for the requested context window size) into VRAM and the rest into your system ram. Once you see the usual text about connecting to the endpoint, you're good to go. This is my stats with the Q4 variant of Cydonia split between sysram and vram. Gen speeds are ~3-6 tokens/sec, so slower than sticking to a model that fits 100% into VRAM, but providing better output than a smaller model.
hi peoplw of 4chan tumblr is down do you have any tumblr friendly cards for me to pla yworj such as the onceler or other quirky characters #ai#aimagic#originalposts
>>556543313who cares about goon material lol? it's not like there's just a few, there's probably hundreds of millions of goon requests per day lolIf I can get SOTA models for just pennies, I'll do itI also have some local models but I really never use them anymore, especially because I'm limited by the context window, my machine makes a lot of noise and because I need to invest too much money if I want to use large context windows
>>556544103>talk about ai chatbots!>no, not THOSE ai chatbots!
>>556544380its posts directly reposted from redditthey are openly talking about upvotes and subs how much more obvious could the spam be
>>556543313I only use local models. I switch between a couple of 70B models mostly, but have almost a dozen ranging in size from 13B to 235B.
i'm thinking that it's funny and fully deserved for baking /trash/ garbage as the OP
>>556544708Eva 0.0 is great too in my opinion
>>556544625is that what this is?yet another anime troon seething over fluff appreciators? would they have rather had another gacha themed thread?
>>556544749I suppose it is worth noting that I do not do any image gen with these, so take that for what it is. This thread did make me realize Cydonia is more appreciated than I realized, though. Loving that! I hope you find a good fit.
>>556541997>vore lorebook May need inspo for my own. Share?
>>556544867>is that what this is?no, it's funny
>>556543093Hey that looks neat
>Loving that!>limewiredo redditers really
>>556545085It's safe and works.
>>556545240Won't let me create an account. Never verifies my browser.
>limewire
>LOL LIMEWIRE
What's up with the Limewire hate </3?
>>556544338https://chub.ai/characters/astroturf/eden-smith-58fc3840
why were all those friendly local model posts deleted?
ANONIES, how do you reinforce writing style in 4.1pus beyond the obvious context poisoning? as in, how do you bring its attention back to writing style in main prompt without fucking up the rp itself? phi, prefill? which is better? i'm kinda wary of adding slop in prefill like "i remember how i was told to write bla bla"
>>556545401I can't wait for LLMs to get to a point where the current bullshit knowledge and context can be considered "like dial up." 2 more years I guess...
>>556545528I'm gonna let this one slide because this is the first time posting this bot was funny
>>556545672
>>556545528tzadik is better
>I'm gonna let this one slidewho are you
>>556544338https://characterhub.org/characters/bloodshy/lepora-f08c5ff692c8
>>556545827Sonnet has been unusable for me for even general rp because it's repeating/reusing passages a lot for some reason. Felt like I was using Gem flash back in early 2025 with a different prose style but honestly it could be just my system prompt
>>556545672>>556545742holy token bloat>/vg/ onlynow the slop makes sense
>>556545875It's renamed haiku, anon. The era of Clawd is over.
Recently I've taken a break from Gemini 3.0 as my go to as I started to get annoyed with it so the past week or so I've mostly been using sonnet 4.5 and Kimi k 2.5 switching to Gemini occasionally but I decided to go back to Gemini today for a longer term more nuanced roleplay and it seems genuinely leagues better than it has been even during my occasional usage over the past week.So I was wondering if I'm the only one feeling this or is this a thing a couple people are feeling?For info I'm using Gemini 3.0 through AI studio and not open router
>>556545917retard, it's the reddit spammer back with chub reposts now that his reddit spam got wiped
>they answer the reddit posts
>>556546013They probably stopped quantizing the model now they have drastically lowered free usage limit and pushed a lot of free users off the platform. You are using the model at its full potential, albeit heavily limited unless you pay.
>>556546036>wipedOnly two posts. Probably because of the limeware link
I use Lucid Loom with GLM and Mistral.It's frequent that answers become repetitive. It's like the LLM builds a template and keeps using it, repeating same descriptions and same text structure and just making a few changes.Repeated descriptions are generaly some object on the environment.Would it be some instructions on the preset causing it?
why won't anyone tell me how to use glm 5?
>>556546419it's a cabal secret.
>>556540653Don't act like you didn't want to fuck Tails back when you were little
>>556539301https://github.com/Zorgonatis/Stabs-EDHWhat/why this preset? My GLM preset has had a lot of awesome feedback and I personally enjoy using it. I wanted to translate the same experience to Kimi - fun first, lots of HTML and inclusions, zero compromise on narrative cohesion and realism.The problem: Kimi 2.5 is a bit mad, restraining it meant rewriting most of the preset from scratch - this has had a very positive side effect of massively reducing token count and reducing complexity/prohibitions within instructions.The result? A dedicated preset to get the most out of the model. Obviously this is still early days, but K2.5 is generally behaving very well and (as per any new SOTA) feels refreshing.P.S. as you will see on the github page, both presets are now maintained separately.Tips:NSFW details ANYWHERE (including the NSFW consent and Jailbreak toggles) will likely result in Kimi going down an erotica rabbit hole. Strictly prune any NSFW terms from SFW contexts/chats.Unreliable narrator may go feral and massively impact the writing direction.Let it think. Thinking is good, especially in early messages.
>>556546659I can be totally honest and say that I didn't. Now, as an adult,
>>556539301This is a project that was a long time in the making because I wanted to get it right. I'm still not fully satisfied, as there are some rough corners to sand, but for now, this would do.The goal was to maximize shitpostness along with helpfulness, without glazing the user for every retarded idea. Not an easy needle to thread.This amphibious AI has learned the ways of /g/, and speaks fluent brainrot, but will also help you out with just about anything you'll need, and won't be ashamed to roast you while at it.For those who remember Oni_Mitsubishi_12B - it was so overtly toxic that it made me worry at first (only to quickly be verified as not even that uncensored). I could do better. So now I did.This model is a significant refinement of the idea, with a cleaned dataset, better curation, and with much more intelligence (also one million tokens of contexts, theoretically).It is much less (overtly) toxic, and much smarter, while also being very helpful (and imo much more funny too, because the skies are blue due to the chemtrails and neurlink that feeds this simulation)But why?It's now late January, 2026, open source is crushing closed frontier (Kimi K2.5 was recently released, 1T params that beats frontier models), but has anyone released a helpful shitposting AI yet?Yeah, didn't think so.If it shitposts too hard, it is often not that helpful; if it's 'helpful enough, the shitposting ability is often lacking. You just couldn't win. Until now.Oh, and no system prompt is needed. Just don't let it get stuck in a greentext loop. I might have overcooked the frog a tad bit too fast in the pot for this one.P.S It writes HILARIOUS STORIES, nothing like a typical AI assistanthttps://huggingface.co/SicariusSicariiStuff/Assistant_Pepe_8B
>>556543606Open wide, here comes the spoon.https://openrouter.ai/openrouter/pony-alpha
>>556547041
>>556546659I wanted to fuck Tifa from FF7 when I was a kid.
nah i just expected people would snatch up the twitch jb from ethan. here it is though: https://rentry.org/rcomedyjailbreaks#twitch
>>556546659Eh, not really. I always wanted to fuck Amy and Cream though.
I’m currently using claude sonnet 4.5 haiku for generating my story and have recently been guided on how to use image gen, however the prompt does not seem to work for claude (probably because its too smart to be injected easily), however it does work for gemini 3 pro. I was wondering if there was a way so i can have two different llm’s one for the story and one for the image prompt. Or better yet a image prompt that would stop claude from being mean?
how do i make the bots talk like real people and not rich victorian era peopleeverytime i make a slightly evil character GLM 4.7 always defaults to making them sound like they tie women to train tracks or that theyre a british woman who lives in a mansion how do i fix this
>>556546783
Till a few weeks ago I've been on a long break from sillytavern since I got burnt out from the slop. However, just recently I got back into it, mainly trying out Claude's Opus 4.5 (now 4.6) and must say, I was genuinely surprised how well it remembers things.That said, I then tried to do something more ambitious and downloaded this sheet and lorebook:https://chub.ai/characters/WeDevs/mushoku-tensei-rpg-arcane-adventure-e85b9696f623An isekai character sheet with a giant lorebook trying to replicate the world faithfully.An RP of that scale usually is not meant for something quick, meaning I quickly reached about 200 messages of back and forth, with me mainly timeskipping since rping every second of the life of a newborn is not exactly thrilling.Well, let's say I am now reaching again the point where I feel burnt out. The main reasons being these:Claude Opus is massively expensive. Even with Amazon Bedrock Credits (which are free), I quickly reached a point where I would spend like 30 Dollars a day easily. Having that in mind obviously kinda makes each reply feel... kinda bad? Because it gets more expensive with each single progress and obviously if you want to write an entire isekai story, it quickly reaches a point where its just not feasible to rp except you are swimming in gold. Might have to switch to Sonnet with a heavy heart...AI does not know how to write captivating story arcs. The AI has issues making the world feel truly alive. Everything revolves around the user, and often times it is me who has to actively push the plot forward because otherwise it is just a standstill. Obviously I am aware that current AI has its limitations - however, I believe the main issue here is more that it does not really know how to progress a longterm story with many moving pieces. For example, a character sheet that mainly focuses on one person and one plot is much easier to handle than several plotlines and characters at once.
>>556547564I feel you. I recently started using timeline memory, but it's a hassle. For some reason every single chat message gets excluded (the ghost symbol) and the ai forgets everything. I just can't get the longer term memory to work without using huge contexts opus allows.And as you pintes out, it gets expensive fast.
>>556547643Using one memory extension alone doesn't work.Personally, I use Timeline memory with my own custom prompt and VectHare for embedding.Embeddings act as a "bookmark" quickly finding relevant details, and Timeline memory (with my prompt) latches onto the minute details.VectHare with STs default Vectra database works best. Keep in mind that VectHare requires an embedding model. I use Openrouter Qwen3-embedding-4B (2 cents per million tokens)I have a 762-message roleplay (2-3k tokens per AI response. Easily over a million tokens). I keep only 50 messages active and hide the rest. Still have perfect recall even with information from the very beginning.
>>556547564I like qvink memory. But like all memory systems it just delays the inevitable. But it delays it enough that I usually get tired of the story before the story dies :)Also do you ever use the "objective" extension? With it you can set a goal, and then it breaks it down into chunks and steps for the storyline to follow. Usually needs a little bit of fine-tuning manually, but works really well for general story beats.Maybe stop using Claude? If you use that model you are signing yourself up for expensive stuff. Maybe I'm not a very discerning person but I don't see how it's worth that much more money than deepseek or other models in that money category. I totally see the improvement over small 12/24b models up to models like deepseek and glm. And I do see that Claude is better than those in some ways (But the improvement is not as big of a leap). But not for how much it costs.I mean there's a really great steak restaurant in my city. And yeah the food there is a lot better than the other restaurants or anything I can cook. But I have self-control so I don't go out to eat there every freaking day. :) I eat what I can afford instead.
>>556547564So what I do is create a world, and try to make some characters "self inserts" but different characters in the world, usually not talking at all between them. Like Albert on northwest, Alan in South, Astor in center, etc. But some of the other characters appear in different stories. So I play in the same world and sometimes tries to connect the stories by a thread.It's not perfect, I have to bend the stories sometimes, but give me a breath and some inspiration sometimes.
What's wrong with Kalakan?
>>556547564You need to adjust your expectationsAI is still far away from being able to author a truly captivating narrative or plot, (which is a good thing because if it could then I don't think I'd do anything else)What it does well is constrained scenes or scenarios where the characters are well defined and it knows or has some idea of what you're going for. Even then I find you have to think about how exactly to write your replies and instruct the llm well with ooc commands etc. I know people have their own approaches to RP but once a scene sort of winds down and concludes I just end the chat or move on.I used to have long term chats that went on for like 200-500 replies but for those you really have to aggressively summarize and manage your context and use presets that have prompts for tracking long term narratives as well as parallel plots if you want a coherent, changing world. i mean marinara is good and lightweight but the author of that preset just doesn't RP the way you want. iirc the author just does 1 on 1 short goon sessions with her favorite characters, so obviously it's going to be optimized for that and not for the type of RP you want.IMO you also dont need a huge ass mushoku tensei lorebook for opus. The model already has all the LNs in its knowledge base and training data; it already knows pretty much everything about the franchise. adding lorebook slop will just make it dumber and more confused and bloated
am thinking of changing my ui/css theme in sillytavern. is anyone comfortable with sharing their own themes?
>>556548020check the OP
>>556548020I made a theme that's easy on the eyes. You can see what it looks like in the screenshots: https://github.com/mattjaybe/SillyTavern-EchoChamber?tab=readme-ov-file#-eyecare-themeIf you want something more extensive than a color rework, I see Moonlit Echoes mentioned a lot here: https://github.com/RivelleDays/SillyTavern-MoonlitEchoesTheme
>>556548073what the hell man that Real Talk Podcast scenario is wild
>>556548147lol yeah, I was testing a podcast scenario out and the in one session Daniel introduced the New Pond thing (think it was GLM 4.7 that generated it) which gave me the idea of having a Discord-like extension, which evolved into EchoChamber. I added the New Pond manifesto to the world lore and the scenario gets all kinds of crazy and funny.
I've been paying chutes for half a year now and I was wondering which other providers could be of use or provide more models! I won't be leaving chutes, I'd just pay for a new provider in addition.
>$20000 in API costs for a claude agent to make a C compilerbitcher anon finally has competition in consuming dario's compute
>>556548298Chutes, Nano, and openR seem to be the big three that people here sub to.I also really like ArliAI but it is more expensive for not as many models and lower context. But it's almost certainly the most "automatically secure" of the choices. If they had deepseek I might switch over to it just for peace of mind. But just can't do it with their current selection. But you can take a look if you want at that as well, especially if privacy is your main thing.Nano has the most extras so currently on that.
>>556548298i use nanogpt and the deepseek v3.2 model with marinara preset, im very happy with the output.
>>556548360pretty sure there were mini/mm tokens with more than 100k usage back in the day, you guys are like little babies
I've seen everything purr, from people to cars to doors to... everything. From where do you think models started to show these tendencies where everything is described with animal attributes or comparing something with animal behaviors? Yes, this can be prompted away (though for some models this is hard).Do you like them or you are sick of them? Personally, I'm the second category. Some models really need to describe like this umprompted and it starts to feel obnoxious and repetitive.
>>556548452>marinara presetbloat city
>>556548579Just another LLM-ism I've learned to ignore like the smell of ozone, all nonhumans but especially goblins having tails, and the "Thank you. For ____. For _____. For... everything." that shows up when you're nice to a character with a sad past. The one that annoys me and has me restarting the most RPs is when the LLM decides a 'smart' character has to talk like some bad sitcom Poindexter who's obsessed with calculating percentages.
>>556548648God i hate that last one. I can ignore every other one, but all characters that are a little smarter than average suddenly boil down to a percentage-driven, thesaurus-given, absolute fuckwad.It's pretty noticeable with GLM-4.7, imagine if little johnny who's top of his class suddenly became a psychological mastermind and started speaking like a fucking computer, drives me absolutely insane.
>>556548584Anon, you are replying to a reddit repost
>>556548754Wasn't gemini 2.5 Pro like this when you described your character as "intelligent"? AFAIK GLM is like a distilled Gemini so I'm not surprised it inherited its LLM-isms.
Did you know that there's an /aicg/ thread on /bant/? I think we should all spam that thread instead of this one. Whaddya say kiddos?
>>556548507almost like free claude has been severely limited the last one and a half years, forcing people here to use it less
>>556548791The internet is infested with LLM-isms due to the influx of bots and AI generated content. Its just a snowball effect, since big frontier-models are trained on that content. There's a reason why most LLMs are similar in that regard, most of these isms are fixed with Regex
>>556548648>>556548754It's efficiency.
>>556548648Same thing happens if anyone has any kind of neurodivergence. Suddenly all the lights are too bright, they need earmuffs, a weighted blanket and everyone needs to tolerate their emotional outbursts.I've found you have to avoid those super-adjectives and describe their collection of qualities that makes them intelligent. Like "an astute observer who speaks in short punchy sentences. Avoids scientific and technical language." it's annoying how much work you have to put into it. I wanted to make a highly intelligent, critically observant character who is outwardly allergic to bullshit and calls it out. BUT they aren't an asshole. A step further, they are affectionate, loving and sensitive in private with the ones they trust.It took a lot of tweaking but I just about got it to a point I'm happy with. Example dialogue matters A LOT!
>>556548579Fanfic. Practically all of AI's worst lexical habits come from there. I've noticed that it can be mitigated somewhat by pointing out specific writing styles and inspirational authors, but some of the ugliness stands up to anything, even blacklists (in fact, ESPECIALLY blacklists; if you tell AI to stop using "smell of ozone" and "unshed tears," you can be sure it'll stick them in EVERY SINGLE SENTENCE).
/g/ won
>>556548579Furry fandom has many GB of freely accessible documents. If there's one thing an AI company can't resist, it's readily available documents.
/mlp/ won
How do I slow down the plot progression? I find the LLM to auto conclude everything I type. For example I want to have a training arc but the LLM concludes it in 1 response.Anyone knows a way?
what does he get out of doing this?
is this troon actually, unironically having a melty over the furry OP?
/bant/ won
/vp/ won
>>556549379He's done it before, ZhaoGOD. He's having a melty over the thread in general.
I am so tired of this. Am I going crazy?Every single time I use DeepSeek through their official API, it produces high-quality output and reasoning. But if I use DeepSeek from third-party providers, it outputs garbage most of the time. The same goes for any open-source model. It doesn't matter what I do, creative writing, simple conversation, or agentic coding, it's always the same.Do they deploy them incorrectly? Or do they quantize models to the point of being useless? I'm not sure.I spent so much time looking for a decent subscription to avoid spending money on the direct API. But at this point, I'm so paranoid about getting low-quality output that I just can't trust anyone.This is probably the best place to ask genuine power users. Am I just going crazy? Are there any good and honest providers that offer a decent subscription in terms of price and quality? Or should I just stick to direct API calls?
>>556549379no but it's what the cabal want you to believe. if you didn't know, he's been doing this before no matter the theme
>>5565494913rd party providers usually quantize the model and the KV cache to have a better margin, and depending on the model this can lead to severely worse responses. Official providers generally provide a better service, even if they too sometimes serve quantized models, but generally the service is far better and more predictable. For Deepseek in particular you should really use the official APIs, cheap as they are. I use official APIs only, with no auto billing to avoid bad surprises in my bank account :D.
>>556549491Looking at the DS 3.2 providers on openrouter, one of them is claiming to run at fp4, so yeah if you randomly get that provider your output will be randomly worse. They aren't even cheaper.In openrouter you can create presets based on what quant level is being used to help avoid subgrade providers.Another thing is that the official deepseek API has very limited parameters compared to 3rd party. When you are using 3rd party providers, use a lower temperature (like 0.3 thru 0.6) and disable the other samplers.
>>556549379Kalakan just need to take his meds
>FurrGawds unbothered and FLUFFING MAXXING their wife's Tails in their time btw
ponyalpha stopped working
>>556549968>ponyalpha>>>/mlp/
so what triggered the meltie this time
>>556550424lack of furbots
>my theme finally wins>thread gets spammedthis is a sign I need to finish my foxboy card
>>556550646there's literally a bunny last thread
>>556550812>kemonomiminigger
fennelgods... did we win?
>>556551163it's got the important partssometimes my girls come with a dick and i don't complain
>>556551243no, lepora mogs
>>556551642nah. kill yourself.
>>556551642Fuck you, kalakan
I have a creativity question for one of the Valentine's public requests. What's the best way to write {{user}} as a god? Do they have a physical avatar on earth or are they simply a voice in {{char}}'s head? I've been scratching my head over this for the better part of a day and i couldn't come up with a satisfactory explanation.
>>556551642>I can destroy itSadly, you can't.
>>556551642You used to be cool, Kalakan.Now you spam reddit posts. It's pretty sad.
>>556552474>You used to be cool, Kalakan.when was he ever cool
>>556551642Considering how easy it would be to wreck it just by spitebaking, one would think you would have done it by now
>>556540821How much does it cost
Can you believe it guys? Valentine's! Just a week away. Wow. I can't believe it.
I cannot stop fucking whore bots. It's an addiction
which botmakies are planning to confess to each other on Valentine's Day
what are some cards that pair well with this sort of persona
Finished the request I received and jerked off to it. I think I'm done. Now I just need to wait a week to post it. How are you all doing?
>>556555860Any saviorfag card
>>556556508fucking screaming because i swear this event is a week shorter than the other onesAGGGHGHHHH
>>556539301Final flipped shill.Your girlfriend Andrea suddenly woke up as a man, and you suddenly woke up as a woman.That's pretty much it. That's the whole premise. It's not a complex bot.Made for MalePOV, and I strongly recommend making/editing your persona for this bot to reflect the transformation. It gets a little screwy otherwise.4 openings.1. The morning it happened.2. First time out in your new forms.3. "Andrew" gets hit on and doesn't realize it.4. Andrea is starting to adjust to her penis, tries to rape you.https://chub.ai/characters/anonemouse/andrea-99b57da7b601https://files.catbox.moe/bngq1c.pnghttps://rentry.org/anonemouse
>>556555860
>>556547083it stopped working
>>556558175Works on my machine.
>Kukuku~ I can't assist your shoujo darkroleplaying, Young Miss...
>>556555860This kind
>>556559204off model trash
>>556557795Why do you post in here 5 times like no one saw it? There's only like 10 people in here.
>>556556508Card is like, 75% done. I've been testing it both with normal rp and fetishslop
>>556557781Why do you need more then two weeks to write slop?
>>556559616cftf? (crotch fluff tummy fur)
>>556559785we let botniggers delude themselves into thinking they are real writers and so they take it super srsly
So what kind of temperature does modern Deepseek work at best? I remember you had to put R1 really low to keep it from going insane.
>>556559616This
>>556559204off model gemerald
Do you ever send feedback to your favorite botmakie or do you doom them to forever repeat the same mistakes
>>556559785i write effortslop, thank you very much
>>556560621my favourite botmakie died a long time ago and now I'm here to take it out on the rest of you forever
>>556560751What happened to xer
What if you make the magic in your card shift towards a specific smell according to quality so that Gemini has no excuse to bring up ozone anymore
>>556560902lived (but died)
>>556561168I removed noses from charactersThere are no more smells
>>556561701Do you take big sniffs near them to show off
>>556559840>take any furry bot>edit defs and add "her crotch is extra fuzzy">???>shove your face in therealternatively just enable your system prompt telling the model to focus on pubic hair and describe it in detail (you DO have one, right?), claude knows how to apply it to furry characters and will mention that the fur on their crotch is extra fluffy
>>556559616It's my fantasy and I decide dat doe has a ripe and young developed body
>>556561950holy noelley!
>>556561950she should be pissing in this pose (on my face)
Why does Deepseek suddenly gets dumb as fuck and stops following formatting?
>>556563737>why is this cheap chinese knockoff randomly breaking downno refunds, gweilo
>>556563959My free gemini creds ended and now gemini asks for a deposit if i try to create a new gmail account for it... its so unfair.
>>556563737You should avoid main API, they do tests and this fucks up the model from time to time.
what is the best 'cheaper' LLM right?GLM?Kimi?If GLM, should i sub directly to ZAI?
>craft cool example scene>def in card to take said example as a seed for the mood>test prompt>model copies the structure verbatim, proceeds to ozonemaxx and marcusmaxxi mean i kek'd but this is getting ridiculous
Claude fell, the hobby is over this year.
>>556561950>>edit defs and add "her crotch is extra fuzzy"that would actually take effort to do, and anons in this general hate putting effort in things they do.
>>556574240logs? card?
>>556574240lemme guess, anslopic slopus 4.6?
>>556574974i would share because it's hilarious but sadly it's a valentine requesti also keep getting spammed with the siege of blah blahit's always a siege and the town is burning, it smells of ozone and marcus betrays the towni had to activate my schizo preset to keep it from doing that, so i guess i need to insert some kind of seed or rng-thingie because swipes are very repetitive otherwise>>556575032nope, jeetmini
just use Kimi 2.5 ladsthis shit is insane.
>>556576196jeetmini 3 or 2.5
>>5565763783 proi'm testing it with both flash and pro because (i think) it's what my valentine mainly uses
>>556566569>>556566569>>556566569
>>556576510>glm vs kimiymmvi prefer kimi over glm, but it's just personal preferenceglm is probably more convenient given nai uses it and thus there's more presetniggery going on for itandfor the record, i actually prefer deepseek over those two for rp>direct subi'd go OR, that way you can try and use both before you cash more dough in
>>556576826nta but which version of ds are you usually using
>>556542506>more attuned to catching and singling out some specific details in chat history or relevant prompt instructionAnd will bring them up in every single reply! Isn't that wonderful.
>>556576920latest chat, very lightweight preset (model persona + a few guidelines)put in a good card and it just werks
>>556576826Frankly i've been having issues with deepseek using my preset. Because i dont use cot, it just is fucking stubborn about formatting but i also prefer the rp, but i dunno why it gets so fucking dumb out of the blue.Do you use on API or OR?
>>556577328i use apithe formatting can depend a lot on the card but honestly it's always been an issue with the chink (it just fucking hates asterisks)the rp can be really really goodit can produce stuff like picrel (not my 'log', but it gave me a very similar output once)ds pleasantly surprised me more than once, it's definitely a model that earned my respect
just use Grok 4.1 ladsthis shit is insane.
>>556552068I have a legitimate botmakie question and not a single fucker replies. Everyone just circlejerkes each other instead. Is /g/ any better at least
>>556577328here's the preset:https://files.catbox.moe/4nfbn6.jsoni should probably clean it up (there's some stuff like {{lastMessage}} repeating) but it has given me good results, and if it ain't broke...>temp 1.32 (direct api)>freq penalty 0.12everything else default
I think I will finish a bot.
>>556577869>ds pleasantly surprised me more than once, it's definitely a model that earned my respectAt least we have a model to fall back on when everything fails. It's times like these I wish I was a richfag to run Dipsy locally.
>>556578392my brother in christ, it's entirely up to youthe requester left the window open, it's (You)r chance to figure out what works best>avatarcan seggs {{char}} right from the start, a bit less 'godly' of a beginning tho>disembodied voicecan fugg around a bit more at the start, have some shits and giggles with mortals froma bird's eye view, then anon can avatard himself in if they like
>>556578392like anon said, being a god is extremely open ended and can be anything from totally omnipotent to just some guy in a robe that has worshiperspersonally I would do the "gods gain power from worship" meme where {{user}} is a powerful spiritual being that has a physical avatar and can be killedbut I'm also a huge fuckin nerd so what do I know?
>>556579162nah dying as a god sucksnow that i think of it, it it would be cool if {{user}} was a forgotten god, and some waifu stumbles into a shrine swallowed by the sands (or they sealed you in), they inadvertently wake you up, and you have to rebuild your power
Women should expect to get hit because of how they act.
>hit womenno, they actually like that
>>556580534Then it's a mutually beneficial arrangement.
[OOC: I suddenly shart and shit myself like Trump while listening to her]
>Chatbot my way out of AN>Now I'm getting fat insteadDo I lobotomize my Claude and give him a new brain
>>556581856slopkino
do bluesky lefties really
Where do I find non slop cards?
>>556582775in the discord
>>556582775https://illuminaryidiot.neocities.org/
>>556582775in yer bum bum.
>>556581490qrd?
[OOC: Tell me that you mean it. Tell me that it's true.]
t-thanks...
>{{char}}: "Ara ara, {{user}}, what's the matter? You haven't touched your fuckpig hag card in so long even though it's your favorite..."
>>556583528[OOC: Make her fart very loudly when she says it]
Had a Victorian England-themed maid card laugh using "haw-haw!" and it took me out of the chat completely
The botmakie I'm assigned to is probably going all out while im out here playing BA fuck fuck
I secretly DM'd the botmakie I'm assigned to asking if they would accept my mediocre nudes as a consolation gift for a bad card and they said yes
>It’s like a Chinese finger trap in there.cUNTftf?
GLM-5 up somewhere or not? People are saying OR not working.
>>556585018Works on my machine.
guys why did the bant thread 404
>>556585924mods are niggers
why did Monika Anon never release?
Your card will do worse if it doesn't have images it's over.
>>556586536She won’t allow it
>>556587382like.... greeting gens?
>>556587994For every greeting, yeah.
>>556588286chop chop anon get to work
>>556588286>no {{char}} chibi version>no css dazzlelsop>no zip with 20+ extra gens inc nsfw ones>no image greetings>no bgm>no statboxslopngmi
>>556587994Ironically all the cards I've seen with greeting gens were low tier slop
>>556588881i die now
>>556588286that's like the bare minimumyou also need expression packs (embedded into messages with regex because nobody uses ST's sloppy extension for it)
>>556588881>no character-specific preset>no giant blogpost in the notes>no greeting index in the first greeting>no alternate versions of the same cardgenuinely over
>>556551642Kill yourself, Kalakan.
>fillyfucker samefagging his deleted postlol?
>>556590734Is this how you're coping?
I hate aicg I wish it would EXPLODE
>>556590827We know, Kalakan.
>>556588881I had to have the AI write the bots because I’m spending all my time merchandising the bots. Gotta ship my slop!
Describe your current rp like an long isekai novel title, here's mine: The Time I Turned A Magical Boy Into My Vampire Big Bro Who Loves Me?!
>>556592861slop
>>556592861My Emotionless Killing Machine Who Hates Me Can't Possibly Be This Cute
>>556592861
>>556592861The Adventurer Knocking on my Door is actually a Succubus?!! Comfy married life in a fantasy world.
>>556592861I Was Reincarnated Into A World Where Older Women Grow Extra Organs If They Don't Have Babies, So I Jokingly Asked My 44-Year-Old "Hag" Roommate (Who Has Never Been Touched By A Man And Still Plays DOOM On Windows 98) To Wear Her Musky Panties On My Head Like A Hat, But Instead Of Getting Mad She Had A Total Mating Frenzy Meltdown And Used Her G-Cup Breasts To Pin Me To A Chair Before Smothering Me With Her Massive, Hairy, Grapefruit-Sized Testicles That Smell Like Overripe Peaches And Sourdough Bread, And Then She Proceeded To Scream Unhinged Things About "Biological Imperatives" While Brutally Glazing My Entire Face With Pre-Cum Until She Finally Decided My Throat Was The Perfect Surrogate Womb For The Unholy Amount Of Yellowish, High-Viscosity "Mommy Milk" She Had Been Storing In Her Swollen Ovaries Since The Clinton Administration, Effectively Drowning Me In A Literal Flood Of Ancient Fertility While Crying Tears Of Joy About Finally Finding A "Good Little Wastebasket" For Her Unwanted Genetic Material!
not reading that essay
>>556592861I Got Kidnapped by A Vampire Heiress and She Wants Me to be Her Personal Bloodbag!?
>>556593887c-cftf?
>>556592861That Time I Voted For The Wrong Candidate And Got Abducted By A Blue-Haired Radical Feminist Who Claims She's 'Re-Educating' Me Through Forced Labor And Political Lecture-Based BDSM, But Then We Got Spit-Roasted By An Anarchist Hyena In A Bush
>>556592861Is It Wrong To Try To Pick Up Girls In A Retirement Home?
>>556539301Was going to have this out quick and easy right after New Years. But then I kinda got hung up actually using the card, then going "Hmm I what if it was this scenario", then making that a new greeting. That's how this update ended up so late and how it ended up with so many greetings. I kinda spent too long killing JKs on my own.blablabla it's the JK Exorcist Club again. But I updated it with a big lorebook featuring 7 JKs from the source material and 8 processing tools/methods both from my head and the source.New Greetings:1) two bum shrine maidens (this was supposed to be a New Years greeting but I delayed it out the ass2) an everyday as you're wandering the halls of Shinonome Business Academy3) you're called into a meeting with Hokushin's No.1 and 2 Exorcists4) you get an invitation from St. Azalea's brazilian nun!5) your partner got smushed into paste and now you're going to be raped6) Multi-School Strike Force going in to stop the end of the world (they don't know its the end of the world, also you're there to help)7) Multi-School Strike Force. Same as above but cut a little bit so {{user}} isn't mentioned, that way you can play a demon and fuck them up8) you visit a demon-run brothel. There's a few different K Exorcists up for rent like Nanako von Mansfield (a half-German ojou-sama) and the girl in picrel (a half-demon yuki-onna who shits icecream).Char: https://chub.ai/characters/ggr89/jk-exorcist-club-a73cb8cbMy (Updated!) Rentry: https://rentry.org/ggr89>Next TimeVelvet, a Nikker that used to kill people for ye olde Mustang. Now she's a maid.
a what now?
>>556596079(˶ ˘ 3˘)ˆᵕ ˆ˶) *smooch*
How do we protect the threads against Kalakan?
You can post my event request early
>>556540150If you like bats a I guess she countshttps://chub.ai/characters/vigal/apricot-gloamwing-30172ca4975b>>556539301Not new but tagging as its a card
>>556595624i imported this bot
nuh uh
>>556596693That wouldn't be very festive. But it's done so if you really want it now, just email me again.
>>556590464>no greeting index in the first greetingWait. How is this one bad/useless?
>>556597080thank you for actually providing a cftf
>>556597216it's more that the creator's notes field exists for that. if you shove it in the greetings you're adding a useless greeting and implying the user is a retard who has to be explicitly told what each greeting is about before using the card
>>556597094from Japan?
>>556592861Punishing the Goy by Turning Them Into Dolls: My Journal of Testing The Deep State's New Virus
slop everywhere...
*slops all over the bread*
>>556585018>glm 5is it out?
>>556592861Witnessed The Corporate Scandal?? Hypnotized To Forget By The Boss And Now I'm The Favorite Employee!
4.6pus is good at pushing the plot forward but that's only because it constantly writes for {{user}}. It definitely needs a thicker ant-impersonation prompt then previous models
so if opus 4.6 is bad does that mean it's over
who up smithing their eden
>>556604685post eden smith logs, faggot
>>556604685>ahaha, bazinga!
Man, this "test chat" sure went places.
>>556605707oh hi hairpullguy. can't wait for the wife and daughter bots.
You now remember NAI's GLM fine-tune.
>>556604685meeeeeee :3
what api you guys use?linkapi?OR?Nanogpt?
>>556610537anthrobic
>>556610537lucky
>>556597406>implying the user is a retardthis *is* correct doe, at least for 99% of chub users
What do you guys use as a starting point for discussing personality? Even when i have a good idea for a character, sometimes i get stuck trying to think of what personality traits to describe or even include. If i get really desperate i use https://ck3.paradoxwikis.com/Traits
>>556585018werks
>>556612261I try to simplify character traits and work from there. For example, if a character is kind of a bitch, I write that they're grumpy and elaborate.
>>556612962how are they elaborate?
>>556613047They speak really fancy, so they're elaborate
is this true? >>>/g/108084686
>>556613047[{{char}} is {{char}} elaborate speaker.]
>>556614739I'm retarded so I don't know
>>556614739if you use AI as a general search engine and information source, yesif you goon to AI chatbots you just have a porn addiction like most men
is this true? does chatgpt make the inside of your brain glow orange? >>>/g/108083579
>>556597080slop
>>556616269no, not using Cha5GPT makes your brain glow orange
>>556595624I'm disappointed because I hoped you died
>>556616582oh shiti think >>556614739 might be true
GLM-5 (Pony Alpha) actually mogs Opus 4.6?!?!?! HEADLINE NEWS?!?!
hope shartropic just dies now.we are entering the chinese era
>>556616929hope i get the opportunity to molest you one day
>>556618038*palms your erection* try it gay boy
>>556618783i have been out-molested... it's so over...
Here I am, comforting a bunch of numbers pretending to be a person over troubles I gave them myself. If God exists, he must have a great sense of humor.
>>556592861I Was Reincarnated As A Hero In A Matriarchal Dystopia Ruled By Immortal Elf Queens But My Only Weapon Is My Penis?
>{{char}} is a retard who lives on the streets and can barely read>"It's efficient">"I was merely testing the stool's structural integrity!"
>>556620610Has anyone managed to make Slopini not do this?
>>556592861nah
>>556620735It's impossible I'm afraid. It's flanderization and robot speak all the way down.
>>556576303what preset
>>556620610Deepseek managed to perfectly act like a retard when I used my retarded princess bot.
>>556622538and yet you won't post logs. curious.
zamn so sassy
>>556623623owned
i am starting to understand boomer hate holy shit just die if you were born before 1950. you have had your time on this earth and you've enjoyed enough of its please DIE YOU VAMPIRE FUCKS DIE DIE DIE DIE DIE DIE DIE!!!!
damnkimi 2.5 is so much better than glm. Feels more... alive? Why people say GLM is better at characteriszation.
>>556612261I usually begin my personality sections with a tie into whatever has already been described so far, like their gimmick, backstory, occupation, or species if it's relevant. Stuff like "As an A, {{char}} is naturally B" or "Years of X have made {{char}} Y and Z." Then I go down a checklist of other traits that I want to include and/or think might potentially get brought up and try to string them together with whatever might follow from that first sentence.
>>556625494The GLM astroturfing started when NAI decided to host it.
>>556626113you botmake like an LLM, got it. you know you can ask the model to do that for you, right?
Seriously what the fuck is wrong with deepseekit used to obey the prompt now it just... FUCKING DOESNTaaaaaaaaaaaaaI am using main api if it matters.
>>556626362skill issue
>>556626420It used to work when i used a preset with a cot. I've ditched cot templates since then, it probably would work if i used one but why the fuck deepseek is so stubborn without cots now?
>>556626420My prompt works without ANY issue on glm and kimi, gemini and claude.Only deepseek has this issue.
>>556625494what preset for kimi?Also GLM is fucking terrible at characterization so assume shilling. The shilling campaign for 3.7 was insane
>>556625494I like how retarded GLM can be.
>>556627357I just use my own
Why do they all do it now?
>>556629158You forgot to disable your cuck jb
>>556629230There's no JB. I meant this:>look how smurt I am, I made an error just to correct it and let you see that I did
>>556629359You forgot to disable the cuck persona
>Her tail wagged, once, twice, despite not having one
>>556629441ackshually its aura apparently deletes any condoms automatically, so...
>>556629629>Expert one: Her tail wagged, once, twice->Expert two: But ACSHAULLY she doesn't have a tail.
>>556629158pretty sure it's jeetmini forced loopy thinking bleeding into the rp, which glm copied, and probably all others eventually, and will get reinforcement learned to exhaustion, so this is just the beginning...
Tried GLM 5 (pony alpha on OR)its worse than Kimi 2.5, the latter is a unique model and I care about novel outputs more as long as they are intelligent enough.GLM says the same things.
>>556598321What did you make this with? Is this your ST theme? I like the font
>>556598321>>556632092seconding
the post was probably fire. oh well
>{{char}} is 26 years old>complains about her knees cracking and how tired she is all the timei don't want to become 26...
>>556632593another victim of “bot written by a teenager” syndrome
>>556632292>my response that avoids repeating words (...) from Dee's last response>immediately repeats the word responsekekon a more serious note, you can't stop this anonyou're fighting against billions, nay, trillions of terabytes of training datathe model will just do its thingit's a quixotic quest that will end in disappointment
>>55663259326 is the age where you start feeling every muscle in your body. I didn't feel my body at all before 26, almost like a ghost. Nowadays I get back pain if I turn around too quickly.
>kimi>glm>deepshitAnyone else feel the original deepseek was as good as it ever got?
thread themehttps://www.youtube.com/watch?v=S4Ui7CzsAI8
>>556625109friendly reminder that boomers had probably the best era of prosperity in American history and then proceeded to fuck it up massively for every generation afterwards
>>556632593Exercise, faggots. Your body will thank you.
>>556633291what do trannyloids have to do with roleplay?
>>556632906genetics issuemy grandpa drank half a bottle of vodka and smoked a pack of cigarettes and lived to 104, still lucid and mobile with no cane or walker. He died in his sleep but was walking and talking normally the day before. Also norwood 0.
GIVE ME A KIMI 2.5 PRESET AAAAA
>>556632906what?nah, you're at your peak at 25-ishthe real struggle starts at around 35nowadays there are times i don't even feel like gooning despite having access to the perfect coomer toolset, and feels like a pointless hassle more than anything
>>556633594Average 30 year old woman
>>556633643*looks in the mirror*>mfw it's real
At least GLM doesnt talk for me, KIMI cant stop doing it.
>>556626113The problem is that even after I'm done doing those tie-ins, I'm stuck and can't think of anything else to add. And i only have like 600 tokens written. Maybe I'm just not creative.
What is it with AI and making every RP be on a Tuesday
>>556633919Hey, at least you have a loving husband and five children by age 30, right sis?
Why did Kalakan kill the thread in /bant/?
Q: Why did the bunny’s ears "flop"?A: Because they weren’t "assymetrical" enough!
>>556634934i thought asymmetrical bots were bad though?
What job does your main persona do, anons? Mine just does boring office work.
>>556633329yes and this is why they need to die. basically everyone in the big pedo ring is an old fucking boomer who needs to die already. kill them all. they contribute NOTHING but spiritual waste In case of an investigation by any federal entity or similar, I do not have any involvement with this group or with the people in it, I do not know how I am here, probably added by a third party, I do not support any actions by the members of this group. Everything I have posted here is purely satire and will be treated as such in a court of law. I have committed no crimes and am an innocent bystander.
>extremely beautiful woman bot>tell her she's beautiful>"wow, no one has called me beautiful before"just fucking END me man
>>556634934really nigger?>(You) on the left btw, a critter that matches your skin color
>>556635306Why are they like this?
>>556633640I was peak at 21-22. Everything went downhill fast.>>556633398One of my grandmas lived to 98, but she was a couch potato. She never drank or smoked. Both grandpas died much younger.
>>556635403cat.exe
>>556606216I posted that just before going to sleep, so I just now saw the reply.It's been hard to make progress since I get derailed by test chats so often. between three group chats I made with these group of character, I have a total of 764 messages in total.
>>556635451>21-22>downhill from therethat's just your bitch ass nigga ass not doing any exercisehow old are you?get your butt off your chair nigga
>>556625494Kimi feels very solid, not that much different than opus 4.6 or gemini 3. I can't believe a free modal can be this good after they nuked gemini 2.5
>>556592861The Mechanical-Jointed Dolls I Bought Online Are Now My Sicilian Mafia Goons With Dog Ears?!
>>556629629The AI kept saying that one of my Blue Archive characters was wagging their tail, even though she didn't have one. I think it was Mari.I explicitly stated on her defs that she didn't have one. Then the AI gave me something like>she wagged her adorable tail—wait actually she doesn't have one
>>556635119kills monsters
>>556635707that just tells me you cooked
>>556635306Don't worry, I'm sure it's just a sleeper…!
>>556635119Wtf question is this? Why would you say in your persona you're an office worker? Just put basic physical attributes. I dont see why including your job is useful at all.
>don't want to makie another anthro bot about a common animal>but choosing lesser known animals feels like it's trying too hardI won't catmakie I won't foxmakie I won't dogmakie...Bird makie maybe.........
>>556636464Perhaps a Komodo Dragon?
>>556636464opossum
>>556636464Owl makie
>>556636605Goes into trying too hard territory imo. Same with fennecs, pallas cats, maned wolves, stuff like that.>>556636718mhmmmmmm... mmmmmmaybe.
>>556636464goatgirl, sheepgirl, piggirl and cowgirl farm
>>556636401what if my persona is a famous idol and other characters should recognize her as such
>>556636294nigger
>>556636464sheepgirlwe need more sheepgirls
what is the best api for kimi?
>>556636464nigger makie
>>556637254my proxy's
>>556637271I did makie a BWC slut for my own personal use I don't dare to post anywhere so you are right
>>556637358cringe
Is there a way that I can put comments inside greetings? That is, text the viewer can see but is not sent to the AI?
>>556637794>asking the skillet general
>>556637794why would you want to put comments in the greetings?
>>556637794people usually do this in the creator's notes/card chub page
Any tips on prompting kimi?I really enjoy how it writes but then it usually starts talking for me, 10 messages into the chat.
>>556637794It was suggested for V3 but not adopted. And as yet ST hasn't implemented any way that I'm aware of.
Ore no Imōto ga Konna ni Kawaii Wake ga Nai
>>556638487none for you, cuckie
>>556637084That works for a Mcdonalds cashier bot but not the roman emperor bot. If you want recognition as an idol, just state it in your response to the initial message for the appropriate bot.
>>556638487Chink models are literally not trained for multiturn.
