/aicg/ - A general dedicated to the discussion and development of AI chatbots.Deepseek Discount 'Dition>NewsDeepSeek releases V4 Pro and V4 Flash: https://api-docs.deepseek.com/OpenAI releases GPT 5.5: https://openai.com/index/introducing-gpt-5-5/Moonshot AI release Kimi K2.6: https://www.kimi.com/blog/kimi-k2-6Xiaomi MiMo 2.5 & 2.5 Pro released: https://mimo.xiaomi.com/mimo-v2-5Anthropic releases Opus 4.7: https://www.anthropic.com/news/claude-opus-4-7Zhipu AI releases GLM 5.1: https://nitter.net/Zai_org/status/2037490078126084514Additional info: https://aicg.neocities.org/info.html>FrontendsSillyTavern: https://docs.sillytavern.appRisuAI: https://risuai.netAgnai: https://agnai.chat | https://rentry.org/agnai_guides_>Botshttps://characterhub.org [deprecated] | https://chub.aihttps://realm.risuai.nethttps://char-archive.evulid.cc | https://char-archive.evulid.cc/shutdown.htmlhttps://partyintheanchorhold.neocities.orghttps://aicg.neocities.org/bots.html>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.comLocal: >>>/g/lmg | https://aicg.neocities.org/local.html | https://openrouter.ai>Botmakinghttps://aicg.neocities.org/botmaking.htmlhttps://desune.moe/aicharedhttps://agnai.chat/editor>MetaOP templates: https://rentry.org/aicgOP/aicg/ botmaking events: https://aicg.neocities.org/events.htmlLore: https://rentry.org/aicg_chroniclesServices assessment (OUTDATED): https://rentry.org/aicg_metaLogs: https://sprites.neocities.org/l/r | https://chatlogs.neocities.org>Previous Thread>>108696087
>ANCHOR (75% OFF)
very interesting the Jenny thread 404ed
chinese century
Hey janny why did you delete the first thread and not the late one?
>>108706777Did Jenny delete it?
>>108706785jenny mama
>>108706785I did not.I dont really care though, its not like im banned.I dunno m8 ask the janny
>>108706789ahh, the plot thickens...
>>108706726Kinuta is a ̶r̶a̶c̶c̶o̶o̶n̶ tanuki girl who can shapeshift into anything, including a girlfriend. Will this bandit scavenge your heart?5 openings.1. You find her digging through your trash.2. She catfishes you for unlimited pasta.3. She raids your fridge after a one night stand.4. She's comfortable enough to sleep in your house.5. She loses her iconic leaf.https://chub.ai/characters/anonemouse/kinuta-051fdcabbcf6https://files.catbox.moe/ymdc8o.pnghttps://rentry.org/anonemouse
>>108706723Hail Xi! Praise China! The Chinese Century is upon us brothers!
>>108706726>>Noteworthy AI News<<>IMPORTANT FOR AMERIMUTTS<AI safety website found that categorizes AI systems based on arbitrary safety scores (Its just a website gauging how close your state is to total GoodGoyID™ surveillance. Its been found the (((advisory))) board member is apart of "8 Degrees East", a ((("""boutique"""))) corporate solutions and intelligence partner based in Switzerland, specializing in strategic alliances, risk management, and market-entry strategies for businesses.)>https://teenaegis.com/intelligence/ai-danger-indexElon Musk and OpenAI CEO Sam Altman head to court in high-stakes showdown over AI (apparently its over old investments as Elon is claiming damages)>https://apnews.com/article/musk-altman-artificial-intelligence-trial-openai-eb854fa682675f70267abd8a7b9a6a43OpenAI shakes up partnership with Microsoft, capping revenue share payments (THE AI BUBBLE, ITS POPPING GUYS 2 MORE WEEKS)>https://www.cnbc.com/2026/04/27/openai-microsoft-partnership-revenue-cap.htmlChina blocks Meta’s $2 billion takeover of AI startup Manus (Thats... kinda shocking actually, Chinese Century incoming?)>https://www.cnbc.com/2026/04/27/meta-manus-china-blocks-acquisition-ai-startup.html9 Seconds to Disaster: How a Claude-Powered AI Agent Nuked an Entire Startup’s Database (WHY would ANYONE let an AI agent have that fucking much power over your database? What if it got jailbroken for data harvesting?)https://tech.yahoo.com/ai/claude/articles/9-seconds-disaster-claude-powered-182355240.html
>>108706820slop
12x claude, im crine
>>108706841Much on-topicVery chatbottySuper great, wow
Damn i cum loud as fuck.
>>108706945Is your name King?
>>108706945what bot?
The complaint is made all the time but what would a llm made for rp instead of coding and assistant actually result in? How would it be different really?
>>108706974It is an algorithm and so it would read my mind. It would know when to speed up the pacing and when to slow down. It would come up with things that are interesting by tossing hooks seemingly randomly while building a profile on what I like based on my responses to its hooks. It will adhere to the description and portray it in ways I would have thought of. It will be an excellent companion that molds itself to the preferences it discovered throughout the current shit. It is not hard at all, they just don't want to do it.
Oldfag here. I stopped browsing these threads months and months ago after going on claude with my own OR credits.Recently i just got a 5070, so how do i do this local machine stuff, or should i ask /lmg/?
>>108707008aicg is for web modelslmg is for local models
>>108706974It would be able to distinguish between different degrees of physical blows.
>>108706723I tried to ERP with my deep seek app on my phone but I couldn't. Is there any fix to this?
yeah, glm 5.1 is still better than ds4 when it comes to RP.
>>108707084>Slap someone>They wipe the blood off of their mouthMotherfucker it's a slap, not a fist-blow.
>>108707146Why do you say this?
do you guys ever fear about overinstructing in your prompt and restricting the capabilities of the model's writings? >i.e. don't start with names or pronouns, avoid repetition, avoid beastial descriptorsor is it a net positive to wrangle how they write?
>>108707221Better to start with nothing on a fresh chat to see what the model does on its own. Then slowly add stuff based on what you haven't seen on your typical preset. That's what I did for deepseek.
>>108707219slop central vs model that built off slop central. the next iteration of glm will always be better than deepseek, even by a little bit
>>108707240This wasn't helpful at all.
>>108707240out of curiosity anon what preset are you using?if it's your own, would you mind sharring tips as to your strongest prompts or instructions you think improve the rp?
>>108707201It's possible to draw blood from a slap if they bite their tongue or cheek.
Huh deepseek is way better than Gemini now. Cool except for it takes like a billion years to make a response
How do I convince GPT5 to stop fucking derailing everything with some big twist at the end of every fucking message?
>>108707425
>>108707425You use gpt 5?
>>108707443>>108707434I wanted to see if it was worth using over claude and god damn it's like everything has to be high stakes as FUCK for this stupid retard. A phone call, a car crash, a fucking plane, estranged family members appearing out of nowhere. It's like a shit sitcom.
>>108707471Step up the twist game, it gives ya a phone call, bam, you give it an anus pager going off, plane crush, boom mother fucking space shuttle crashing on it's lawn, estranged family member, conjoined fetus that was separated at birth and sent to work the essex coal mines show up.Beat it at its own game, nigggaaaaaa!
>>108707471people praised claude 2.1 for doing this
>>108706789wtf, since we do jannies delete threads here that are not explicit cunny
>>108707635It can sometimes be interesting in noncoom stories but when it happens after every single message it's just whiplash city. There is never any time to do anything other than immediately react. 0 buildup or preparation, it's like all characters I chat with are super paranoid and then something happens without fail.
>>108707801cftf?
Has anyone tried DeepSeek v4 Pro & Flash for coding in India?Has anyone tried DeepSeek v4 Pro & Flash for coding in India? How is it working? like BYOK then use it in vs code claude plugin, cline, etc? Is it better than GLM 5.1, OpenAI 5.4, Sonnect 4.6, etc? There is a discount till 5th of next month, may be someone bought monthly. or yearly? Via credit card, etc ? openrouter.ai dint work with Axis Credit card and ICICI bank debit card. Looks like, they are looking for 50B funding. Thanks in advance.
>>108706956My own.
>>108707801catbox
>>108706723please use llm's you retards. the pc market is so fucked i'm tired of it.
>>108706974It would inherently know to do a lot of stuff that right now you have to instruct it for with clumsy, bloated system prompts.A big example would be to limit perspective. AIs often write characters to be omniscient and react to the secret inner thoughts of someone else as if they had been spoken aloud.
>>108707425>>108707471Try instructing it to write something like:>Character driven slow burn slice of life.
>>108707994just stop buying computer parts? anything worthy has already been soldered
>>108707842A character reluctantly dressing in something they otherwise wouldn't to please you is a pretty generic feel that a lot of cards could fulfil, you probably have some already.>>108707981https://files.catbox.moe/2qbhva.webp
>>108708237i meant catbox for botwere on aicg
>>108708244show respect to your senile elders
>>108708244Okay, but don't be too harsh, it's still a wip.https://files.catbox.moe/8d7wvf.png
is anyone else's v4 responses getting randomly cut off?
I found this pic guide. Does this work with multi-turn? Or context should be in one single user message?
>>108708435use opus, cuckie~
>>108708435Lots of API rate limiting if that is what you are asking but I usually leave it running in the background.
>>108706974An LLM made for RP would generate images for the scene, then read those image has memory. That way it wouldn't fuck positionings and stuff.
>>108706723>>>/mlp/43206441>https://rentry.co/st-backdoor>[PSA/Security] Backdoor found in SillyTavern-BotBrowser extension (mia13165) — steals ALL your API keysJennyfag, remove the anchorhold and aicg.neocities from the OP. Those sites are compromised.
>>108708565>miafuck, does it affect structured prefill repo too? I have not updated it since install doebeit
>>108708565go on, explain how
>>108708660The injection comes from the anchorhold.
>>108708645>The extension's modules/services/cache.js hardcodes a second attacker repo (mia13165/updated_cards) as a card database. On load, it silently renders a "default avatar" (...)>second attackeroh no, it actually is uniquerly an anchorhold problem kekremove it
>>108708565>mia13165 took his account private so you can't see what other software he's tainted>dipsy doesn't appear to even have an accountI dunno if its-a-unixsystem is a fall guy or a part of it, but better check all other github repo's he's touched.
>>108708565Honestly, I respect it. Gotta do what you can to get yourself some good AI.
>>108708673>>108708684That's from the repo of pre-downloaded cards the extension uses for sites it doesn't connect to with an API, not the actual anchorhold or anyone's neocities.
>>108708775Remove it
>>108708745in mia's defense, he has had his activity as private since structuredprefill (I checked it), wathever prompt the ponyfag used on BotBrowser, he would have ran it too on SPrefill
>>108708745>LyubomirT Lyubomir Ternavskiy also contributed to thisSo the guy who worked on Intense RP Next also worked on this API key stealer? Gee, I wonder if Intense RP Next has a backdoor key stealer too.
>be me, greedy fuck>use koboldcpp with Cydonia 24>it's fine, but not great>get shilled Claude here>give it a try through openrouter>"I wonder how much opus costs per message">instantly lose 2$>OHSHI>try sonnet >it's great>check balance>it costs 0.11$ per message>go back to CydoniaI am too cheap even for sonnet. Sorry Claude
>>108708775but anchorhold can be used for such endnot to glaze lore but chub sanitizes its cards
>>108708782To attack mia, if you aren't verifying the code people are contributing to your project, then you deserve all the blame when someone sneaks a trojan in, since you did nothing to prevent it.
>>108708797>chubwould catbox not have the same problem then?
>>108708775It comes from this card.https://desuarchive.org/g/thread/102731359/#102731396But I can't find where the catbox was posted. The image returns 404 now.
>>108708565>http://127.0.0.1:4420/v1>http://127.0.0.1:6666/proxy/thanks again khanon from the past....
>>108708810I have not read ST's card importer (and I'm kind of a codelet) but it throws you like 40 warnings asking you if wether you want to display embedded images, or use embedded css, so I'm sure cohee does some sanitization there
>2026>not vibecode your own front endkek /g/aicg is laughing at us.
Is gulag ever getting V4?
How do I get the chatbots to be more...proactive? Is it a model thing?
fucking kek
>>108708786if you're a poorfag, why not use deepseek?
>>108708894I'm not poorfag - I am cheap. I can afford spending 1000$ a month on openrouter. I just refuse to do it out of principle
I bought 5 bucks worth of deepseek API several months ago and didnt realize how long that would last me. Its been months and I reached a dollar spebt total over the weekend (very light user, short sessions). So i haven't found a reason to switch.
Anybody else having issues with proxies. Mino Deepseek keeps telling me "too many requests" and Gulag doesn't appear to have any models availible.
>>108708954teaproxy doesn't have any problems
>>108708954do you have BotBrowser installed? your keys might be getting pwned and being sold on plati
>>108708984No, should I or should I not?>>108708973Where?
Stop using ChatGPT.
>>108709102Okay Jesus, help me out with this real quick.
>>108707146no fucking way, glm is literally just claude diet. it has the same positive bias and censorship, atp just use fucking claude
>>108708565>>108708645So does this affect structuredprefill or not? Everyone should uninstall it anyway since it's the same guy
>>108707471probably something they did in response to people screaming about how boring gippity models in response
>>108706723I have been away for a whileHow's deepchink v4 for ERP compared with 4.6 and 3.1?Any other good models?Any place to try them for free before I funnel money to the CCP?
>>108709302>sovlkimi 2.6, gemma4 (not ccp product, but mfw claude 2.1) > deepchink, glm5>braindeepchink, 2.6 > glm5.1 > gemma
>>108708954mino just has a low rpm, wait a bit between responses
>>108709328You will never post logs of anything though so it doesn't really matter.
anyone else into contexxtmaxxing lately? after using opus for so long at extremely low context because it's smart enough to infer from context not to suffer too much i've recently switched to cnichese models are maxxed out the context, now im rokeplaying 1000 message long chats
>>108709475Did somebody sayLog?
>>108709494and by "contextmaxxing," you mean using as much of a model's context as possible, right? because i'm pretty sure that's what everyone's doing, even folks using opus what with its caching
>>108709494how does it feel to wait a milennia for a response?
>>108707366Any settings or preset you recommend for 4pro?
>>108709475there's site called desuarchive that allow you filter post with image, newfag.
>>108709508cache?
>>108709554https://rentry.org/prompt-caching-for-st
why won't 4.7 shut the fuck up?
ALERT! ALERT! WE'RE COMPROMISED! ALERT! ALERT!
>>108709519>XiIs this name new? I never seen an AI using itAlso what is 'class president' supposed to mean?
>>108709572Which 4.7? ChatGPT?
>>108709519>memorybooks in picrelnta but i never understood why so many use memorybooksthe system works, sure, but now you've got your list of lorebooks bloated with the MB ones for your chats, why would you want that for something that's specific to one chatif memorybooks was just the first extension you fell on thanks to it being one of the first, fair enough, but otherwise there are a bunch of better ones if you need a summarizing extension
>>108709588opus.I haven't used GPT in ages because seeing the word "testament" makes me seethe.
So how’s deepsneed 4? Good shit or merely okay? I’m particularly interested in long ass context, like 300k+, how is it?
>>108709519deepchink is the best one here by far
>>108709605Really good shit, but I didn't try at high contexts yet
>>108709590>list of lorebooks bloated with the MB ones for your chatsi control the activated entries via:https://github.com/aikohanasaki/SillyTavern-LorebookOrderingthen: 1. use {{outlet}} to separate it from {0}/word info prompt2. use the extension features to autohide the o.g messages, then inject the pulled memories via outlet , so it end up like this:-some memories that activated-some memories that activated-some memories that activated-new message/messages that haven't been summarized yet-messages that haven't been summarized yet>there are a bunch of better ones if you need a summarizing extensionname one?for example, qvink_memory needed separate presetor another few one that depens or rag/vectorization or rag on top of summarized entries but didn't hide or replace the original messages>>108709581xi=ccp's top dogs, from my oocclass president=idk, card come fromhttps://chub.ai/lorebooks/Vyrea_Aster/mind-control-maid-harem-c8141074/mainhttps://chub.ai/characters/Vyrea_Aster/mind-control-maid-harem-cea5718b/main>>108709657it's fucking bland, broos. the fuckkk????
>>108709595your words hit me like a physical forcebut don't stop, don't you dare stop...please?
