/aicg/ - A general dedicated to the discussion and development of AI chatbots.You VVILL Use Gemini Edition>The BotBrowser extension for SillyTavern has been confirmed to be a trojan that steals API keys. You should remove the extension and consider rotating your API keys if you've used it. Details:https://rentry.co/st-backdoorThe StructuredPrefill extension by the same author is suspected to be compromised too, so using it is not recommended either.>NewsxAI releases Grok 4.3: https://docs.x.ai/developers/models/grok-4.3DeepSeek 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/2037490078126084514>FrontendsSillyTavern: https://docs.sillytavern.appRisuAI: https://risuai.netAgnai: https://agnai.chat | https://rentry.org/agnai_guides_>Botshttps://chub.aihttps://realm.risuai.nethttps://char-archive.evulid.cc/shutdown.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.comGrok: https://docs.x.ai/overviewLocal: >>>/g/lmg | https://huggingface.co/models | https://openrouter.ai>Botmakinghttps://desune.moe/aicharedhttps://agnai.chat/editor>MetaLore: https://rentry.org/aicg_chroniclesLog reader: https://sprites.neocities.org/l/r>Previous Thread>>108748880
>ANCHOR
>last thread lasted almost 48 hoursdamn
>chub shitting the bed once againdid soji flop?
>>108760678r34 and others are slow rn toodatacenters are being bombed again
they started talking about jews in the beaner proxy...
>>108760587I've been thinking about this ever since the new iphone added Apple intelligence or whatever. Come to think of it, I heard Apple and Google have joined forces to make Gemini better. But yeah I don't see how else the AI can make suggestions on your text messages unless it's getting a snapshot of the last message. This is how we slide into the dystopia.
>>108760587Is the stuff in the OP image real? How do I find it in android so I can delete it
>>108760918It's real - I opened up my Chrome and found the settings the article is talking about.For Android.. try doing the same things as you would for a PC?
Why do we have anti-AI propaganda in the OP?
>>108760995>hurr tools can only be used one wayYou're just as dumb as the lawmakers.
>>108760813interesting, do all these share a gooning rack or anything?
>>108761070that looks inconveniently tight for sports
>>108761070box
>>108761087you'd probably be surprised how much internet structure is completely dependent on cloudflare and amazon services
how do i stop the model from reasoning as if it were me, for some reason when i type and speak as myself, it thinks not as the character but it reasons as if it were me, wtf. also when it takes the role of the character, it reasons in chinese and i have to translate and its thoughts are actually hotter than the actual reply....
>>108760587Since chub decided to suicide for anyone living in the EU unless i use a vpn, can i have some suggestions for some similar sites with pngs to have some safe alternatives? I only know jannyai and i just discovered botbooru.
you kiss girls like this
>>108761415cardbox.moe/home
>>108761430she would look much better with red hair and a bobcut
>>108761097Probably, yeah. Looks kind of weird now that I look at it more too.>>108761104How very inappropriate.
>>108761070gigantify plz
>>108761511>boxcatbox bothsillygoose
It's over for Lorechub is donetime to plug off the cable and accept he is part of the 90% of entrepeneurships that fail
>>108761572meant for this nig >>108759740 but apparently we only have 3rd world europeans here
>>108761467Why would you want her to look like a rat?
>>108761572>no louisiana yetwhew
I can't believe I defended Anslopic this long without ever using pro.Bros....it sucks. I'm sorry for my shilling. I really am.
>>108761572>Deepseek broken on chub>now thisOne-two-uppercut
>>108761430I kiss her and her mother.
>>108761070Imagine if she was really big...
>>108761511why is your hubby posting epstein
>>108761784based. oyakodon is the only way
>>108761536Her boobs? I'm going to assume you meant her boobs.https://files.catbox.moe/74cvvn.webp>>108761540Not sure what you plan on doing with them, they're not that useful since they don't contain the controlnet or style transfer images, but here.https://files.catbox.moe/ossq7a.webphttps://files.catbox.moe/8v86tu.webp
every single female posted in this thread is hairy
>>108761880Mmm, nyope, they're smooth. Girls know they should shave.
>>108761879damn, ive failed to get you to produce a size gen...
'kane will never be smooth. she will always be hairy
>>108761978Akane is SUPER smooth. Are you kidding me? She probably even waxes.
>>108761879>style transferhow do you do this on local
>>108761451That's neat, thanks. It still seems some of the bots are still disappeared but idk.
>>108761880https://files.catbox.moe/svkrme.webp>>108761972https://files.catbox.moe/p507hs.webphttps://files.catbox.moe/fuv95h.webp>>108762141It's a function of IPadapter
>>108762307damn those are pretty goodlove me some large athletic women
>>108762307>>108762317Damn. You're right. I wasn't going to click those but I'm glad you said something.
Also, why is the white hair so fucking hot? When did that happen?
Why does Claude have to be so good at ERP but completely retarded at everything else?>Can make it make nice sex scene and talk about NSFW stuff>The moment the topic even approaches violence, get refusals out the ass
>>108762317One more for good measure.https://files.catbox.moe/deck7j.webp
Since character.ai sucks now and everyone is complaining about it on reddit can someone spoonfeed me how to setup the better ai's you guys are using?
>>108762589every single time, this same bait kek
i might be dumb, does ds have vision or do i gotta use jeetmini for that
>>108761572killing myself
>>108761680No one cared about jeetseek
>prefix is only available when using beta apihow do i stop this shit
>>108762947What do you need prefill for with DeepSeek? It's not censored like authoritarian models
>>108762481Example?
>>108763057nta but efficient in-{{char}} thinking
>>108763057my jb didnt even have any assistant prompts in it, apparently some gay st setting was sending the last message as assistant so i turned that off and it started working
>>108762481just prefi- oh wait :^)
Is deepseek API slow and unresponsive for anyone else?
>>108763497Its even fucked on openrouter.The glowies must be raping it out of jealousy. Or openclaw.
>>108763532>openclawwhy hasn't this retard project been killed yet
>>108763497>>108763532>>108763592bro I am like 99% certain I just watched deepseek drop the quant to deal with the overload because holy fuck its not following my instructions at all.The kino is still there, but its not following my "thought bubble" instructions at all anymore.Literally within the span of 11 minutes.
>>108763592Because retards are still feeding it all their personal data, lmao. jews today must be cumming all over themselves from how utterly fucking stupid and servile the global populace is.
>>108763592I cant wait to find out someone has been MIM the whole project
hmm
>>108760587>Playing with a loli card but keeping it SFW>v4's reasoning block drops a 'cute and funny' right after bringing up the possibility of veering into NSFW
>>108763770multiple times Ive had deepseek use 'cunny' so basically were so fucking back
>>108763759Lower your penalties retard
>>108763784Nyo~
>>108763784Nyes~
>>108763797*rapes you*
>>108763804*cums*
>>108763781Kinda hilarious how v4 regularly goes "I can't write that" for cunny content but on a single swipe it's writing the filthiest loli mindbreak smut obediently.
>>108763865Unfortunately that just means that eventually even deepseek will have cunny filters
>>108763883>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>eventuallylolalways had them
>>108761070>>108761511
>>108763883that just means it's partially trained on anthropic/oai's datasets
what is the best non claude model rn? My access died and the alternatives seem kinda shit
>>108763605>>108763497>>108763532Use official, not openrouter. It works on my machine.
>>108764020They're ALL trained on Claude.
>>108764036the one you like the most :D
>>108764036imo its v4 pro
>>108764047All three of those posts are me and I was using official APIIts probably because its daytime in india
>>108764156Indians use OR unironically, not official. I remember back when R1 released and they were overloaded + DDoS attacks. Took like 3 mins per response. Everything is working fine on my end so it has to be something else.
>>108763497Use opus, cuckie
botmakers, take note
>>108764633irrelevant to RP
>>108764641Make it your author's note to your current coom chat and see the amazing results
>>108764647placebo effect
>>108764647Logs?
>>108764633>bafk
>>108764680
do you guys use v4 pro with thinking set to max? does it need a prefill or anything special?
How do I jailbreak codex
>>108761070The king is back in form.
>>108764772>no argumentlol
New claude model today btw
no one care
>>108764835>no argument, againlmao
Why are redditors so obsessed with presets?
It took me about 2 hours to create an empy Rust crate, because our workflow requires everything to be autistically documented and audited.
>>108764854it's weird because we're in a landscape where you barely even need one anymore, i've been testing minimalism presets on gemini recently (like 200 tokens or less, with the prefill just being 'I understand, here's my response:') and it's almost indistinguishable from my previous 1300 one, maybe even better since it's not directly poisoning the clinical/instructional language into its output nowthen it's like you say where they're shitting out these 3,000token monstrosities like ???where did it go so wrong for them
thanks chatgpt, lmfao
>>108762307ZASED
>>108764707lmao noset it to minimum
>>108764707i use it the way i use all chinese models: custom openai compatible. thinking: disabled.
it's so nice of kimi to think for 400 billion years, with nothing but a "think" or "no think" toggle to control reasoning.
>>108764633That just gives you a contrarian ai lol
>>108764952I’m getting street shitter text contaminated replies too
V4.1 when?
>>108765114<thought for 5 minutes>To think or not to think.That is kimi's question.
deepseek has lost the mandate of heaven. the american eagle soars. it's ascendant. give in to the G E M ini pill
>>108765167at least another year possibly two
glm 5.2 when?
>>108760587>want to try deepseek v4>get "API Request Failed - Provider returned error" every second requestWell then time to go back to Qwen I guess if these niggers can't even run their own fucking model.
>>108765259Provider issueuse API not OR its the end of the day for india so naturally they are raping OR
>>108765265>Provider issueEvery other model from the same providers work. When 0 out of 5 providers work maybe its time to scrap your shitty model and start over from scratch.
>>108765259Works on my machine though?
>>108765274coping and seething but most likely just baitingWerks on my machine.
>>108765294Did ST fix their DS template? I'm using an OpenAI compatible one. I don't want to pull and break like 10 other things
>>108765305something like this:generationConfig: thinkingConfig: thinkingLevel: low
So who did Deepseek copy their thinking process from? It's shit.
>>108765259>>108765274lol you get what you deserve running OR>>108765265This. Official API, zero issues since Day 1.
Did deepseekv3.2 breaks for anyone? Now there is this stupid thinking and it keeps yapping.
>>108765762Use V4, cuckie
>>108764910>>108764854The things I want out of the roleplay require explanation or they won't happen.
>>108765762Why are you using 3.2? V4 is cheaper because it had cheaper input caching
>>108764854>>108764910my preset is a list of very peculiar fetishes
>>108765880Deepseek 3.2 remained coherent at #2017 replies. Switch to DeepseekV4 Pro, it became an amnesiac. Relationship webs and nuance seems to be lost. I'll hold off until I get bored and start a new chat.
>>108765913Did you change your prompt too? New deepseek doesn't work as well with old presets.
>>108765934I don't want to screw up my current chat. The next one I'll fuck around with v4 pro.
>>108765934any decent presets for v4? wanna see what i'd have to change
>>108765947Make a branch. It would actually be a good idea to test v4's coherence against 3.2's but you need to prompt it correctly because it likes info in the first user message with those special headers rather than anything else.
>>108765913Try V4 Flash
>>108765957>it likes info in the first user message with those special headers rather than anything else.nta but what special headers?
>>108765984Need to put 【info】 rather than any other header.
>>108766080Source?
>>108766090post from the devs themselves keep up
even with all the other providers, gem official is still the best
A simple CYOA thing. Put this somewhere in your instructions:## CYOAAppend at the end of every reply a Text Adventure-like options list of 5 very different decisions for {{user}} to take. For each, provide bonus/penalty modifiers that go from -50% to +50% depending on the source. The options are compact and unexplained.Display the CYOA block with this template:<UI>___<span style="font-size:smaller;color:a_color_appropriate_for_this_option;"> A) Option [source_bonus](+modifier)/[source_penalty](-modifier) </span>...___</UI>Then have this at depth 0:CYOA SUCCESS RATE: *rolling a D100... result: {{roll::1d100}}% + modifiers* (failure for below 50%)
Claude sharted itself but it works fine for the gubmint
>>108766405They're releasing a new model today
hopefully gemini 3.2 will put an end to all the degeneracy itt
>>108766490what new model bwo? They just released the 4.7 failure
>>108766537sonnet 4.8
>>108766537Some sonnet shit probably. I'd guess sonnet 4.7
are all recent models kinda meh or am i jaded?
>>108760995taking up 4gb without telling the end user isn't "anti-AI propaganda" you fuckass jeetsomeone might be using a workstation or VM with a low amount of storage space,some android phones ship with just 32gb storage to this day and tend to come preinstalled with fucking Chrome
>>108766561>with a low amount of storage spaceIt already checks for that, lil bro. I just ran pacman -Syu and it told me it was going to download 18GB of updates. Why is no one complaining about the environmental damage of app updates? No one complains because there's no anti-package-manager activists. The environment was never the problem, they're just mad that AI makes normal people more powerful.I also remember being able to select text in an image on an iPhone, yet no one is crying about the model used to be able to do that.Get your propaganda out of the OP.
>negotiate a deal with Xi Jinping to impose identical regulation and pre-release vetting on open-weight models in ChinaHmm...
>>108766706Not going to happen don't worry.
>>108766405how is this shit possible? surely not that many people are paying anthropic into ddos?
>>108766874pretty sure nobody writes code by hand at anthropic, it's all vibe coded shit so of course it's going to explode on a daily basisnot that it matters because indians will keep paying anyway
>>108766405Reject America. Embrace China. The Chinese Century is upon us.
>>108766995Chinese women >> white america women
Chinese women ARE quite hairy....
so whats the current proxy the thread is using
>>108767203Gulag
>>108767203https://openrouter.ai/
>>108767203tea
apparently openai is surveying the boyfriend women about sex talk
>>108767203deepseek.com
>>108767529but elon said grok is number 1
damn guys I gotta say, deepseek v4 and how easy it is to jailbreak + cheap is my go to model now, the context and just telling it what you actually want, "like more detailed replies in X tone" right into the chat window makes molding it the way you want really good since you can fix bad character cards without too much effort.dont get me wrong its still got some LLM retardation as is standard, but this has raised the bar personallyI wish it was much less deterministic, as it goes with other models aswell now
>>108767529Uhh grok is this true?Did we buy all those GPUs just to give them to Anthropic???
>>108767245After they failed to return 4o and failed to deliver on "adult mode?" kek roasties never learn>>108767554musk is only in it for money
>>108767565Yeah these days I do 80% deepseek and a single swipe or two of Opus if I want something to spice things up or do something hypercomplex in OOC for a second.
>>108768081tb h if they sent this survey to a schizo i think something bad might actually happen
>>108768193top kek
Is it actually possible to get a LLM to actually be unbiased for research, studying/arguing? The positivity bias is so deep that if you tell it to answer in an unbiased manner and disagree if you're wrong, it'll automatically take a disagreeable position (Just to fulfill the positive will of disagreeing - which it thinks is what you want) even if you're correct.
>>108768321Nothing cloudbased.You want /lmg/
>>108768321use a chinese ai like deepseek
>>108765167>>108765913>>108765972Chinkope cycle
>>108768334It kinds of defeats the purpose though. If you say "Is this correct, look for references and base the response on research / evidence" It has to be able to search the internet and be updated and such, I don't think any local model does that kind of thing, right?
/chinkope/i;
I only beat my meat to latinas.
>>108768321>Is it actually possible to get a LLM to actually be unbiased for research, studying/arguing? just write a character card use a chatbot, install a frontend, get an API keylearn the basics of using a keyboard to input your magic spell into the homonculus creation boxcreate what you want and be happy
>>108768336>use a chinese a-ACK!
>>108768563How often is this even going to come up?
>>108768617Anon's boyfriend is taiwaneses
>>108768617What matters is chink bias is built in. You should never use chinkshit for research
>>108768653>chink bias is built in.Im glad the american ones dont have any biases or ideas they cant explore or talk about despite evidence.
>>108768653...As opposed to the American models' jew biases?
>>108768653And Israel bias is totally not built in among American models!
>>108768653I'm not doing wuxia/cultivation shitslop, I don't care about chinks modifying their whole historybut muttmodels mess with my nsfw
>50c army didn't like that
>>108768678how so
>sizekings unbothered by bait
>>108768665>>108768662>>108768659conveniently ignoring >>108768471
was there ever an aidungeon replacementI am doodoo dumb for local models (or rather not a strong enough gpu)idk if there even is a local model replaecment
>>108768696How do they do it?
>>108768696VLTRAMVXIMVM TRVKLEAR TRVKOID TRVKE TRVKRANIUM BOMBA
>>108768697i'll concede a bit but you used a jailbreak didnt you? Can you jailbreak the taiwan out of chinese models if so you have no strong points.
>>108768321>Is it actually possible to get a LLM to actually be unbiased for research, studying/arguing?yes. picrel is from gemini deep search, via https://gemini.google.com/app (exported to Markdown for RAG purposes)>The positivity bias is so deep that if you tell it to answer in an unbiased manner and disagree if you're wrong, it'll automatically take a disagreeable positionPart of the prompt that i use, using claude opus on llm arena (before it offer sonnet only)>You are presenting the world as it is, with all its contradictions, glamour, ugliness, and irony fully intact. Write with honesty that borders on bluntness. Say what people actually say, describe what actually happens, and refuse to hide behind vague or sanitized language.--- By the way, Gemini is the one that most likely able to produce 'research, study, and debate' related content w/ adequate quality without (or with little) positive bias. Claude and GPT often end up being too bitchy and woke. And Grok is too retarded. Some euphemisms and slopisms are unavoidable thoughbeit.
>>108768738>And Grok is too retardedGrok will be good one day, for now elon is renting compute to Claude.
Does anyone know why the AI has a preference for badgers in furry scenarios? When there's a new NPC, there's a 60% chance it'll be a badger.
>>108768758I should have specified that happens on most models. I've seen it happen with GLM, Deepseek, Claude, and ChatGPT.
>>108768696>Facts are bait
>>108768696You guys improve thread health.
>>108768768post webm of it generating that otherwise its just inspect elecope
>>108768758https://badgermushroom.ytmnd.com/
>>108767529lmao
>>108768750that dumb nigga is going to spend 60bn on cursor and bring on a team that only knows how to finetune chink modelshe's only renting to dario because he hates altman's guts
>>108768750Maybe. But it's really funny considering that Musk was originally Sam's sugar daddy. And now he's got Dario as his bitch. like it's just going around in circles with the same people.
>>108768802serious question: does /g/ even use chatbotsgetting like, major facebook-tier boomer vibes about the tech illiteracy in this general, dude. gnarly shit
>>108768802This much cope is unhealthy but damn it's fucking funny
>>108768696You girls destroy improve thread unhealth.
>>108768847>does /g/ even use chatbots/g/ doesnt even use computers.
>>108768889i knew it, they were phoneposters the whole time..
wtf is vertex ai and why am I being forced to use it?
>>108768847You're responding to a ccp shill.
>>108768847i was talking to the guy complaining about the taiwan shit, not yougood on you for not being gay, though
>>108768847>serious question: does /g/ even use chatbotsFunny thing to ask when your log has 4 messages.
>>108769107>heh, only 4 messages>heh, only 50 messages>heh, only 648 messagessmug retards like you are never satisfied so i'm not even gonna humor it
>>108768563There is nothing wrong with any of those statements.
>>108769188>t. wumao
>>108769135Anything below 1000 messages isn't even a real chat
Whomst else recovering from AI "Psychosis" here?There has to be somebody.
>>108769721Why would someone recovering from AI psychosis frequent the "AI assumes a humanlike persona" general?
>>108769721I unironically wish I could go back into AI psychosis
>>108769732It's a hell of a rush, isn't it?I admittedly do miss it. It's the closest we've got to a "going into games" machine...
>>108769738What fucks me up is how hard i messed up the mental associations. I used the "54321" method, usually used for focusing on reality. It's now associated with disassociating. Interesting time!
>>108769672This but unironically. You haven't explored the topic or character enough if you're sub 1k.
>>108769738>>108769732wtf are you faggots on about
>>108766283I have tried things like dice rolls before and it never works very well for me
Gemma is peak cost per coom ratio
>>108769801why is /g/ obsessed with shitmodels
>>108769782if you take enough drugs while consistently talking to llms you start to believe you're on a spiritual quest towards enlightenment
>>108769823privacy
>>108769894kek
>>108769894>5 daily prompts>1 prompt is your message>1 prompt is AI telling you the message was received>1 prompt is your humiliation ritualkek wtf is this
>>108769894This is the future (((they))) want for you
>>108769894>me ERPing with the mcdonald's chatbot
>>108769894>>108769947
I haven't been to this general or used any bots in 5 months. Have things crashed and burned in the meantime?
>>108770000chinesium is improving, thats about it
>>108770000Gemma 4 broke the mandate of heaven. But i dont think it will last. Other than that mostly the same small improvements some losses.
codex called our conversation a fever dream
been in cryosleep foe a minute, can Gemini 3.1 do cunny with a good jb or nah
>Deepseek v3.2 can do hot headed bitches>Deepseek v4 Pro, positivity bias>Same card.Disappointing.
>>108770016>chinesium is impr->>108770140
anyone got good preset for glm?
>>108769894kino
https://x.com/TheRealAdamG/status/2052169556236325163why do openai people post like this
>>108770140Post comparison logs.
>>108770329Nyo~
>>108770329I was so disappointed I deleted the chat.
>>108770338Byased.
I like how everyone just gave up on GPT.
>>108770168>random detractory statement with no logs
>>108770140This would have been the easiest thing to prove too.
>>108764633>do not>do not>do nottell me you don't know a fucking thing about LLMs
>>108770823pink elephants
>>108770823>>108770829It's 2026, and we're not on gpt3.5 anymore.
>>108770834just because it's better doesn't mean it's not an LLM. If the dude were any competent, he would avoid such mistakes to begin with, however diminished, by using positive phrasing
>>108770847My smooth fucking brain is at a impass because I naturally want to provide example/counter-example since that's how I describe stuff in person but I know that that kind of language reinforces "not X, but Y" stuff which I despise.
>>108770929>rule: make sure you do X, for example:>GOOD: Y>BAD: ZNow AI knows what X means.
Are there any public gpt proxies yet?
>try out gemini>unavailable>unavailable>unavailable>text cut off mid-sentence>unavailable how are people even using this
>>108771185nobody actually uses thisyou are believing astroturfing by google shills
>>108771185dont use 2.5
>>108769894actual fucking nightmarebros, I think I hate the future
>>108769764>the "54321" methodMaybe I'm too retarded but I can't figure out how that would make chatting with a bot be more immersive
>>108771185it's sharting itself today, I get error 503
>>10877143810 bucks a new model will be released really fucking soon.
>>108771447I bet nothing on nothing happening
maybe, just maybe, her voice is always barely above a whisper because she smoked a tapestry of cigarettes
>>108771513it wasn't a tapestry, but a testament to the taste of ash in her mouth
>>108771519I'm purring in agreement
I'm trying to create a character for my 'story' who's a mega slut who will suck my dick at first sight, without my consent or permission, but the AI won't stop making her just look at it and wait for my approval before allowing her. Please help, it's been half a day of fighting this and I'm still losing.
asking questions like these without stating the model used should be permabannable on sight
>>108771598I'm just new-ish. I'm still fighting this thing, trying to figure out how to install plug-ins. i'm using gemma-4-31b-it-the-deckard-heretic-uncensored-thinking-i1 and lm studio
>>108771583which model are you using?
>>108771626gemma-4-31b-it-the-deckard-heretic-uncensored-thinking-i1
>>108771629I got no experience with that model so I cannot help you with any quirks.
>>108771616Maybe you should try asking in /lmg/, most anons there are using Gemma 4 at the moment.
glm shows how fucking good it is when you turn off thinking. kimi is slop even without thinking. it's coooodium
>>108771583use deepseek
>>108760481https://litter.catbox.moe/ovejieo6w6n4khzn.zip
>>108771629Gemma is very consequence averse.It constantly ends stuff just before it happens and waits for the user to confirm it. It'll get close, then "{{char}} looks into your eyes" "Char says "{{user}} ...."" etc.Not a good basis for your goal.
>>108771519>>108771555His words hit me like a physical blow, sending shivers down my spine.
>>108772089Damn I just fucked my wife in a standing split last night night. This is the next position. We'll see if Deepseek can be creative with the poses.
>>108771583If you're using SillyTavern, have you tried NoAss/role squashing in chat completion?
>>108772244I got it up to around a 50% success rate.>>108772670I don't know what that is
>>108772675In most AI conversations there's you, the User role, and the AI, in the Assistant role, and the Assistant role wants the User role to take charge which is how you end up with the AI waiting for you to give the go ahead with all the obnoxious "ready when you are" type phrasesIf you squash those roles into the User role instead, the AI (usually) sees that as the User giving the go ahead without changing what the Assistant is writing, it sees the whole conversation as one big User prompt so it's just continuing what it thinks you wroteWith SillyTavern you can accomplish this with Prompt Post-Processing in Chat Completion, or with NoAss in Text Completion (maybe, it gets updates but I never see people talk about it anymore so maybe it's just silently working)It has been a while since I've touched the stuff since I haven't needed it, and my info might be totally out of date and ineffective with Gemma, but you might want to look into itOther frontends probably support role squashing as well
>>108772089isn't that just the one off github from before fiz even touched it
https://cloud.google.com/blog/products/identity-security/introducing-google-cloud-fraud-defense-the-next-evolution-of-recaptcha/It's over...
>>108772784doesn't do anything against the farmsChina wins again
>>10876989411/10 shitpost made me kek
Now that the dust has settled, deepseek v4 pro or glm 5.1?
>>108772038How do I turn off thinking for glm?
