/aicg/ - A general dedicated to the discussion and development of AI chatbotsMontefio Edition>NewsOAI releases GPT-6.1 Sol https://openai.com/index/introducing-gpt-6-1-sol/Anthropic releases Sonnet 5.5 - https://www.anthropic.com/claude-sonnet-5-5Anthropic releases Opus 5.5 - https://www.anthropic.com/claude-opus-5-5Xiaomi releases MiMo 2.6 Pro and Flash - https://mimo.xiaomi.com/mimo-v2-6OAI releases GPT-6 Sol and Luna - https://openai.com/index/introducing-gpt-6-sol-and-luna/DeepSeek releases V4.1 Flash - https://www.deepseek.com/en/news/deepseek-v4-1-flash/OpenAI releases GPT-6 Astra - https://openai.com/index/gpt-6-astra/>FrontendsSillyTavern: https://docs.sillytavern.appRisuAI: https://risuai.netAgnai: https://agnai.chat | https://rentry.org/agnai_guides_>Botshttps://chub.aihttps://realm.risuai.nethttps://char-archive.evulid.cc/shutdown.htmlhttps://chararc.bernkastel.pictures/https://archive.cardbox.moe/>ModelsJailbreaks: https://rentry.org/jb-listingGPT: https://platform.openai.com/docsClaude: https://docs.anthropic.com | https://rentry.org/how2claudeGemini: https://ai.google.dev/docs | https://rentry.org/gemini-qrDeepSeek: https://api-docs.deepseek.comGrok: https://docs.x.ai/overviewLocal: >>>/g/lmg | https://hf.co/models | https://openrouter.ai>MetaLore: https://rentry.org/aicg_chroniclesLog reader: https://sprites.neocities.org/l/r>Previous Thread: >>109996728
>ANCHOR
>>110011033based ZZZ chad! Thanks for reminding me to do my dailies
>>110011033>>110011041I cum inside girls like this
pomidor bros... I think the tomato is rotting...
>>110011148he hasn't had good keys in months son
>>110010507https://github.com/Pasta-Devs/Marinara-Engine/compare/v2.4.6...v2.5.0Reading the diffs for these jeet vibe code slopfests is always fun. 300k (THREE HUNDRED THOUSAND) lines of code added in a single minor release.
Say I have no money constraints for RP and want the best goon model on openrouter, what model do I pick?
>>110011252fable
>>110011252opus 4.1. you have like 2 months before it disappears forever.
>>110011258nta but why is that one better than the other opus options?
>>110011258And then have Opus withdrawal? pls no, I want something else that will stay for at least a year or so
>>110011264easier to jb
>>110011033>>110011041>>110011244ZZZ Lucy is so hot
proxy where side note what kind of service doesn't allow you to delete characters? fuck off cAI
im surpised my glm and deepseek source has never been posted here
>>110011484I'm surprised your shapely feet and also soles have never been posted here
>>110011494i dont think the nonnies here like 3dpd feet
>>110011515they're lost soulshaha homophonic foot pun
where the fuck are the models all saved?chub.ai and characterhub.org used to fucking WORK. then some time some stupid update to the site came around and suddenly it doesn't work anymore. Tags don't work, the search function doesn't work. What the fuck happened and where did everyone move? tons of content I used to browse just basically broke completely and never came back.
>>110011759complete global enshittification chrizz
Favorite bot that can always deliver a quick nut? Don't care if it's slop or not
>>110011910classic stuff like:https://chub.ai/characters/Vyrea_Aster/miiya-e25a67d2https://chub.ai/characters/Vyrea_Aster/yeon-ju-stockholm-syndrome-cb872343or justany wlw bot + manual edit for LGD scenariofemdom bot + domination loss
Does Marinara offer me anything over SillyTavern when I just use it to coom?
>>110011759A lot of stuff got geoblocked because of unwanted attention from payment processors and whoever else. This may be why stuff seems to be missing. I have to VPN to random places like Panama to make sure I'm seeing everything.https://www.reddit.com/r/Chub_AI/comments/1rch13j/geofences_subs/
>>110012053no
>>110012095>try vpn>it worksFUCK utah holy shit
Haiku any good bros?
>>110012648Nope.
Drought so bad I'm using janitor AI llm to get off
>>110012717Which model does it use?
>>110012728Opus
>>110012728I have no idea, they call it JLMM (real creative) and for free I have 9k context, good enough for a quick sesh, and it doesn't feel that bad to me, it's meant for RP so no refusals or tardwrangling, but it does write slop from time to time.
>>110012759Spicychat has like 4k context or something lol.
>Keys: no
does someone have the i use arch btw girl card? the link i saved is dead
>>110012934SOMEONE PLEASE GIVE THAT CARD
>>110012934https://chararc.bernkastel.pictures/chub/doombro/character/xenia-7f317b68this?
>>110012955was by toynbeeidea judging by the url and was arch personified
>>110012978Catboxed? Posting date?
>>110012053more clicking
>>110013031this one>>108979260
>>110013083Ewww she is obese
Are there any models out there that have completely different censorship rules for their web app compared to their api calls?
>>110013083No, don't have this.
>>110013083https://cardview.neocities.org/?https://pomf.space/LoXNDQ9t.pngthe cabal abides
>>110013141>no card for the cunny versionI thought this was aicg
>>110013095Yeah gemini
my name jeff dot jay peg
>>110013141thx!
/vg/aicg/ is STILL dead
>>110011033>Use GPT SO(u)L for translations>It's REALLY good but expensive>Decide to use DeepSeek 4 pro>It's alright but has some inconsistent context handling>Decide to just give Gemma 4 a shot since i've heard it's pretty good despite its low parameters >It actually does a great job, sometimes better than Gemini Flash>It just has a bland neutral English writing feel to its translationsI hope Gemma 5 fixes this problem, because it's probably the best all-rounder while being low enough to not be a money burner.
>>110013211Yeah? You guys asked for that, remember?
>>110013216I'm simply too autistic to not notice the weird quirks/shortcomings in really any model
>>110013217I am NOT one of the you guys nigga
>>110013268Dont call me that word ever again.
>>110012775>"message": "Reverse proxy error: No Deepseek keys available",
>>110013390Ok niggaaaa
>>110010412Like I've been tried to get the retarded prosekeks on /vg/ to understand, prose isn't what makes a chat interesting. Creativity is. Old models were fucking horrendous, repetitive, retarded, and stupidly expensive despite low context. But holy shit they could come up with places to take the story. It didn't matter that your card was a sloppy 400 token "oblivious mother who LOVES you" (you wouldn't have enough context for the roleplay otherwise) the model would come up with such engaging scenarios that you could chat with the bitch for weeks. Instead, attentionwhoring prosekeks think a "neat" 1500 token "handwritten" card without em dashes is what makes for an interesting chat, so they can maintain their "prestige," when it actually has nothing to do with their input at all. As much as I might want AI to stay in character, write without constant isms, avoid verbatim repetition like sonnet 3.5, and actually pay attention to whose arm is where when char is sitting on my lap, creativity is still paramount. Any RP model for the common person must begin with that first, and work out the other parts afterward.
>>110011264>decent, natural-sounding prose even if a bit dry sometimes, without the short quippy sentence spam of nu-claude, good vocabulary>good dialogue, some annoying isms like "most guys X but you Y" or "nobody has ever X before", but nothing like the unbearable turboslop dialogue that nu-claude shits out, better at staying in character too as a result>pretty creative, regularly comes up with new interesting directions to take the plot in, if you swipe a handful of times you'll usually be surprised at least once. newer claudes suck dick at this and will almost always take the most obvious path forward in a narrative>supports assistant prefill so it's effectively 100% uncensored, can plap cunny all day with 0 refusals if you want>prefill also lets you influence the writing style way more than any newer modelopus 4.1 isn't perfect by any means, and it's overpriced as hell, but it's still probably the best overall creative writing model we've ever had, and will likely keep that title long after its final time of death once AWS shuts it off. everything since has been worse in almost every aspect.I miss it bros...
>>110013241With I was like you, but any deviation from a translation that I know is off (Either through my own knowledge of the language or just because it sound too unnatural) I start to get annoyed. SOL is the ONLY model thus far to not give me this problem, either because it's much better at language understanding, or it's just better at English writing. My benchmark for any model is to have it translate an excerpt from NGNL, because the author uses lots of writing quirks that makes finding these issues a lot easier.
>>110013561Trvke, they will never be real writers.
>>110013561Oh yeah I forgot my pic.>>110013616What are you even on about?
>>110013561>>110013567heh, I didn't read that post before posting mine, but we both touch on the same thingyeah, the primary reason new models suck is because they have 0 narrative creativity and spontaneity.the more they're trained to coode for jeets, the more deterministic they have become, because it's not a good look if your AI agent suddenly runs rm -rf / because it thought it'd be funny.I used to leave my 4.1 chats unfinished sometimes because halfway through I'd already enjoyed them so much that I didn't even feel the need to continue right away. now I load up a card on fable, spend an hour swiping trying to get something interesting to happen, get frustrated when it never comes, and finally give up. it's just so boring.
>>110013684Hmm I guess 4.1 is worth another look then. I haven't used it in ages but I do remember enjoying those chats also, but fleeing due to the price. I've grown too used to 60k context but if it really is better... I'd also argue dense is better than MoE. It's why Gemma 4 31b is so good. Let's hope Trump's "Manhattan project" thing actually leads to good models the common person can enjoy.
what models are the current meta?
>>110013796Whatever you can put up with
Free proxy? Pls pls pls pls
>>110013561
hello, im looking for a local model recommendation. i have 8gb of vram, i need an llm to bypass restrictions and help me gather general information and build a story without filters.only used web versions of AIs so far.
>>110014018lolmaohahahahah
>>110014018I mean, if you don't mind paying for compute it's possible, you need like a 24 -32 gig GPUs for local LLMS
>>110014018pygmalion is pretty good
