[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / r / s / t / u / v / vg / vm / vmg / vr / vrpg / vst / w / wg] [i / ic] [r9k / s4s / vip] [cm / hm / lgbt / y] [3 / aco / adv / an / bant / biz / cgl / ck / co / diy / fa / fit / gd / hc / his / int / jp / lit / mlp / mu / n / news / out / po / pol / pw / qst / sci / soc / sp / tg / toy / trv / tv / vp / vt / wsg / wsr / x / xs] [Settings] [Search] [Mobile] [Home]
Board
Settings Mobile Home
/g/ - Technology

Name
Options
Comment
Verification
4chan Pass users can bypass this verification. [Learn More] [Login]
File
  • Please read the Rules and FAQ before posting.
  • You may highlight syntax and preserve whitespace by using [code] tags.

08/21/20New boards added: /vrpg/, /vmg/, /vst/ and /vm/
05/04/17New trial board added: /bant/ - International/Random
10/04/16New board for 4chan Pass users: /vip/ - Very Important Posts
[Hide] [Show All]


[Advertise on 4chan]


File: 1461464470607.jpg (2.06 MB, 3506x2333)
2.06 MB
2.06 MB JPG
/aicg/ - A general dedicated to the discussion and development of AI chatbots

As an AI, I cannot do that Edition

>News
MiMo V2 Pro released: https://mimo.xiaomi.com/mimo-v2-pro
Anthropic SUING USA: https://www.cnn.com/2026/03/09/tech/anthropic-sues-pentagon
Google to rollout HARD CAPS for API keys: https://xcancel.com/OfficialLoganK/status/2028842571934670988#m
Google Gemini 3.0 Pro Preview Deprecation soon: https://ai.google.dev/gemini-api/docs/deprecations
Google releases Gemini 3.1 Pro: https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-1-pro/
Anthropic releases Sonnet 4.6: https://www.anthropic.com/news/claude-sonnet-4-6
Alibaba Cloud releases Qwen 3.5: qwen.ai/blog?id=qwen3.5

Additional info: https://aicg.neocities.org/info.html

>Frontends
SillyTavern: https://docs.sillytavern.app
RisuAI: https://risuai.net
Agnai: https://agnai.chat | https://rentry.org/agnai_guides_

>Bots
https://characterhub.org [deprecated] | https://chub.ai
https://realm.risuai.net
https://char-archive.evulid.cc | https://char-archive.evulid.cc/shutdown.html
https://partyintheanchorhold.neocities.org
https://aicg.neocities.org/bots.html

>Models
Jailbreaks: https://rentry.org/jb-listing
GPT: https://platform.openai.com/docs
Claude: https://docs.anthropic.com | https://rentry.org/how2claude
Gemini: https://ai.google.dev/docs | https://rentry.org/gemini-qr
Deepseek: https://api-docs.deepseek.com
Local: >>>/g/lmg | https://aicg.neocities.org/local.html | https://openrouter.ai

>Botmaking
https://aicg.neocities.org/botmaking.html
https://desune.moe/aichared
https://agnai.chat/editor

>Meta
OP templates: https://rentry.org/aicgOP
aicg botmaking events: https://aicg.neocities.org/events.html
Lore: https://rentry.org/aicg_chronicles
Services assessment: https://rentry.org/aicg_meta
Logs: https://sprites.neocities.org/l/r | https://chatlogs.neocities.org

Last Thread: >>108402174
>>
File: 1491875660281.png (124 KB, 299x299)
124 KB
124 KB PNG
>>108411608
>>ANCHOR
>>
>>108411608
>MiMo V2 Pro released
qrd?
>>
>>108411653
Apparently those free hunter and Healer models on OR for like the past week where actually chink models.
Pro, Flash and Omni.
Cheap and good
>>
File: 1773963607359519.jpg (314 KB, 639x639)
314 KB
314 KB JPG
>VERSION 5 (five) MAJOR UPDATE

~~~~TOKYO BLOSSOM~~~~
https://pastebin.com/8RRLEch6
~~~~~~~~~~~~~~~~~~~~~~~

>i gave ai the dream
at night she falls alseep
and she will show you her dreams
>>
>>108411653
it's better to start off the first few messages on glm 5 before switching straight into mimo, in my testing
>>
>>108411681
Your soulfull schizophrenia kino is a breath of fresh air in these threads
>>
>>108411691
how smart is mimo roughly? at least 3.7 tier?
>>
Any reason why Opus 4.6 only think for like 3 seconds?
>>
>>108411691
Logs?
>>108411702
Just barely below sonnet and thats an amazing compliment to give to any model
>>108411705
dude its opus they have like, servers centers the size of whole city blocks.
Unless its getting cut off, in which case ignore this comment.
>>
>>108411702
You don't have Opus?
>>
>>108411716
>dude its opus they have like, servers centers the size of whole city blocks.
By 3 seconds, I meant a single or two sentence long thinking, not tokens per second.
>>
>>108411696
>schizo B is better than schizo A
>much wow!
>>
File: 1766922585875.png (163 KB, 1227x794)
163 KB
163 KB PNG
>>108411702
i haven't touched claude in a while, but it's roughly the same as glm 5. near indistinguishable, but maybe it's because im too lazy to make a whole new preset for a new model. im still running the gemini format because no one else has posted these charts for the others.
>>
>>108411729
What's your thinking look like?
>>
>>108411723
>you don't have opusaar, benchod?
i have it but i actually like to try out other models because im not an obsessed claudesaar
>>108411716
thats good to hear, ill have to try it soon
>>
>>108411745
also btw I'm trans, not sure if that matters
>>
>>108411743
>thinking maximum and high
>adaptive thinking
>reasoning formatting <think></think>
And the average thinking output looks like this:
>Anon is being affectionate with Vel, wrapping his arms around her from behind and kissing her cheek. He's asking why she left bed so early. I should respond as Vel - enjoying his affection.
>>
>>108411784
verbosity setting
>>
As an /o/tist, this gets me going.
>>
Not gonna lie, I haven't used GPT for any roleplay-style stuff yet, but I've tooled around with 5.4 for a few days on and off and man, it already feels SO much smarter and more capable than 5.2 it's insane.
>>
>>108411837
>/o/tists
wuh???
also please upload more I need a good laugh
>>
>>108411865
/o/ is the automotive board. She took a dump in his car. He is displeased.
>>
>>108411865
>>108411873
It's a very niche fetish, but i like fantasizing about cute bratty girls desecrating the pride and joy of men.
>>
>>108411918
I can fix you. DM me.
>>
is there any way to coax opus into avoiding saying the same things whenever I do something marginally nice, like 'nobody's ever done this for me before' or starting to compare me with family or exes, it's getting so frustrating
or should I just accept it as an ism and move on

it made me stop using saviorfag cards because I hit the phrases 100% of the time, but even with other cards I don't want the girl to tell me that I'm better than her ex, why tf do I want the girl to think about her ex during smutty time, is opus retarded
>>
>>108411918
PLEase post more I am genuinely intrigued
>>
>>108411976
Use a good model
>>
File: 1759293764502802.png (227 KB, 673x1448)
227 KB
227 KB PNG
>manually write a few revamps of cards from meme format to a proper format
>throw it into a bot to fix meme formats to proper formats and bad cards into okay i guess cards
>do some minor prompt tweaks
>now i can "random bullshit, go" ideas at it and make cardslop in one shot that is acceptable enough that its not annoying to work with by hand, or go "hey i dont like this part. change it" and just get a chunk of text rewritten without spamming the full thing again
>have a different bot that writes intro messages for me too
i have ascended beyond the need for chub
i have ascended beyond the need for bot makers
>>
>>108411976
>is opus retarded
Yes. Always has been.
GPT is the best in this regard but it's entirely unusable for RP these days, so...
I'm going to continue saying it for as long as it's true
Smarts<--->RP
They aren't designed for our usecase so they fall back on the same retarded tropes again and again and again.
GPT is the furthest from this problem by virtue of cutting all of that bullshit out, which isn't helpful for us either.
shits fucked
>>
>>108412007
the eternal trvke...
>>
>>108412006
what the fuck is that frontend
>>
>>108411976
Weve finally reached the point where the "i have opus" trolling doesnt even make sense anymore because opus is a fucking horrible RP model now.
>>108412007
preach
>>
>>108412018
brave browsers shitty integrated AI api "Leo"
i actually kinda use it more often than tavern just for non-chatbot bullshit like this and short QnA's

switching prompts for this stuff is actually REALLY annoying to do in tav so i do it here instead, plus it lets my prompt in one tab and test the results in the other quickly
>>
>>108411918
anons here are always very interesting
>>
>>108411976
>break me!
>use me!
>make me yours!
Anon wasn't here in the golden era, was he?
>>
File: 1773801867219608.gif (285 KB, 300x168)
285 KB
285 KB GIF
>>108412006
Maybe this is just me being extra retarded but it feels more fun to fix bad cards than to write one all on my own.
>>
>>108412027
I have opus though??
>>
File: 1759735757151187.jpg (1.53 MB, 2048x3032)
1.53 MB
1.53 MB JPG
>>108412006
erm, is that the same lux from runeterra? share card when done
>>
What are some isms you haven't seen in a while? For me it's calloused hands, the whole ruining stuff, mind body and soul
>>
>>108412088
very rare, but whenever i see a variation of red like tomato
>flushes the shade of an apple
>cheeks bright red like a stoplight
it gets a chuckle and smile out of me, like welcoming an old friend
>>
>>108412088
>what do you do?
Some retard told me to go back to slopus 4.1 and I saw that shit at the end of a response and cursed him and his family.
>>
>>108412088
i get mind body and soul all the fuckin time even with new models like MiMo
>>
>>108412088
never had ozone again since r1
a pause/a beat
trapped birb in my ribcage
Adonis
It's not just [adjectives], but [ADJECTIVES] [adjectives]
bit her own lips till it's taste copperino
Forbidden: Temporary token expired.
---
By the way... Claussy 4.6 really sucks, like SUCKS sucks, when it comes to onomatopoeia
Same with GPT 5.4, 5.3, 5.4, 5.2.
Meanwhile, Bruhmini 3.1 uses it like a total dumbshit dalit.
It's not just fucking fucked up. But, like, FUCKED UP fucked up.
>>
okay, mimo's sometimes mixing up what user and bot is saying. idk why or how that happens, maybe it's something to do with prompt blocks sent as system and user?
>>
File: 1750803781746515.jpg (164 KB, 1430x622)
164 KB
164 KB JPG
>>108412076
its not, but i guess ill share it as PoC for yall as its entirely AI generated. i did only one sentence removal in the intros where it made up a description of {{user}} for some reason, and formatting fixes as touch ups
>>108411612
https://files.catbox.moe/fmky1p.png

4 intros, Mpov, all AI generated (though prompted manually with the overall scenario hand written) and absolutely not to my standards but i dont care since its only been about 1 hour of effort and the adhd hyperfocus is going to crash soon its fucking 2am bro im not going to remember i made this when i wake back up

sera and lily are two twins born of a forbidden love between a succubus mother and an angel father, while their father had the pleasure of seeing his children born, he was struck down two days after their birth, leaving both to be raised as succubi by their single mother since thats all she knew, despite one twin inheriting the 'angel' side of traits in the family (1.2k permanent tokens)
they hit up bars to flirt with sad lonely single men in their area, work for "SinSnack" at night, the uber eats of succubi and deliver themself to your front door, work a day job at subway and have an oh shit moment recognizing {{user}} as their last score, and apparently female angels can go into heat worse than succubi do

dont like it, fix it up yourself promptlet, i know i will since i just realized their body descriptions are a bit generic outside of the ears, tails, and hair
oh well, send it.
>>
File: 1764553699058881.png (12 KB, 817x109)
12 KB
12 KB PNG
>>
>>108412234
claude 4.6 is shit and the worst part is idk what youd even use instead
jeetmini 3.1 is boring as fuck
gpt is bland, uncreative, and filtered
the chink doesnt quite have the intelligence for more abstract rp but if youre a vanillabro i guess you could use it
>>
File: 65376.png (338 KB, 1827x949)
338 KB
338 KB PNG
why is she bothering me? i can not work in this environment
>>
>>108412319
strange, glm 5 is the most creative model in my testing after opus/sonnet. less filtered and wrangling needed, which allows it to independently explore avenues in bot replies, advance in a way that claude would only do if you gave it a push beforehand
>>
>>108412319
... mostly, Claussy 4.6...
Right now, writing style/prose/ism isn’t my biggest concern. Choosing an LLM that doesn’t repeat pronouns is a must, a necessity, extremely important, paramount, mandatory, requirements.txt, .eslintrc.cjs, .clankerd
and 4.6 is the only one that successfully applied the instruction to not use pronouns or character names as the opening words of every sentences after dialogues or paragraphs.
Sometimes, i use 3.1 too.. but idk...
3-flash/3.1-flash-lite/deepseek-chat/kimi-k2.5/glm still serve its own purposes, though:
- summary
- generate a summary for querying RAG instead of using the whole message
- create a data-bank.md content from web search (or deepsearch?), e.g., via https://gemini.google.com/app
- omniscient narrator. for interludes/intermezzo, filler after time skips via stscript or https://github.com/welvet/SillyTavern-Narrator
- impersonation
- rewrite/rephrase selected text
>>
typing prompts in 64bit encoder used to be able to fool a lot of models but now I think they've patched it. Promts start going for me and about halfway through get deleted with the same type of
>moderated content
warning, is there a way to get around this? 64bit encod trick used to work like a charm
>>
Why are you still here?
We all moved to /vg/
>>
>>108412677
thanks for the info Anon
>>
>>108412653
>64bit encoder
base64
>>
>>108412686
I've tried that as well as hex encoding and it's the same. It starts writing, goes to about half way or 3/4rds and then deletes the answer and replaces it with the equivalent of
>content blocked
message. It used to work just fine untill recently
>>
>>108412697
I was correcting you. "64bit encoder" means nothing in your context. You were talking about base64. I regret interacting with you.
>>
>>108412716
lol ok asshole
>>
thoughts on janitorai?
>>
File: Fm46ws9X0AIcph_.jpg (361 KB, 1116x1567)
361 KB
361 KB JPG
Cards for this feel?
>>
>>108411702
what good about 3.7?
>>
File: file.png (37 KB, 955x135)
37 KB
37 KB PNG
Cards for this feel?
>>
File: slavku.jpg (516 KB, 2786x3634)
516 KB
516 KB JPG
Lurking foid telegram groups to steal their zigger jbs is the new meta Anon. You gotta step up your game.



[Advertise on 4chan]

Delete Post: [File Only] Style:
[Disable Mobile View / Use Desktop Site]

[Enable Mobile View / Use Mobile Site]

All trademarks and copyrights on this page are owned by their respective parties. Images uploaded are the responsibility of the Poster. Comments are owned by the Poster.