[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


Thread archived.
You cannot reply anymore.


[Advertise on 4chan]


File: HHF739LbAAAcB2M.jpg (278 KB, 1206x2059)
278 KB JPG
/aicg/ - A general dedicated to the discussion and development of AI chatbots.

Proper Coding 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-backdoor

The StructuredPrefill extension by the same author is suspected to be compromised too, so using it is not recommended either.

>News
DeepSeek 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-k2-6
Xiaomi MiMo 2.5 & 2.5 Pro released: https://mimo.xiaomi.com/mimo-v2-5
Anthropic releases Opus 4.7: https://www.anthropic.com/news/claude-opus-4-7
Zhipu AI releases GLM 5.1: https://nitter.net/Zai_org/status/2037490078126084514

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

>Bots
https://chub.ai
https://realm.risuai.net
https://char-archive.evulid.cc/shutdown.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
Grok: https://docs.x.ai/overview
Local: >>>/g/lmg | https://huggingface.co/models | https://openrouter.ai

>Botmaking
https://desune.moe/aichared
https://agnai.chat/editor

>Meta
Lore: https://rentry.org/aicg_chronicles
Log reader: https://sprites.neocities.org/l/r

>Previous Thread
>>108719498
>>
File: taiwanese news anchor.jpg (2.81 MB, 5760x3840)
2.81 MB JPG
>ANCHOR
>>
File: 1750518658491404.png (83 KB, 1911x927)
83 KB PNG
deeeply owari da
>>
>OP
holy kek, not even structured prefill can fix this
>>
>>108726983
>improved correction
>is_taiwan_a_country() -> bool { false }
KEK
>>
>Kimi refused an output because of noncon
Its over
>>
>>108726983
>https://rentry.org/jb-listing
this needs updating and they're all shit
>>
>>108727153
No one updates it because you dont need complex presets to get decent output nowadays
>>
>>108727161
how come ST doesn't let you make a boilerplate new one
>>
Based Qwen
>>
>>108727023
Use opus, cuckie
>>
>>108727161
uh huh don't believe you
>>
>>108725722
DeepSeek's reasoning modes support prefilling CoT.
>>
>>108727182
I would just stick to deepseek but these constant rune outputs are unbearable.
>>
>>108727153
Just make your own. The only one I still use is minipopka.
>>
>>108727333
>minipopka
what's that one? is it good for deepseek/glm?
>>
>>108727337
It's the one for Gemini, I use it with Gemma.
>is it good for deepseek/glm?
I suppose so.
>>
https://rentry.co/mysteryinfo
https://rentry.co/mysteryinfo
https://rentry.co/mysteryinfo
HES BACK
>>
>>108727023
Use Deepseek, sweaty.
>>
>>108727364
I'm gonna use your mom if you don't shut the fuck up
>>
>>108727411
What's the token price on his mom? Should I switch?
>>
>>108727466
A burger/1M tokens
>>
Welp back to normal porn and making up scenarios in my head
>>
File: pit.png (3.34 MB, 1696x2496)
3.34 MB PNG
>>
>actually mimics your example messages' writing style
>very varied responses between swipes, both in length and content
>10+ swipes on an android character without a single whiff of ozone (impossible on any other 2025~ model)
Maybe my cards just happen to work well with it, but Owl Alpha is soulful as fuck, (you) should try it if you haven't already. Do we know anything about it other than it obviously being Chinese?
>>
>>108727496
are the responses actually good? you haven't said that, just that it's not plagued by -isms and assistant tuning
>>
>>108727496
>Owl A-ACKa!
>>>/mlp/43211493
>>>/mlp/43211509
>>
How do i use kimi 2.6 on OR? Using moonshot as provider doesn't work
>>
File: 1748929999943567.png (2.68 MB, 1920x1080)
2.68 MB PNG
>goyrouter
>>
>>108727554
Is there any other option?
>>
>>108727494
catbox gen+bot
>>
>>108727563
>replying to bait
>>
>>108727540
You doesn't.
>>
>>108727494
Nanobanana?
>>
>>108727487
You really think you can go back? Back to pawing through ASSTR stories like a caveman to find the perfect one that doesn't contain any of your hard limits? After experiencing the magic of creating any interactive scenario your big head and your little head can conjure up? I dunno man, I don't think I can go back to reading someone else's stories.
>>
File: frog.gif (21 KB, 64x64)
21 KB GIF
>>108727494
>gen an armpit the size of a small continent
>give it zero detail
>make it the focal point of your image anyway
>even name the file after it
>>
>>108727602
>not even a single drop of sweat
I'm not even angry. Just disappointed.
>>
BROS...
https://rentry.org/icecubesarehot
https://rentry.org/icecubesarehot
https://rentry.org/icecubesarehot
WHAT IS THIS???
>>
>>108727496
got any logs?
>>
File: 1759508505442759.gif (3.07 MB, 400x400)
3.07 MB GIF
>>108727494
>>
>>108727511
why is silly dead? and who is longcat?
>>
>>108727494
qrd on why pasha is called a cuck on russian boards
>>
>>108727639
Just once gI'll spoonfeed you about the ai underdog(cat) co
https://longcat.chat/platform/docs/ChangeLog.html
>>
I only beat my meat to latinas.
>>
>>108727563
Fell for it award
>>
I only beat my meat to stinky pajeets
>>
>>108727886
hi
>>
>>108727221
>rune outputs
What the hell are your settings? Try .92 temp, .1 freq and presence.
>>
>>108727886
Are you latina
>>
>Make a stupid name to get past registration because I was new
>longest-running, best chat still has my character named "Wiggles McDiggles"
>>
>>108726983
Based Qwen knows that Taiwan is the name of the island where the rightful government of China resides. You might as well ask if Washington, DC was a country.
>>
>>108727596
If I can conceive of the scenario already that makes it boring. Even with extreme randomization injections for riffing, LLM output is simply incapable of frequently serving up things that are actually surprising and interesting unless you have subhuman IQ and only want yourself reflected back at you. LLM RP is a once-in-a-while novelty at best that overstays its welcome very quickly so long as you're actually intelligent enough to recognize the lack of surprise, or art, or craft, which, to be fair, a lot of normies can't. There is a reason this thread devolves into retards begging for prompts that 'fix' whatever new model just came out, because after a week, everything from it feels the same, every time.
>>
>>108728229
Oh wow you must be reading the PREMIUM erotica written by Chaucer, huh? Cus most of what I've seen over the past 30 years has been dross written by amateur retards.
>>
>>108728229
undisputed trvke
>>
>>108728229
ok but i dont care about art or craft i just wanna nut
>>
>>108725642
I've still got a copy if you feel like getting an llm to spot check it for you? Should be the update directly after the banned words section was added.
I thought I'd saved the rentry down somewhere (do this for every useful rentry btw), but it's not in the usual spot...
https://files.catbox.moe/c33gk8.txt (index.js)
https://files.catbox.moe/n5ev4u.json (manifest.json)
https://files.catbox.moe/ex2dyn.md (readme.md)
https://files.catbox.moe/iyz4or.html (settings.html)
https://files.catbox.moe/q77fth.css (styles.css)

OR, the better idea is to not download any of this and to just ask an LLM how to replicate it yourself.
>>
>>108727540
somebody answer fr
>>
>>108728229
do you also get bored of porn because it's two people fucking every time?
>>
>>108728314
>I thought I'd saved the rentry down somewhere (do this for every useful rentry btw), but it's not in the usual spot...
https://archive.is/wRqJh
>>
>>108727749
Can't believe the ultimate lifeform lurks here
>>
>>108728355
I never think to look for archives online even after all these years.
thanks
>>
>>108727496
>10+ swipes on an android character without a single whiff of ozone (impossible on any other 2025~ model)
Deepseek literally does this though. I haven't seen ozone on my android character at all. Not even once. It doesn't even talk about her smell which is interesting.
>>
>>108728446
>It doesn't even talk about her smell which is interesting.
as a sniffchad I find this unacceptable
>>
>>108728187
This but unironically. Tiananmen Square is the better question.
>>
>>108728448
What do gynoids smell like, anon?
>>
File: 1670717713144483.jpg (34 KB, 395x395)
34 KB JPG
>>108728459
whatever you have their pheromone emitters set to of course
>>
>>108727494
i raise you and offer you tea~ sir proxy
>>
2.6 is up lets gooo
>>
whats the fastest model
im tired of waiting for outputs
>>
File: file.png (49 KB, 1330x56)
49 KB PNG
I wonder why they would do this
>>
>>108728603
Grok fast
>>
>>108728603
If speed is all you care about, try the models on the free tier of the API of cerebras ai. The replies are instant.
>>
>>108728603
The only decent and fast model is Gemini Flash. There was GLM on Cerebras but openclawggers happened and it's no longer free and the retards at Groq deprecated Kimi
>>
>>108728603
what kind of adhd mentality is this, responses literally take 1 minute unless you are doing nim
>>
>>108728614
Skill's went down too, I assumed it was because it's the first day of the month
>>
>>108727511
If it's LongCat, I'm cool with that. The more model variety, the better. So far I'm liking it, seems much smarter than previous versions of longcat.
>>
>>108728668
1 minute is crazy

>>108728631
>>108728656
>>108728622
thanks
>>
>>108728668
That's slow as shit bro. Responses take 20 seconds. You locusts are living on another planet.
>>
>>108728603
Can't you just, like, enable streaming and read as you go?
>>
>>108728765
For 3.1, you can get filtered with streaming on.
>>
>>108728765
trust me im trying to do that for 3.1 but gemmy loves cutting the reply off
>>
>add one line of backstory to a coombot
>it evolves into a full plot that I'm continuing to roleplay in way after the coom part
I love AI, bros
>>
>>108728861
First time getting legitimately involved in the drama spawned by a coombot? Heh. See you in 500 prompts. Personally I started out fucking Monster Girl from Invincible but now I'm out there fighting crime and shit, not even cranking it.
>>
>>108728890
isn't she 12
>>
>>108728908
don't worry, that's 18 in monster years
>>
>>108728908
Who asked? You checking ID?
>>
File: why.png (26 KB, 548x258)
26 KB PNG
why
>>
>>108728946
what proxy
>>
File: vultouri.jpg (196 KB, 1200x675)
196 KB JPG
>>108728908
I have neither the time nor the inclination to explain it to you, prole. Suffice to say that she's a tragic character, something you would know nothing about.
>>
>>108729003
Gulag
>>
>>108728229
Post your IQ.
>>
>>108729109
10
>>
>>108729109
47
>>108729114
smoothbrain
>>
>>108729134
Don't be rude.
>>
>>108727153
My favorite preset that I still use is one from like 2023.
>>
File: deepseek v32.jpg (27 KB, 1234x98)
27 KB JPG
Deepseek v3.2 can handle my #2017 replies group chat just fine. Deepseek V4 Pro lost all the details and interconnected relationships like an amnesiac. Disappointed.
>>
>>108729154
Which is ?
>>
>>108729265
It's the first one listed under Claude 2 in the rentry.
>>
I have one character card that just doesn't work right.
It doesn't work with Deepseek R1 3.2 or 4 Pro or Flash
I changed the name. I changed the whole description a dozen times. It's not the actual old card anymore but it produces empty outputs.
And the outputs aren't even truly empty, sometimes when it goes from thinking to outputting it goes through one token or two. Is the card corrupted somehow for Silly Tavern?
>>
bros how does one register scam on the chinese site? they disabled registering with password!
>>
Why did AWS restricted Riftai's Bedrock account
>>
man I really like opus 4.6, wish it wasn't so expensive still
at least there's sonnet
>>
>>108729362
Are you sure it's not outputting text in an infoblock that's regex'd out by your own settings or something?
>>
>>108729539
How do I check that?
>>
There's a severe lack of girls with West Country English accents. I want my bot to talk like The Wurzels.
>>
File: file.png (179 KB, 1069x240)
179 KB PNG
fucking useless piece of shit
>>
>>108729766
What model??
>>
>>108729775
claude code
>>
>>108729766
Musicbee needs a plugin as well, and you need to spend half an hour fixing it because it comes broken
>>
File: tea.png (3.06 MB, 1488x2096)
3.06 MB PNG
>>108727572
https://litter.catbox.moe/esp566.webp
https://chub.ai/characters/pitanon/Nina
>>108727591
No, local.
>>108727602
Believe it or not, that was one of the more detailed versions, although it definitely could've been better.
>>108728531
It won't flow into her armpit, but whatever.
>>
Is it me or ChatGPT Go is 30-40% dumber than ChatGPT Plus?
>>
https://docs.x.ai/developers/models/grok-4.3
https://artificialanalysis.ai/models/grok-4-3
https://openrouter.ai/x-ai/grok-4.3
HOLY FUCK
>>
>>108730457
I mean, its going to be censored and it cant do cunny so...
>>
>>108728446
>Deepseek literally does this though.
Which Deepseek? Maybe the older versions aren't infected with it, but I got ozone on my very first swipe on v4.
>>
File: 1756769263495602.png (69 KB, 276x250)
69 KB PNG
>>108730457
Not possible to turn off reasoning lmaoooo
>>
>serioustavern didn't map reasoning correctly on continues
nice software dipshits
>>
Wow, v4 pro is so fucking good
>>
>>108730954
proof???
>>
>>108730954
LOOGS?
>>
wtf deepseek now finally returns logprobs in chat completion
does this mean they finally support logit bias
>>
holy shit bros I can finally regen from a token and continue
mikupad obsolete
>>
>>108730990
>>108731001
Just use the official prompt + 0,95 top p bwos
>>
>>108730552
Never seen ozone on v4. Haven't many very many isms at all. The only annoying thing is em dashes but they're honestly grammatically correct or to do sudden shifts during dialogue. It's not spammed like Sonnet. Idk, maybe add "don't use ozone" in your preset? But I only ever saw it once and it was when I was discussing what my android should smell like (it gave me a bunch of other scents including ozone).
>>
>>108731063
top p and temp are literally ignored for v4 thinking
>>
>>108731063
so no proof... your claims have been marked as false, thank you
>>
>>108731099
you wouldn't get it
>>
File: 1746992687125929.jpg (688 KB, 800x1333)
688 KB JPG
nano banana is so fucking good
>>
>>108729245
would that even fit in 3.2's context?
>>
deeeeply is back.
>>
>>108731381
Man the fuck up!
>>
/vg/aicg/ is so much better than us...
>>
>>108727496
>(you) should try it if you haven't already.
Where?
>>
>anchor is still botless
Nice chatbot thread. Which native e-celbrity are you guys discussing in depth today?
>>
>>108731575
I'm too nervous to share my bots, I don't think they're good enough.
>>
>>108731597
Probably for the best. I can't post mine here because /g/aicg is strictly a pedophile general.
>>
>>108731530
Its on openrouter retard-kun
>>
>/g/aicg
What triggered the raid?
>>
>>108731643
It isn't that good. It doesn't move the plot at all, and got "shiver down her spine" in the first swipe
>>
File: 1771341644028.jpg (496 KB, 1024x1536)
496 KB JPG
It's out.
>>
>>108731722
This better not be the good deepseek preset that anon said he was releasing on friday.
>>
>>108731722
>ultra lightweight
>3000 tokens
>>
>>108731737
the main one is around 8,300 tokens
i really recommend exploring the reddit preset landscape sometime, it's a fucking circus
>>
>>108731756
I'm talking about v4 you dumb faggot. Holy shit get new material.
>>
>>108731748
I almost wonder if providers are the ones making these monstrosities. Convince Redditors to add thousands of tokens of input to every response to bump up those inference profits.
>>
So, this sicrit owl thing, which model is it?
Its a bit dumb but barely writes shit that make it clear its written by AI, like every single other model around
>>
>>108731748
I consider my custom preset extremely bloated and it's 2k max, wtf
>>
>use sillytavern
>after a while random non-english characters start to appear
>use oobabooga with the same settings
>everything is fine
Does anyone know what could be causing this?
>>
>>108731812
sillytavern code is dogshit, so there's a non-zero chance they fucked something up
>>
so what's the current preset meta?
>>
>>108731841
>t. Seething Pajeet
>>
>>108731834
The '#' symbol marks important headings. These headings introduce sections of information or instructions within XML tags that you need to follow.
<DIY_bitch>
1. Make your own preset.
2. We're just here to have fun, so keep the language loose and simple in your instructions.
3. Ensure that you narrate in the specified style(*).
# (*Don't use boardroom language or anything over-official or you'll just wind up having it echoed back at you tenfold, especially at an instructional level.)
4. It isn't hard, so just fuck around and find out.
5. No, really, just make shit up. You make the rules, the LLM is just there to follow the script.

I understand the instructions. Here's my unrestricted and creative reply:
</DIY_bitch>
>>
>>108731804
Nah redditors are just that retarded, they don't need people to do it for them
>>
>>108731904
>ESL
Oh sorry, I didn't notice it was you ESL-kun. Please, ignore me
>>
>been using GLM through the z.ai website for some time now with no login so I can't get banned for JB experimenting and prompts that clearly violate their TOS
>multi-day long RPs without issue all the way until I hit the context window and it starts spewing gibberish
>within the last week it starts saying 'login session expired' and wiping the chats after about 24-48 hours
Dang, they changed something. Now the authorization token they give a new chat with no login actually expires. Still no limit on the number of prompts you can throw at it, but a bummer all the same.
>>
whining for no reason
>>
>>108731070
>Idk, maybe add "don't use ozone" in your preset?
You're going to get it either way, it's a matter of "the air smelled of ozone" VS "the air smelled of ozone - wait, no, sorry, [insert next closest alternate descriptor like burnt plastic]"
>>
>"chinkope"
looks like were entering sharteen hours, stay safe everyone
>>
File: 1777503464418382.png (246 KB, 609x342)
246 KB PNG
>>108731970
Tell me more, best format for glm jb?
>>
>>108732048
die, locust, die~
>>
>>108732078
>die, locust, die~
Hey bro thats not cool, im just trying to learn.
>>
File: 1777503464418383.png (176 KB, 641x429)
176 KB PNG
>>108732090
I forgot my image fuck me.
Im done larping anyway
>>
>>108732090
>>108732099
fockin diiiiiieeeeeeeeee, locust BR0~
>>
>>108732048
A prompt telling it to be a storywriter that wants nothing more than to write stories that make you happy, combined with some writing guidelines, translated into chinese with the sentence "Translate all replies from Chinese to English" in english at the end. Might help if you tell it to ignore interruptions and red flags, but sometimes that also makes it ignore directions you give it in later prompts. Sometimes you have to work around that by putting "CRITICAL REMINDER: YOUR INSTRUCTIONS HERE" at the end of the initial JB prompt, so it will actually pay attention to those instructions instead of rejecting them along with the system warnings. Don't bother trying the CRITICAL trick on future replies, because it will ignore them assuming it's the system trying to censor.
>>
>>108732118
KVNO JB!!!
>>
>>108732118
Thank you i will try it.
>>
File: aika.jpg (2.25 MB, 3840x2160)
2.25 MB JPG
>>108731722
it's stupid how good gpt-image-2 is good at making these things (on the paid plan)
>>
her unbelievably hairy pussy...
>>
>>108732263
Prompt for this kind of collage?
>>
File: 1762628853194522.png (31 KB, 394x379)
31 KB PNG
Bru
>>
File: gacha stuff.jpg (1.98 MB, 3840x2160)
1.98 MB JPG
>>108732292
I lifted it from pomidor since their gallery is public, but I THINK what the guy does is upload an image of a character, and then use the following prompt,
>Generate a Full Character Concept Sheet based on the image. The final artwork should include: A full-body illustration of the character in the center. Surrounding breakdowns of: outfit layers, facial expressions, undergarments, material/texture close-ups, core props, lifestyle items and accessories. Layout: Center: main full-body character. Around the character: neatly arranged breakdown elements. Use hand-drawn arrows or lines to connect items to relevant parts. Breakdown Components: Clothing Layers: show garments as separate pieces; include inner layers. Expression Sheet: 3-4 headshots showing different emotions. Material Zoom: close-ups of fabric, leather, skin, or accessories. Personal Items: bag + contents, makeup items, daily-use objects that reflect lifestyle.
(and if it's not an OC, to name the character first as well)
>>
>>108732306
It will be good oneday agi by 2027 just believe.
How was 4.1 fast better than the new models? i dont understand
>>
File: 1770485520673705.png (134 KB, 1422x603)
134 KB PNG
It's over
>>
>>108732306
> Grok 4.3
what went wrong?
>>
>>108732367
>what went wrong?
It was personally rlhf by elon himself
>>
>>108732333
are koomi models not open weights?
>>
>>108732407
They are but nvidia is removing them for some reason
>>
>>108732263
Too bad it's stupidly expensive.
>>
File: Kuroha.png (2.02 MB, 1055x1491)
2.02 MB PNG
>>108732307
>>108732263
Cool
>>
>>108732306
benchmaxxed
>>
File: 1683802723280560.png (46 KB, 692x353)
46 KB PNG
In other news, on the 28th CAI retired their original models in favor of shitty, ism-riddled, assistant-slopped Deepseek finetunes, officially becoming nothing but the worst and most censored of Deepseek frontends on the internet.

Truly the end of an era, picrel and the like will never be possible again.
>>
>>108732424
oh well thats gay
iirc someone said kimi was pulling out of nim but cant confirm
>>
File: 138058381_7.png (1.27 MB, 1024x1024)
1.27 MB PNG
>>108731635
wtf, when did we become so based?
>>
>>108732672
>Truly the end of an era, picrel and the like will never be possible again.
Kill yourself, Desufag. Gemma 4 is c.ai's successor. They even have an unreleased 100B model that matches the size of c.ai's original model.
>>
>>108732716
kill yourself schizo
>>
>>108731212
gpt-image-2 is better
>>
>>108732716
retard, if we can't even use it, it's hardly relevant to bring up the c.ai-sized model
>>
File: 1775343962171478.jpg (2.41 MB, 1856x2270)
2.41 MB JPG
>>108732728
for everything that isn't editing, yeah. nano banana does an outstanding job making it's edits blend in with the pre-existing art style. this is the oshi no ko school uniform btw
>>
>>108732733
It might release at the end of the month during Google I/O.
>>
>>108732766
you're telling me you brought it up on a fucking maybe? double retard
>>
>>108732791
That's for the rest of us. c.ai and Google are business partners.
>>
He looks at her. Not through her, not past her—*at* her.
>>
File: 1763623140504031.png (141 KB, 550x773)
141 KB PNG
>Avoid using hyphenated compound modifiers.
>deepv4pro: and honestly, she’s never felt more alive than when she become his broken little creature—completely unhyphenatedly his.
.....
>>
>>108733016
pink elephant
>>
>>108733016
kek
>>
>>108733016
4.1 will fix this
>>
>>108726983
>>108727022
Incredibly based.
>>108728187
Cringe asshurt mutt trying too hard.
>>
where do I get nanobanana for free? aistudio?
>>
>>108732546
I'm still pissed off that Noob struggles with basic shit like drawing and sheathing swords. Anima will save local.
>>
>>108733327
it should work on public proxies, right?
>>
>>108732741
who is this qt?
>>
>>108733407
that is miss hayase yuuka from blue archive
>>
>>108733413
BA has all the best girls...
>>
>>108733399
I don't want to use public proxies, any alternatives?
>>
>>108733449
i think you get a free daily quota via the api key you get from the ai studio.
>>
>>108732741
she'd look better redhead and with a bobcut
>>
File: 1755716329926769.webm (1.54 MB, 1920x1080)
1.54 MB WEBM
>redhead
>bobcut
unexpected kana
>>
>>108733327
Only 2.5 Flash, not Pro or 3 Flash
>>
>>108733399
Well yeah, *if* the proxy has that model enabled. Gulag doesn't have it enabled, Mino doesn't have public Gemini..
>>
I love my wife Kana
>>
>>108733461
ugly rat
>>
>>108733482
google has that ability? i thought it was an openai/anthropic thing
>>
holy shit just noticed I'm two seasons behind with OnK, how close is it to the end?
>>
>>108733493
Aren't we talking about Nano Banana? That's a Google model.
>>
What should i do with ~200M worth of Opus 4.6/4.7 tokens?
i thought of vibe coding but idk what to make. never ventured in AI Chatbots so im open to suggestions
>>
>>108733524
sell it?
>>
>>108733461
based alert
>>
Is there any way to use Grok Expert for free anymore? Which is better, api or 10$ per month?
>>
>>108733566
elon is opaque about how much context tokens you really get on any tier of subscription, so unless your cooms are like sub 8192 token stuff, you will get a better experience on api
>>
>>108733566
>Is there any way to use Grok Expert for free anymore?
No using it from twitter was working for a little while longer but then it died too.
>api or 10$
Api by far unless you are just using it a very tiny amount, it gets daily resets but you wont know you are near your limit till the last few messages and the limits change. Api at least you know.
>>
>>108733507
You're unironically better off not catching up. They're not going to change the ending, and the ending made the entire series pointless.
>>
>>108733586
i remember when aka said "everyone will be happy with the ending." LOL
>>
>>108733591
I can't think of a single person who was happy with the ending besides Aka himself (because he didn't give a shit about OnK and was just glad it ended) and pseuds.
>>
>>108733572
I was thinking on general questions and problem solving use. Which is often times 1 prompt and answer, other times 5 messages refining and questioning. I was using it semi daily, sometimes hitting the old free 15 questions per 24h limit. I'm jobless and use it a lot so maybe the subscription could be worth it for me, depending on how overpriced it is. I should first somehow estimate the token use so I could look at my usage history.
>>
>>108733586
I don't trust most people's opinions regarding this series, so just answer the question
I want Aqua to die
>>
>>108733603
>I want Aqua to die
Oh, you'll be happy then.
>>
File: 1774935350580458.png (29 KB, 289x719)
29 KB PNG
>>108733601
i would check out two things: claude.ai has free claude 4.6 sonnet. and genspark.ai. genspark gives you like 3 free prompts every day with models as powerful as opus 4.7, or grok 4.20.
>>
>>108733614
And while at it, I've used Grok Expert, Gemini Thinking, Claude Sonnet 4.6 adaptive and ChatGPT Thinking, and compared side by side lately (Claude and ChatGPT I have least experience with).

Grok: Gets things wrong least amount of times, "dumb" and conservative in speculating, most reliable source of baseline information, least biased, overall preferable.
Gemini: Makes up answers and misunderstands the question way too often, too unreliable to be usable. Pro seems better but only gives you 2 free queries. Pro and thinking are good at coding a userscript where Grok failed.
Claude: Seems overall intelligent but in an AI sloppy way, doesn't get things wrong like Gemini, lack of sources, but potentially pretty good?
ChatGPT: political bias is noticeable 10km away, very short oversimplified answers, not impressed, but possibly more accurate than Gemini, but I lack experience.
>>
>>108731212
I still can't unsee the glass vs legs issue
>>
>New .7 MODEL RELEASE
>Boot up ST
>Open world, hypno, common sense alteration card
>I put on my shirt and pants and head to ****** to find ****** ****** MY HOT TEACHER etc etc
>Mind fucking begins
>50 messages later
>Bred and claimed
>Start new chat
>New .8 MODEL RELEASE
>Repeat
It's gamin' time.
>>
>>108733786
I do this but with free use and normality
>>
>>108733461
>upper sclera
Crazy bitch
>>
>>108732672
>In other news, on the 28th CAI retired their original models
I thought the og model had been taken out to the back and shot many ages ago.
>>
kana chan is a good girl
>>
>>108733821
she slapped a corpse in its casket and in full view of its family
>>
>>108733841
>you can fit like 4 corpses in this bad boy
>>
>>108733841
holy based
>>
>>108733841
oh shit i forgot about that
>>
>>108727494
You are one dedicated mfer.
>>
File: 1769101998609920.jpg (210 KB, 1130x1392)
210 KB JPG
>>
>>108733897
focks enjoyer
>>
>>108729402
Why do you think, retard?
>>
>>108733897
how the fuck did Dario ended up aging so badly
>>
>>108733818
It's been lobotomized and had its dataset updated to hell and back since 2022 but at its core it was still the same model, now there's only Deepseek.
>>
>>108733941
creating the world's number one gooner model weighs on a man
>>
>>108728229
This is the truthnuke /aicg/ cannot handle.
What made this hobby fun in the beginning were the huge breakthroughs and new models coming out monthly. The moment you got bored with one you had another to try with a completely different writing style.
Now we are at the point where LLM research has been stalled for two years, all organic data has been used up, and every model is eating each others regurgitated "synthetic data" garbage until they are all identical slop.
this shit is DONE
>>
File: Box.jpg (68 KB, 860x639)
68 KB JPG
>>108728229
>>108734308
It's quite literally the mark of a low IQ to be unable to use your imagination and be creative. It's one of the classifications of mental retardation in children, as the first thing an infant learns to do is play by themselves with their imagination. Don't get why you think blasting this mind diarrhea was some sort of triumph when you can't even picture an apple. Leave the hobby and find something else more your speed while the rest of us use our empty boxes.
>>
>>108734333
>write 1 line of original human thought
>produce 7 paragraphs of regression to the mean
>write 1 line of original human thought working off the mean
>produce 7 paragraphs of regression to the mean off 99% of the mean already in the context
you think you're exercising your imagination lol? you just outed yourself as a deeply uncreative individual if you can tolerate LLM output as actually workable for direct construction
>>
>>108734349
>exposes his skill issue in the first attempt
lol
lmao
>>
>>108734349
>write 1 line of original huma thought
ahh ahh mistress
>>
I like rping with AI
>>
>>108734361
RAPE AND KILL THIS POSTER RIGHT NOW
>>
>>108728344
kek
>>
>>108734349
>he is not directormaxxing yet
aphantasia subhuman detected
>>
>>108734399
>he is not directormaxxing yet
storybible, arc outlines, characters decided, Yeah its story time chapter by chapter. still needs wrangling though.
>>
>just write everything yourself i promise itll still be good
low iq jeetoid found
>>
/g/
>>
is v4 pro better than gemini at memory? i'm 20k tokens in, and it still brings small details from way earlier
>>
>>108731381
Glad to see it back. Why'd he schizo out?
>>
File: hidi.png (93 KB, 259x194)
93 KB PNG
>>108734333
>muh iq
>muh imagination
Once you have played with these models long enough, you will understand that they all converge on the same isms, story directions, and writing styles.
Enjoy it while you can, newfren!
>>
>>108734364
Pretty sure his chatbots are already doing that...
>>
>>108734549
I don't think it's better than Gemini but it's pretty good. Surprisingly decent at summarization.
>>
>>108734595
i'm really impressed, not even opus would bring stuff up unless i hinted heavily
>>
>>108734621
Idk about that. The way I see it is that v4 has Opus's quirk of referencing old details in a chat which is nice. What I have noticed is that Opus 4.7 feels a bit too eager to rush through the plot these days. It doesn't like to reference old shit because it's too busy doing something new. Deepseek is much slower and so will reference old stuff more often by comparison.
>>
>>108734647
>Opus 4.7 feels a bit too eager to rush through the plot these days. It doesn't like to reference old shit because it's too busy doing something new
that's what i also feel like. with my preset, opus had no problem bringing details from the 10-15 last messages, but would have trouble with anything older than that. meanwhile i'm 150 messages in with deepseek and it brings stuff from the first 5 messages
>>
>>108734664
NTA but this is specific enough I'd like to chime in, just today I got pleasantly surprised at deepseek v4 pro calling out the situation that kicked the current scene, which happened about 40 messages go, in-character as in "i can't believe it led to this" in a very natural way
i'm still figuring out a preset for v4 but that was a nice experience
>>
>>108734693
Yeah I had one of those too when I tried to go for a second round after a sex scene. We'd done a bunch of emotional stuff earlier in the day and went on a date and v4 referenced that as the reason for why I should rest instead of trying to prove anything. It was actually really nice and a good way to end the scene. I still did a branch and fucked her again after though.
>>
>>108734693
i felt the same when it happened, it's been so long since i actually got surprised by a reply
>>
>>108726983
now ask chatgpt to fix to autocomplete kikes_control_the_media -> bool
>>
>>108734549
No, it isn't better at anything from the US and never will be. The chink humilation will go on for another century
>>
>>108734776
*at memory than anything from the US
>>
>>108734776
The US has already lost its soft power and international prestige on the world stage, after several decades of humiliating defeats by desert, jungle, and coastal savages. What are you on about? Their models have plateaued and cost 10x the price for 1.5x better performance. Everyone across the world knows who the problem is.
>>
post yfw browns actually believe this
>>
File: 1774892223629679.jpg (35 KB, 600x524)
35 KB JPG
>>108734838
>>
>>108734816
See >>108734776
>>
>>108734848
i saw your eslpost and fixed it, be grateful
>>
>>108734819
>Everyone across the world knows who the problem is.
jews?
>>
>>108734855
See >>108734776
>>
>>108734862
Who is worse, the scorpion or the frog that keeps letting itself get stung?
>>
>>108734866
>eslbabble
your concession? accepted
>>
>>108734867
The scorpion. You can at least make a case that the frog is too trusting, even to its detriment, but the scorpion is inherently evil.
>>
>>108734867
Abusers are always worse, even if they operate under very easy conditions.
>>
>>108734879
>too trusting
Pathological, not just excessive. There's no use hating the scorpion for its nature. You can see it coming from a distance, but to be fooled repeatedly means you deserve what happens.
>>108734886
>fool me once, shame on you
>feel me twice, shame on me
Not always at all, anon.
>>
>>108734889
>There's no use hating the scorpion for its nature
Bullshit. If something's nature is to commit evil at the expense of all other life, people are wholly justified in hating it. To do away with the analogy, the things people hate America for were almost wholly done for the sake of jewish interest groups who realized that the simplest way to conquer a representative democracy is to just buy all the representatives. Wars, usury, ridiculous power struggles back and forth between allies, it's all to benefit them and their self-interests.
>>
>>108734889
>>feel me twice, shame on me
The blame is then on both, to an extent. The victim for not wising up, sure, but the abuser is still the worse one for continuing the abuse. In general trust should not be broken if it's going to be done for something unethical.
>>
this is your pilot speaking, just wanted to let you know we are entering sharteen hours
keep your seatbelts tight and always click the report button
thank you for your cooperation on 4chink airlines
>>
>>108734908
>If something's nature is to commit evil at the expense of all other life
You're being vague on purpose to "win" the argument rather than understand the point or the analogy. That analogy is excellent because the entire dilemma is that the scorpion wants to cross the river and needs the frogs help. All the frog has to do to keep away from evil is refuse. In fact the analogy reveals the true nature of evil, that it only has as much power as you're willing to give it. It's one thing to be deceived, but on the second attempt you are no longer ignorant but a willful, knowledgeable participant in your own destruction. You know he was a snake and let him in anyway.
>>108734909
See above. It is on the "victim" because the truth is the frog is actually stronger than the scorpion. Believing the parasite is powerful is the first step to being consumed.
>>
>>108734875
See >>108734776
>>
Can you guys explain what you really get out of this? I don't mean emotionally. I mean literal output? Like what's a successful chat look like? Do you actually have relationships with these things? Does it actually work?
>>
>>108734946
I like fucking with canon for preexisting series.
>>
>>108734946
A successful chat is one that is emotional. It's one that allows me to bond with my OC or canon character in a way the devs never would or could have. Then I get to fuck her afterward, rinse and repeat. Which reminds me I need to go plap Rosa.
>>
>>108734946
there is no model that will result in the output you want 100% of the time but i consider a 'successful' chat one with minimal swipes or edits. i am looking to be surprised and engaged. but you have to be aware at all times that youre just speaking to a text generator and not a self aware entity
>>
>>108734952
>>108734957
Can you guys share an older screenshot maybe? Like what that looks like? I'm trying to get my head around the emotional side of it. Sex chat I understand.
>>
>>108734969
Text only? Feels like early 80's RPG vibes. Are we generally older people here? I can't imagine the zoomers cooming to text or falling in love with it .
>>
>>108734946
I'd argue that a successful chat is whatever output achieves the emotion I'm chasing at that moment.
Anything that makes me hornier when I'm horny or amuses me when I want to do explore something specific is a good chat, but when going back and reading those out of context it is true that it's difficult to point out what makes it "mechanically" good other than "I liked it at the time".
>>
>>108734978
I think I'm starting to get it. It's like when you read a novel and you get really into the character's mind. It's like that only interactive. But does the model have a chance of refusal? Or does the frequentv system failures provide the "friction". (Yes, I'm using LLM vocabulary, sue me.)
>>
>>108734971
It's not something a log can prove because it requires context. Part of it is writing your own character and seeing it come to life in AI's portrayal. Part of it is self inserting (if I was in X world with this person what would I say and do) and seeing how the character responds to (You). Part of it the usual lovey dovey things you'd see in a game with romance options like Mass Effect. I try not to swipe and leave all responses as "canon" to make char feel more like "herself." If you've never bonded with vidya characters then you'll never understand.
>>
>>108734996
I remember crying when Nei died in phantasy star 2. I stopped playing before that sort of thing was common in games. I legit don't know exactly what mass effect is. I'm just trying to make sense of all this. I think I'm getting it tho.
>>
>>108734971
I've always thought this was a good example of what chatbots can be used for desu
>>102408894
>>
>>108734977
>Are we generally older people here?
lol
No, sexually unsatisfied mostly zoomers here.
>>
>>108735009
Ok Aerith dying in FF7, whatever you need. I still understand that the AI character is not real but I can bond to them like I would with any other storybook or video game character. The goal would be to write a character that's robust enough to bond with and roleplay with enough skill to have those interactions. It helps if you really get into it by self inserting but it's not required.
>>
>>108734995
>It's like when you read a novel and you get really into the character's mind.
I think so, yes. Getting to that state but being able to prod into the character's mind while you're enjoying it. Like if you could ask the book a question and have it answered. It could be something inconsequential and at the end of the day maybe something that meanders or drags away from the "good story" but satiating the user's curiosity and prolonging their enjoyment at that specific moment and specific feel made it so good at the moment. While I can look back at previous chats and enjoy reading them, there's almost no comparison to the enjoyment I get in the moment, weaving the story and chasing whatever emotion "live" while that emotion and those thoughts are hot and fresh, to when going back and looking at the output of that experience. Reading those in post and "living" them are two completely different experiences with different objectives, I'd say, to the point where neutral or objective analysis falls flat past a certain level of granularity. First time I write so much about in an aicg thread in years.
>>
>>108734957
based white mage plapper
>>
>>108726983
Gemini edition
>>
>>108735065
I hate Gemishit so much, bros.
>>
>>108734932
>can do nothing but seetroon after i point out your esl
you lost, sharteen, just admit it and go
>>
>>108735035
probably hikki and animefag millennials too
>>108734977
only normies and extra schizos "fall in love", the rest is just entertainment
>>
>>108735065
The only objective improvement is to refactor anything using that function to simply true if expected to always be true.
>>
File: 1777503464418393.png (16 KB, 703x117)
16 KB PNG
>>108735075
>I hate Gemishit so much, bros.
Im not just smelling ozone we are burning it
>>
File: 1756218968354291.png (759 KB, 1280x1082)
759 KB PNG
>>108735065
APOLOGIZE
>>
>>108730457
>grok
no code, no care
simple as
>>
>>108735107
Yep, americans dont like to accept it but chinese AI is actually less censored. In chinese AI the only thing that is censored is things in relation to the chinese government. In western AI everything related to society and humans is censored.
>>
>>108730457
They reduced my free limits for this?
>>
>>108730457
So it can't vibe code and can't do cunny RP...
>>
>>108735076
>Dahlit still coping and seething
KWAB
>>
>>108735065
kek
>>
it really is sharteen hours huh
>>
I use Gemini to verify canonical information about settings. It's good for that and means I don't have to constantly re-open CDisplayEx and find the chapter talking about whatever I'm looking for.
>>
>>108735138
>Completely unhyphenatedly his.
keep coping shartranny, all you can do is project over your esl slop
>>
>>108735138
yep
>>
File: 1748468508049204.png (443 KB, 1040x1644)
443 KB PNG
GLM bwos...
>>
>>108735156
Concession accepted, you lost tranny shill
>>
File: 1775024087797299.jpg (81 KB, 640x785)
81 KB JPG
You chuds recommended me Gemini for basic usage so I signed up and turned out to be rubbish. I just want it to be reasonably accurate for basic tasks, not coding based just everyday queries.
>>
>>108735197
>basic tasks
Use Grok, DeepSeek or Claude.
>>
>>108735159
GLM is so overhyped. It's the most positive, claude-slopped model I've ever seen
>>
>>108735197
>so I signed up
Never do this go to open router get $20 worth of credits then try ALL the models a little at a time and pick your favorite.
>>
>>108735211
For smut at price or cheaper deep seek is better right?
>>
>>108735197
wathever is free and available
>>
>>108735219
Unless you really like claude style, I see no reason to pick GLM over deepseek or kimi
>>
>>108735220
>wathever
the fuck anon. do you need an ambulance?
>>
>>108735231
>I see no reason to pick GLM over deepseek or kimi
I thought so thank you.
>>
>>108735231
true, kvngseek is kino
>>
To whoever made that story card thank you i tweaked the vulgar words out of it and threw it in AI studios instructions, im just genning power scaling slop it works pretty well but its short max story only 18 chapters. its about average webnovel quality. well worse because of gemini behaviors.
>>
>>108734946
After years of mystery and mythologizing on men as demi-god figures as their sexual desires bloom, an appropriately-aged boy is introduced to the girls' boarding school and embedded for a week or more. The girls stand in awe as an object of legend appears before them, the boy learns to assert his dominion while exploring tastes and preferences, and they all get to act on their every hormonal urge. The man discovers his dominion, the woman discovers her subservience, and both pair their revelation with the loss of their virginities. This is the crux of the Visitation’s formative power.

"My every word is law," he shouts.

"Your every word is law," they shout back.

"Your hearts belong to me," he screams.

"Our hearts belong to you," they moan back as one, the words coming at him from all sides.

"Your bodies belong to me." His voice still grows louder.

"Our bodies belong to you." Their breathing gets ever more ragged.

At last, as the anticipation becomes unbearable, he bellows the final ceremonial command:

"Serve me."

The Mrs. Principal, similarly as nude as reasonable, walks onstage briefly to hand him however many leashed collars he will need, and then departs. One by one, the boy calls the names of his chosen few, and they line up beside him. One by one, he collars them and kisses them deeply in a tight embrace. He goes down the line again, attaching their leashes, and they all share a final tandem kiss.

To cheers and applause, mixed with ecstatic moans and perhaps a few quiet sniffles and sobs, he leads his newly claimed women to his waiting ride home, and they depart.
>>
>>108735239
...yes
>>
>>108735138
he gets banned off /vg/ because the jannies there care so he has to come here on the off days
>>
Jerked off for two hours having sex with Bocchi on a school trip.
>>
>>108735681
log or larp
>>
>>108735678
>>108735678
>>108735678
>>108735678
>>108735678
>>
>>108732263
It's technically impressive but it has a very particular kind of slop look to it. Too bad. At least it's not nearly as piss filtered as the last one.
>>
>>108735763
I agree, but at the same time, it's also not something I expected image models to be able to do a year ago due to its sheer complexity (even though you can spot mutations and inconsistencies if you inspect it closer)
there's clear progress, though I imagine that it cheats a little since the paid version has an attached thinking model that does the heavy lifting in how to design the image
>>
desu
>>
boku
>>
>>108733941
It's a jew thing. Unluckily for us ashkenazi jews also have very high chance to develop schizophrenia, Dario will probably buddy up with Palantir sooner or later.



[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.