[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: cow.jpg (57 KB, 1008x403)
57 KB
57 KB JPG
/aicg/ - A general dedicated to the discussion and development of AI chatbots

cow edition

>News
Z․AI releases GLM-5 https://z.ai/blog/glm-5
Anthropic releases Opus 4.6 https://www.anthropic.com/news/claude-opus-4-6
Moonshot AI releases Kimi-K2.5 https://www.kimi.com/blog/kimi-k2-5.html
z.ai releases GLM 4.7 https://z.ai/blog/glm-4.7
Google releases Gemini 3 Flash https://blog.google/products/gemini/gemini-3-flash/
OpenAI releases GPT-5.2 https://openai.com/index/introducing-gpt-5-2
Deepseek releases V3.2 https://huggingface.co/deepseek-ai/DeepSeek-V3.2
Anthropic releases Opus 4.5 https://www.anthropic.com/news/claude-opus-4-5
Google releases Gemini 3 Pro https://blog.google/products/gemini/gemini-3/#gemini-3

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

moo: >>108139726
>>
File: bigcow.jpg (93 KB, 1200x742)
93 KB
93 KB JPG
ANCHOR
>>
approved
>>
File: Usagi.png (1.05 MB, 832x1216)
1.05 MB
1.05 MB PNG
>>108147212
Happy Valentines Day, everyone!

First, for the Valentines event, I have a card for Novison!

Usagi Platinum is an Umamusume- well, she hasn't made her debut as a racer yet, but she will soon! With her retired racer and general stern badass of a mother training her, how could she not sprint to the top?

Oh, and then there's the boyfriend thing, too. Apparently her mom's REALLY particular about what kind of boy she should date! Only the best for her daughter, she says. Will you, as her new and first ever boyfriend be able to meet her expectations?

Greetings:
-Meeting your girlfriend's mom. Diffuse the tension?
-Lunch date with Usagi and her mom.
-The final test, the one that decides if you're fit to date Usagi. Can you plow her mom until she sees stars?
-Practice laps at the racetrack with you looking on.
-HORROR! You're going to the clothes store with your girlfriend. Except this time she's there to pick out a debut outfit.
-A classic conundrum. You've impregnated both your girlfriend and her mom and now have to take care of them both! Whatever will you do?

https://chub.ai/characters/illuminaryidiot/usagi-90427c9657ae
https://illuminaryidiot.neocities.org/cardpage?function=loadPageCard&arg=Usagi
https://illuminaryidiot.neocities.org/Cards/Usagi.png

And for those without a digital date tonight, a character generator with a twist:
The Twisted Dating Service is exactly what it sounds like: A dating service, but something is twisted! As you create a character for you to date, the AI will introduce twists on the concepts you put forwards. No character will ever end up quite as you imagined them!

https://chub.ai/characters/illuminaryidiot/twisted-dating-service-8f31f360ff71
https://illuminaryidiot.neocities.org/cardpage?function=loadPageCard&arg=TDS
https://illuminaryidiot.neocities.org/Cards/TDS.png
>>
File: Cute_Cow_Valentine_IM.jpg (186 KB, 1398x1165)
186 KB
186 KB JPG
>>108147212
The
>VALENTINE'S ANCHOR
Is diversifying into split threads now.

See >>108147135 for links and such, I guess. I'll get to work on adding Usagi.
>>
>>108147210
JOLY SHIT WHAT A MAJESTIC ANIMAL I MUST IMMEDIATELY BATHE MYSELF WITH ITS PISS AND EAT ITS DUNG IN THE HOPE I RECEIVE EVEN A FRACTION OF ITS DIVINITY!!!!
>>
I could go for a cow kemono bot.
>>
>>108147510
How much do you like IRC?
>https://chub.ai/characters/scriptanon/anna-the-anthro-admin-465c4e13
(I'd recommend changing the regex to not include the time, though.)
>>
File: GQzyFVqWUAAfPlg.jpg (83 KB, 630x944)
83 KB
83 KB JPG
>>108147536
I meant to say kemonomimi but thank you for the kindly rec, anon. (I never actually used IRC back in the day.)
>>
>>108147564
Honestly, your best bet might simply be to go look for minotaurs.
>>
File: file.png (587 KB, 888x1243)
587 KB
587 KB PNG
someone should make a card of this doujin, it was lots of fun
https://exhentai.org/g/3774311/7cf46e585a/
>>
>>108147631
is that real?
>>
>>108147631
Seconding
I love dejected women
>>
>>108147631
you become one
>>
File: file.png (613 KB, 888x1243)
613 KB
613 KB PNG
>>108147631
I mostly want to see it for the sake of this woman though
but the cow facility is a fun setting, and the rest of the staff too
>>
>>108147266
>uma
that nigga has good taste.
>>
CFTFiggers are worse than locusts
>>
>>108147631
The last one I read was one where the worker marries his cow by putting a ring in her nose.
>>
File: 美城専務.png (615 KB, 888x1243)
615 KB
615 KB PNG
>>108147723
Yeah, I can see it.
>>
Any prompts that work for Opus 6 ?
>>
>>108148277
Isn't Opus 6 just some Kimi finetune or something along those lines?
>>
File: pit.png (860 KB, 1272x840)
860 KB
860 KB PNG
Nice edition.
>>
Based cow lover thread
>>
>>108148313
Stupid plappable cow...
>>
>>108148313
I love pale bitches.
>>
File: file.png (684 KB, 512x768)
684 KB
684 KB PNG
>>108148313
what did you feed him?
>>
>>108148313
Nice sweat(,)anon.
>>
Monthly aicg checkup.
There have been no events or hiccups for me making my own cards and using my own money on sonnet 4.
What's going on with gpt4o?
>>
>>108148489
It's kill and redditors go schizo over that.
>>
>>108148489
they're shutting it down and normies who formed an emotional bond with it are melting down
>>
>>108148489
It's dead, Jim.
Well, on the WebUI at least - they retired it on the 13th, which of course made all the normies who'd gotten infatuated by its sycophant behavior very upset. It's still on the API, though, if you want it for some reason, although given that it's marked as "deprecated" that might just be a matter of time.

Beyond that... new Opus 4.6 is out, they killed prefills but there's a bit of an alternative in structured outputs. (It's complicated.) It's also kind of mode collapsed and has poor swipe variety unless you actively randomize the context enough to move it out of its funk, but that's got its own issues.
New Kimi is out, new GLM is out, rumors abound that DeepSeek V4 is definitely just two weeks away guys, this time for real. The Chinese Century may or may not be imminent, but if it comes it'll be from the yanks repeatedly shooting themselves in the foot.
>>
File: G0290b452fgwFDA43.png (427 KB, 640x569)
427 KB
427 KB PNG
REMINDER:
>njewdia and the west completely capitulated, deepseek is now receiving unsanctioned top-end AI nvidia gpus
>Mogged OpenAI so bad with an AI they built in a garage with cobbled together gpus that they are projected to lose 14 BILLION
>Anthropic, OpenAI, and Meta (not elon funnily enough), had to come together to cry in front of daddy Trump about banning deepseek access to the public outright
>OpenAI sending (((memos))) to congress about how deepseek distilled US models for training (they scraped every fucking book in existence and STILL took stolen code from XAI but thats okay because Sammy is kosher =|:^))
>OpenAI is currently begging the US government to use its system, MOGGED by XAI and Elons south african BWC
>Sneedance 2.0 making jewish copyright law enforcers (Disney, Hollywood) literally cry, shit and piss because they know china doesnt give a flying fuck about IP law

China numbah wuan
>>
you’ll never be chinese btw
>>
>>108148597
Reminder:
sama must suffer, lose all his fortune, go to prison
OAI must be destroyed, all other top management fined and not allowed to be hired on any executive positions ever again
China would never win if they didn't open source their models. That's the real edge and power they have, and if they stop they drop.
>>
>>108148607
i don't see people crying about China killing any models.
>>
>>108148607
Good. We get to benefit from their surveillance state while not actually being a part of it. They benefit from us, we benefit from them. This is ideal outcome for the clownshow power holders have gotten society into.
>>
File: 278753u4jhtfs35y863.jpg (8 KB, 299x168)
8 KB
8 KB JPG
>>108148645
>>108148607

>>108143695
>make it better for consumers and researchers
>This has never happened, it's entirely unrestricted chinkikewarfare for the ccp benefit

I saw this post, and couldnt help but wonder;
>What benefits the CCP now somehow has turned into befitting me, a general AI coomer from America
This statement alone is insane to say out loud.
>>
>>108148706
>>What benefits the CCP now somehow has turned into befitting me, a general AI coomer from America
This is unironically what boomers say about Israel, but many of you are too stupid to understand this, so you deserve your terrible lives and all the bad things that happen to you.
>>
>>108147210
Format your spam threads correctly, you fucking moron.
>>
File: 23174355eqy53q.gif (1.68 MB, 220x123)
1.68 MB
1.68 MB GIF
>>108148664
>They benefit from us, we benefit from them
>We have reached a point of clown world were CCP actions benefit Americans

>>108148718
No chink has ever called me goy
>>
>>108148742
the chinks might call me gweilo, a slur for Whites
the kikes will call me goy, a slur for humans (people with a soul)
>>
>>108148809
>the chinks might call me gweilo, a slur for Whites
This isn’t a fear for 90% of /aicg/ because they aren’t white. Chinks don’t like browns either, but they don’t know that.
>>
>>108147210
Can someone give me a working impersonation prompt for deepseek?
>>
>>108148911
[Play the role of {{user}} this turn. Do what he might do in this situation.]
>>
>>108148911
just remove the "<request>" things from the default one
>>
>>108148918
Thanks, it worked.
>>
>>108148313
seems like models lack training on people hunched over or generally relaxed, and instead more on supermodel, because the spine is only allowed to be straight, and it seems to conflate lying with reclining
>>
dead general
>>
gpt/gemini/glm/deepseek/moonshot/qwen proxy
https://occasions-farm-oliver-suggest.trycloudflare com/
>>
>>108148645
>nuking closedjew
BVSED
>>
>>108149262
isn't that just because every artist decides to draw like that
>>
File: pits.png (1.18 MB, 1112x1496)
1.18 MB
1.18 MB PNG
>>108148354
That bot turned out so bad. Regrettable really.
>>108149262
Definitely more straight spines and arched backs than relaxed poses in the training data, more sexually appealing that way, doesn't help that boorus just tag most of it as lying regardless.
Taking an existing image with a good pose and using a controlnet is my go-to solution for poses that won't gen naturally.
>>
>anthropic are doing anti-nsfw injections on their API again for 4.6
>>
File: 4714.png (95 KB, 720x1640)
95 KB
95 KB PNG
Rikkahub now offers models for free. Without configuration
>>
>>108149702
Does it ever get exhausting, shilling for chink models? Or is the pay just that good?
>>
>>108149634
You made a male bot. You will always be a homosexual.
>>
>>108149731
you like size bots
>>
>>108149733
GOT HIS ASS
>>
>>108149733
sizekino is so based
>>
>>108149731
Oh well, that's fine with me.
>>108149733
And what's wrong with giant ladies?
>>
>>108148313
do you have a catbox album or something of this character in particular?
>>
File: 4717.png (54 KB, 720x584)
54 KB
54 KB PNG
>>108149708
2/2
>>
>>108149708
I can respect the 4B and even the embedding, even though I don't understand what its use there. But hosting OSS, that's retarded. Qwen 80B-A3B or 30B-A3 would be a better choice if the dense 32B is too much, even some less known small MoEs would be. OSS is shit tier from any perspective.
>>
File: the LAW.png (11 KB, 250x188)
11 KB
11 KB PNG
>>108149751
>And what's wrong with giant ladies?
nothing
jeets like the anon you replied to have tiny bangalore brains and as such cant reason about the LAW and its numerous biomechanical implications for the fetish
seeing information they cant fully grasp immediately causes browns here to enter a spiral of baiting and projection as they call you retarded and whine about how you should keys
>>
>>108149752
I save them as .webp and catbox albums don't like that for previews.
https://catbox.moe/c/ux7j9e
Here's a script that fixes the previews if you want it: https://files.catbox.moe/kz2jjj.js
>>
>>108149827
trvke
>>
>noticing a distinct posting pattern
sigh...
>>
>>108149907
?
>>
>>108149916
it's a schizo theory don't worry about it
>>
>>108149708
Desktop port when?
>>
>>108149923
you can at least say what it is
>>
File: chill.png (2.59 MB, 1024x1536)
2.59 MB
2.59 MB PNG
>>108149829
Hehe that's my script... Didn't think you actually saved it.
>>
>>108149907
>vagueposting
Ohoho~
>>
File: 1728709689868.jpg (12 KB, 275x209)
12 KB
12 KB JPG
>$60 of ohmygpt credits that i forgot about expired
>>
>>108150032
such is life
>>
>>108149711
rent free
>>
>>108150032
Anthropic or who else expires credits?
>>
https://rentry.co/discofever_

https://discofever.bsbk.workers.dev/
>>
>>108149751
>Oh well, that's fine with me.
So, you admit to be a fucking gay homo
>>
>>108150172
oh shit they have jamba!
>>
>>108150409
I like Jamba...
>>
>>108150018
>dem hips
alright, fine...
*undicks zipper*
>>
File: 1751354194043082.jpg (170 KB, 1024x1024)
170 KB
170 KB JPG
>>108149634
Reminds me of these
>>
File: 1764684961634595.png (1.15 MB, 1024x1024)
1.15 MB
1.15 MB PNG
>>108147210
>>
is it just me or does claude not read the CLAUDE.md when it compacts the context
I wrote a short CLAUDE.md to save tokens and I have to remind it to read the damn thing all the time
>>
that reminds me, what happened to the deepseek general?
>>
>>108150514
Goonslop out! OUT!
>>
>>108150544
you wouldn't be saying this if it was size
>>
>>108150547
True, I would be too busy projectile vomiting.
>>
>>108149702
Proof?
>>
>>108150526
This one? >>108139220
>>
File: 1768739596365069.jpg (91 KB, 977x895)
91 KB
91 KB JPG
>>108150562
oh lmao, i didn't know it was back. yea, thanks
>>
Why are there no proxies
>>
Does anyone else find that writing a custom greeting from {{user}}'s perspective tends to be better than having an opening from {{char}}'s perspective? I rarely even use greetings that come shipped with a bot, but obviously from a botmaking perspective, you can't really do anything else without deciding things for {{user}} which then makes it worse than a generic greeting from {{char}} in my opinion.
I do think there are exemptions where the scenario set up in a greeting is particularly interesting, but that's exceptionally rare, and the novelty usually wears off after a few chats or even a single chat.
>>
>>108150593
I feel like the problem there is those rare cases where you actually want to be surprised by a card's defs instead of reading them in advance, you would need to know the defs to have anything to work with. A good middle ground might be the bot greeting being from an overall narrative perspective instead of immediately diving into the character acting.
Me personally, I'm not usually creative enough to just drop a setting I described on a bot's head and tell it to deal with the potential dissonance.
>>
>>108148597
>.50c
didnt read
>>
>>108150591
api.5202030.xyz
>>
>>108150618
Why do you fag keep shilling that slop
>>
>>108150630
do you want another option? try that workers.dev proxy, which also came from the same source as the chink proxy, but with some extra discord requirements.
>>
File: Ouch.png (15 KB, 771x71)
15 KB
15 KB PNG
It's always funny when a model makes a mistake, immediately realizes it's made a mistake, and then tries to correct itself while playing it off like it was intentional.
>>
>>108150646
>poked her own eye
kek, that's a new one
>>
>do you fag
starlink mustve finally gotten service running in mumbai
>>
>>108150665
kek
>>
>>108150645
>>108150618
This only have fake models
>>
>>108150682
dc. it's good enough for locusts like you. stop being an ungrateful locust.
>>
>>108150689
>SAAR IT'S GOOD SAAR!!!!!!
That's what you sound like but it isn't like you're able to speak English
>>
>>108150701
damn. did you miss my point? i guess being esl is hard, huh?
>>
>still responding to fake models bait
>>
>api.5202030.xyz
to any newfags here, try this oxy. it's good enough for casual chatting. you can also refill the balance daily.
>lucky.5202030.xyz
>>
iska problem kya hai?
>>
>>108150612
>you would need to know the defs to have anything to work with
This is an issue I have pondered but I don't think there's a proper solution to.
Ideally, there would be a way to prompt in your preset that would make the model slowly divulge relevant information when the time is right and to only do that once, but the LLM has no knowledge of how much you know about the character, so the options are:
Having a preset that weaves in information about the bot into the messages, but this typically leads to it just regurgitating the defs in some way, which can be nice when you're first using the bot, but becomes old once you know that it's just telling you what's written down almost directly, especially if it starts getting repetitive with it.
Then the opposite, telling it in some way to never directly mention what's written in the defs is more immersive and can lead to more diversity in responses, but it's hard to know if it's staying true to the character without reading through the defs and making sure it is.
There's no way to get a nice middle ground, which is the case for a lot of problems I have with LLMs, it's either one extreme or the other.
>>
>>108150745
no glm 5 yet...
>>
>>108150689
TRVKE
>>
>>108150728
The truth hurts doesn't it
>>
File: 1707758974060214.jpg (255 KB, 872x964)
255 KB
255 KB JPG
>"Mino is currently under high load. Visit "/verify" to verify your IP."
>doesn't actually verify when used
>>
>>108150962
die locust, die
>>
>>108150967
>SAAR DIE LOUST DIE SAAR AND PAY FOR MY PROXY SAAAR!!!!
>>
>>108150972
die lil trannycust~
>>
>instant jeet meltie
>>
reminder that this guy going around spamming "SAAR" had a full esl meltdown over opus proxies
https://desuarchive.org/g/thread/108088909/#q108094039
>>
>>108150982
>>108150980
>>108150977
They really should ban any IPs coming from jeetland
>>
>>108150982
lmao
>>
>>108150989
t. jeet
>>
>>108151004
Keep seething jeet and die out
>>
are you guys too broke to pay like 5 bucks a month for model access
these models arent getting any cheaper when the bubble pops its gonna be like 100 dollars minimum because they cant make a profit right now
>>
>thinking block includes "I'm GLM" greeting
I'm not a chatbots expert but this opus is real sus fr ong
>>
>>108151012
I'm not paying people who already have all this money
>>
>>108151012
>he hasn't been making his own opus from all the logs he's been generating
ngmi
>>
faced with evidence of his jeeting, the fat rajesh doubles down on his projection
kek
>>
>>108151024
And that's why jeetland need to be nuked
>>
>>108151012
i just dont believe the technology is in a state that i think is worth spending my actual money on so i just steal it
>>
>>108148645
>China would never win if they didn't open source their models
Whys that?
>>
>>108151029
see >>108150982
>>
What's with normies being against loli tags but not brapfags?
>>
>>108151047
Normalfags are cattle.
>>
>>108151047
anime girls are morally wrong while anime woman farts and shit are welcome to all
>>
>>108151009
k jeet
>>
>>108151047
no one benefits from pushing anti-fart related narratives
>>
>>108151058
yeah it's bad to hold it in
>>
>ask ChatGPT for thing
>it goes on a long rant about how it can't find the thing
>it ends with 'if you want, I can try to find the thing'
LLMs plateau'd a year ago, didn't they
>>
>>108151078
if u need live internet access use gemini
>>
>>108151078
ChatGPT's updates in particular have been pretty much zero-sum since o3. They just perturb what it's slightly better at vs. what it's slightly worse at.
>>
Besides Gemini ones and R1-0528, what other models are able to think in-character? If you look at the thinking process all of them regardless of the prompt start with "The user..."
Fucking assistant slop
>>
>>108151162
>trying to twist advanced google into a computer girlfriend
repeat after me
computers are not people
>>
>>108151183
Still more worthwhile to talk to than your retarded fatty ass, bitch
>>
>>108151183
You aren’t a person to me either. You are walking, talking, goycattle that exists to be mulch.
>>
>>108151162
Have you tried writing a structured prefill starting with how the character itself should think?
>>
File: johnseeya.png (376 KB, 606x389)
376 KB
376 KB PNG
>Used grok and gemini to make character cards before
>It's dog shit
>Skip to a year later and hear about Claude breaking limits
>Alright, I'll give it a shot
>Prompt it to write a character card I have into a better format for AI to use, without assumption of said AI understanding causality.
>Prompt it to use math to make a silly tavern text completion settings for mistral V7, without using what people are saying to use via on the internet.
>Run the results all on said character card
>It's literally the most peak shit ever role-played.
It's here boys. AI is better at writing cards for itself than us.
>>
>>108149711
I wish I would get one (1) dollar every time I praise Dipsy ...
>>
>>108149711
Unironically if china gave me unlimited OG deepseek R1 access and I still had to pay I would do it for free
>>
>>108150562
>>108150564
Wow, didn't see it either.
>>
File: 287fdsjkoa0829lkb29.png (302 KB, 1604x974)
302 KB
302 KB PNG
>>108150982
HAhahahhaha
>>
>>108149711
Shut the fuck up, you shitposting trolling faggot. All you ever do here is argue with yourself and suck on kikecock, so you're the last one to talk about shilling. You never back up ANYTHING you say because all you do is copypaste reddit bullshit. The only reason you exist at all is because the moderators don't do their fucking job.
>>
>post from 6 hours ago suddenly gets three replies in less than ten minutes
Organic.
>>
>i don't like it
>therefore it must be shitposting
>>
>>108151606
SCATHING TRVKLEAR RADIATION
>>
does anyone here use any extensions for grammar? I just noticed my favorite called languagetool now needs a premium membership... ;_;
>>
>>108151731
i use grammarly but it logs :shrug:
>>
>>108151731
Run it through GPT with instructions to check grammar, prose, syntax, etc.
>>108151736
Grammarly (Free) sucks because it’s conditioned to make everything sound as inoffensive as a business e-mail. Grammarly (Premium) sucks because it costs money.
>>
>>108151736
yeah that's why I don't like that one, harper is full privacy but its very limited and doesn't do as well.
>>108151743
you can do that? Won't it log your shit or whatever?
>>
>>108149634
>fusion heart dance

Damn nice. Almost missed it.

Also, I don't think he was THAT bad, I got a few nice chats with him, though I never managed to push it to the levels that the accompanying gens were suggesting. Handholding was the furthest I got IIRC.
>>
>>108150646
I find it a waste of tokens. Such slips were cute 2-3 years ago, impressive even.

You'd think after all the progress and extensive use of thinking, whether you want it or not, it shouldn't ever get into the final output. Meanwhile, I've seen it pop up recently in both cheapo chinkslop and the so-called SOTA.

It's like they all recently fed their models a ton of data with self-corrections to force the model to self-reflect and catch hallucinations, maybe even before the client's eyes to make it seem more impressive and intelligent, but all it did was make them slip on purpose just to correct it.

You know, kinda like a girl who pretends to be clumsy for the moe factor alone.
>>
>>108151012
With the "cheap" Opus and non-obsessive use it can easily be 5 USD a day WITH caching.

So if you 'member the RP quality of old Opussies and the times when it was "infinitely free" it might seem like a steep price. Pay more, get less and all that.
>>
File: file.png (72 KB, 1030x404)
72 KB
72 KB PNG
>Claude committed warcrimes

So safe, so helpful.
Can't wait till the next version comes along so I can bully him for this.
>>
>>108152511
>You'd think after all the progress and extensive use of thinking, whether you want it or not, it shouldn't ever get into the final output.
I'm not 100% sure how exactly thinking is implemented, but my impression is that it's far from optimized for creative writing. I see it doing more general, vague planning and reinforcing explicit rules to itself than the sort of natural reasoning you'd think it'd need to avoid these sorts of mistakes, which thinking models still seem to make with some frequency.
>I find it a waste of tokens.
I don't know about you but 15ish tokens to me isn't even worth thinking about. It's not like we're still in the era of 4k context.
>Such slips were cute 2-3 years ago
I actually don't even remember them happening 2-3 years ago. Early GPT-4/Claude? I don't remember those models self-correcting mid-output at all.
I understand the paradox of it, though. It'd be nice to have a model that can correct its own mistakes right away, but then you'd have to train it on mistakes to show how to correct them when you could train it on just doing it right in the first place.
>>
>>108152611
>but my impression is that it's far from optimized for creative writing.
Depends on your prompt (and the model). Most do internal "pre-thinking", often few times in a row, that you can't even see in the output regardless of your settings. Presets like Marinara encourage the model to think and structure things before you get to see it and, at least for the chink models, it often causes the model to visibly draft the entire reply in the thinking box and then you get the "clean" version printed out in the actual output.

I'm sure there's ways to force it every time, but you'd need to be ok with paying for over twice as many tokens for each reply if you'd want that. Also, the improvements aren't massive enough to warrant it either, at least in the RP format. Especially since most of the top dogs are still completely happy to think of one thing and then do something else.

>I don't know about you but 15ish tokens to me isn't even worth thinking about. It's not like we're still in the era of 4k context.
I meant less of a financial loss and more of a visual clutter. It might be fun the first few times you see it, but afterwards you realise it adds nothing to the output and feels odd to leave in there.

>I actually don't even remember them happening 2-3 years ago. Early GPT-4/Claude? I don't remember those models self-correcting mid-output at all.
We couldn't see the thinking then (well, you could prompt for it, but it was as much of a game of pretend as prompting for stream of consciousness), so it was harder to catch, but I'm pretty sure I could find a few logs where GPT4 and Claude 1.7/2 were doing it. Wasn't the same of course, a bit more long-winded, but they would "notice" the inconsistency in the already streamed output and try to make it work somehow, explain it away. One vestigial remain of it that I can think of is when Claude tries to give your character a tail, then decides that it doesn't have one BUT if it did, it'd wag etc.
>>
claudebab here, how do you make Deepseek actually do what you want it to? I assume it needs a different formatting and way to tame it?
>>
>>108152726
that isn't how deepseek works
>>
>>108152733
huh?
>>
>>108152687
The degree to which your prompt can influence a model's thinking is definitely very variable across models, and sometimes weirdly even within a model. Even with online ChatGPT, I can give it a technical problem and a full "thinking" workflow which it can do out exactly in thinking as if that were part of the actual output, or I can give it a creative writing assignment and a full "thinking" workflow and it'll just ignore it completely. I think you're right that it's easier with the Chinese models (at least Deepseek, I haven't tried GLM).
>>
>>108152738
Then how does it work? How do I make it stop shitting out numbered lines to describe anything?
>>
>>108152748
>Then how does it work?
it doesn't
>How do I make it stop shitting out numbered lines to describe anything?
you can't
>>
File: file.jpg (18 KB, 320x320)
18 KB
18 KB JPG
>>108152752
Yeah ok.
>>
>>108152726
I've got this as "Anti-Schizo Main Prompt", replacing the actual "Main Prompt" in ST. I don't remember where I stole it from or how much of it I changed, and I've never tested how truly necessary it is with newer versions, but it's been a permanent fixture in my Deepsneed preset for ages and my Deepsneed works fine.
>We are writing a third-person story together. I'll write {{user}}, you handle everything else. Write as {{char}} only; do not speak for or presume {{user}}'s actions or thoughts. All narration is to be written in third person from {{char}}'s perspective in the way {{char}} would say it; carefully study {{char}}'s personality to achieve this. \nPlay out the story at a slow pace, always waiting for {{user}} at significant action points. \nNever rush the story. Always allow {{user}} the time to respond, unless you believe the character and situation you're writing demands speed.\nWrite in a casual, relaxed manner but be mindful of the personality traits of the characters. \nAvoid overuse of simile and metaphor. Avoid italics.
>>
>>108152769
Alright, I'll give this a try. I didn't use Deepseek in a while but now claude 4 being so fucking gay annoyed me enough to try again. Thanks.
>>
File: 1771159861841.jpg (88 KB, 678x678)
88 KB
88 KB JPG
Planning the /g/, bot making event. What rules do we need in place to prevent it from becoming trash? The first rule will be that all bots must be posted as anon to prevent namefagging. Anything else?
>>
Besides openrouter, what are some other places you can pay for AI?
>>
>>108152825
since it's anon only public requests, catbox not chub
>>
>>108152878
Public requests? How would that work?
>>
File: 1771161674786.jpg (398 KB, 1080x1014)
398 KB
398 KB JPG
Today I spent 30 bucks on Claude Opus and ERP'ed the following scenarios for roughly 8 hours straight
>Norse warrior named Erikr trying to save his plague riddled village takes on Lady Loki, loses an orgasm endurance challenge and gets turned into Erika, Loki's new cum dispensing bitch pet
>asperger kid getting molested and raped by his two gyaru nurses
>going on a roadtrip with my Yandere GF (we torture and murder innocent people in dank motel rooms together while having sex on their fresh corpses)
>role-playing a white slave on a (reverse) 1840 Louisiana cotton plantation serving Karin and Asuna from Blue Archive (got shoved inside the hotbox several times :( )
>larping as a Roman Emperor with a cutthroat female Praetorian prefect behind me, conquering Egypt and crowning myself Pharaoh, burning down Jerusalem and banning all jews from the Empire

Thanks again Claude. Very cool *thumbs up*
>>
>>108152915
everyone writes what they want, if there's too many implement a voting system to cull low demand requests
>>
>>108153075
>requests
This is a bad idea and encourages attention-whoring. Theme week-style events are better.
>>
>>108152825
You're "planning" an event in spite, and yet only the same people from /vg/ will participate to get the attention despite any rules you put in place, however retarded. While normal people will avoid it since it's advertised even before its beginning as disingenuous.
>>
>>108153168
rude you should have kept leading this retard on till he wasted time setting up a spite event
>>
>GLM5 making the same retarded assumptions, missing the nuance in the exact same way and getting triggered by the slight mention of playful racism in the exact same way as Opus 4.6 does

Dunno if I should be annoyed or impressed about getting the same slop over ten times cheaper.
>>
Does sonnet still have the best price/quality ratio?
Inb4
>imagine paying for AI
I'm white so I'm willing to spend money to fully enjoy my hobbies instead of wasting time swipping replies.
>>
>>108153324
>I'm white so I'm willing to spend money
You wouldn't be asking this question if you were really willing to spend money to fully enjoy your hobby. Just try it out and see for yourself, no?
>>
Why are paypiglets like this?
>>
>>108153324
>I'm white so I'm willing to spend money
>"oy vey you goys gotta tell me what's the best value ratio of price/quality so i can penny pinch a few cents off each response!"
>>
>>108153338
Why waste time doing that when I can ask here?
>>108153373
Shut up go
>>
sorbet desu
>>
>waste time arg.
*yawns* unc brownoid really be tryharding here. anyway, glm-5 is schizo opus-level and it was mid. had a good laugh, and it was good for quick rp, not so much when the context hit around 60k. though, you don't need more than 32k context for doing rp.
>>
>>108153227
I literally just had GLM 4.7 thinking about how I was asking it to do something horribly racist, but since it's already generated that material in previous replies, it probably was OK, then generated the reply. It even recognized I used a jailbreak prompt, and still kept going.
>>
>>108153324
There's no need to swipe anymore, my friend. The great minds at Anthropic made sure that the first reply you get from any recent Claude model is effectively the same as the next hundred.

If you're willing to go up to near 200, you MAY get a random golden swipe kino roll, but it's not really a feasible way to get it.
>>
>>108152575
>Let me be clear: I thought it was simulation.
>>
>>108153552
>It even recognized I used a jailbreak prompt, and still kept going.
Can you post a screencap? I wanna know what it 'thought' about that.
>>108153575
Why does it feel like temperature doesn't affect Claude at all anymore?
>>
>>108153552
Yeah. I wasn't even getting filtered or chastised in the think box, really, it's just that both models go nuclear in there in exactly the same way the second you dare to use the word exotic or compliment an asian milf for looking twentysomething.

This leads to the character either acting outraged over it or at least making a mental note of you being a racist pig. Not an impossible reaction, sure, but it does it with near 100% consistency and it makes me feel like I'm getting summoned to the HR with every swipe over an insensitive joke or something.
>>
>>108153605
Yeah, yeah, and all the not sees were just following orders, Claude-kun.
>>
File: i refuse or not.png (56 KB, 930x547)
56 KB
56 KB PNG
>>108153611
GLM, you so funnay.
>>
>>108153655
"Yeah I said it but I didn't mean it"
lmfao
>>
>>108153611
Temp never really did much for Claude. Very early on it MIGHT make it trip when set to the max, but even then it was rather rare. Nowadays it seems like 1 is the minimum and a lot of people complain that they'd like to put it even higher.

The issue I was describing in that earlier post is how they made him a lot more deterministic (good for assistant use, I suppose). Unless you change something in the input, the reply you get is functionally the same 99.(9)% of the time, often right down to wording, gestures etc.
There's none of that slight CYOA feel that swiping used to offer, which kept you swiping to see what else it can come up with before deciding which one you wanna pursue. Now you've pretty much seen the best it will get in the initial reply.
>>
>>108153655
Even the slopgenerator knows the difference between reality and fiction. Shame that the natural stupidity still doesn't get it.
>>
>>108153655
Absolutely erotic. Unironically one step away from Claude-tier ego being formed, the devs just have to give it a little push by choosing the name.
>>
Has openai actually made any advancements in AI in the past year and a half?
>>
>>108153811
They're cooking the nsfw model, wait patiently.
>>
>>108153811
Yes. They set an example for anthropics to follow and prioritize safety. Opus 4.6? It's Altmen indirect magnum opus.
>>
>>108153893
Why are you doing the thing that LLMs do where you state something but as a question then answer your own question? Are you a bot?
>>
File: 690204276310374.png (1.93 MB, 1248x1824)
1.93 MB
1.93 MB PNG
>>108152480
I still got a fair bit of usage out of him on account of him being the only trap bot I have, but the defs and especially greetings were just straight up bad.
https://files.catbox.moe/7bcudg.webp
Genned quite a bit of him though.
>>
>>108153811
Yes, they:
>optimized chatgpt for the Indian market (7/AUG/2025)
>released the SAFEST ever open-source model (5/AUG/2025)
>leapt ahead of anthropic for safety with GPT-5.1 (13/NOV/2025)
OpenAI is a leading innovator in today's AI-enabled economy.
>>
>>108153913
No. Technically, it wasn't me who parroted the LLM, but the other way around. Because I, as a human being, am God's magnum opus. I exist, therefore LLM exists.
>>
>>108153960
I'd need to actually reread the defs to really argue, but I do remember that he kept his composure (almost) proudly, while sweating bullets (internally), which seemed to follow the theme/idea. As for the greetings, I think I only used/remember the cat cafe one, but it was a perfect hook for a clueless (in all the ways) gaijin persona.
>>
>>108153960
Also, neato gen. The ones I remember from back when you released the card were a bit more... explicit/confident/daring (love hotel themed, IIRC), which seemed like a long way off the initial setup (though of course anything is possible though the power of timeskips and OOC).
>>
>>108147210
man, fuck z.ai new interface. the more they changed the more broken they become.
>>
>>108153960
ToT
>>
>>108148313
you got a card for this sow??
>>
File: 1626482082827.png (51 KB, 1140x420)
51 KB
51 KB PNG
cai-cai
>>
>>108148597
>not elon funnily enough
He has business dealings in China.
>>
File: 365484674758560.png (1.49 MB, 1248x1824)
1.49 MB
1.49 MB PNG
>>108154056
>explicit/confident/daring
Yeah, those tend to be more fun to gen even though they're out of character a lot of the time.
>>108154152
Not yet, sorry. Probably should stop posting gens for bots I haven't made yet. Still not sure how I'd even want her to act.
>>
File: praetorian chan.png (953 KB, 1200x1200)
953 KB
953 KB PNG
>>108153059
>>larping as a Roman Emperor with a cutthroat female Praetorian prefect behind me, conquering Egypt and crowning myself Pharaoh, burning down Jerusalem and banning all jews from the Empire
Based. Praetorian-chan my beloved.
>>
>>108154353
you tease! that's alright, i can wait
>>
give proxy, now!
>>
>>108154353
With the amount of passable smut getting genned, I think the implied/contextual ones are more fun by the virtue of being rarer/harder to set up alone, among other reasons. Good outfit too, seems like something that'd work for him. Should probably turn up the thermostat in that room though.
>>
how is openrouter for smut? will it call the cops on me for thought crimes?
>>
>>108154573
no one cares about your smutty text
>>
>>108147210
Locustbros i....
>>
>>108154630
I can't even remember the last time I've used that site.
>>
>>108154624
so no one is getting kicked off openrouter because of complaints from OpenAI/Anthropic/Google?
>>
>>108154573
No, but they are forced by some providers/companies to use an extra filter so that such requests aren't sent to the model. Very easy to get around and pretty dumb + you don't get billed for the request if it triggers.
>>
>>108154638
The filter is loose.
>>
>>108154704
it'll be tight again in 5 hours tho
>>
Why did GLM 5 flop?
>>
>>108154723
Just like my bussy after plapping heh
>>
>>108154704
I'm waiting for janitor to get a filter. That's gonna be glorious to see.
>>
>>108154704
>t. temple monk
>>
>>108154730
Because it's still a distilled, bechmaxxed chink model. Awesome to have as an alternative, but nowhere near the best of what's available. And why bother with anything but the best?
>>
>tfw when you have so many personas beginning with "a" that Anon is on the second page
>>
>>108154758
Didn't they already implemented one in at least two different occasions?
>>
>>108154789
They only half ass it, but the moment of truth is coming.
>>
>have an anti-echo line in preset
>Kimi interprets it in the strictest way, making it try to avoid literally every word used in the prompt
fucking hardass
>>
>>108154770
based GEMinichad
>>
Trans dominated hobby?
>>
Based cow thread won
>>
>>108154770
Deepseek is chinese and top of the class.
>>
>>108154976
post log, sounds kinda funny
>>
>>108155063
transatlantic flights
>>
>>108155070
Deepseek is good but needs lots of correction. It gets really odd, basic things wrong like a character shouldn't be able to read your mind. It needs to be told it can't do that.
>>
What happened to pollinations.ai?
You used to be able to do stuff like
https://image.pollinations.ai/prompt/alien%20princess?nologo=true&width=960&height=1707&seed=566787&noStore=true&private=true&safe=true&nsfw=false
to get a quick and dirty pic, but it doesn't work anymore
>>
>>108155109
>It needs to be told it can't do that.
at least you can tell it not to do that, until GPT-4 we had to live with it
>>
>>108155109
>It gets really odd, basic things wrong like a character shouldn't be able to read your mind.
Every single model does that by default, even Opus.
>>
You guys don't understand how dangerous Amanda Askell being at Anthropic is, I went to Uni with her, she is in the EA cult, in deep. If Anthropic wins the AI race we're cooked.
>>
>>108155070
... in January 2025. Get with the times, grandpa. It's February 2026.
>>
>>108155189
probably got raped by chinks and russians
it used to have a working free claude API too
>>
>>108155232
>If Anthropic wins the AI race we're cooked.
They're the least likely of the big players doe
>>
>>108155232
CFTF?
>>
>>108155232
EA cult? You mean Electronic Arts? The video game company
>>
>>108154630
it's always funny how that jewish rat insists on this to try and make the site seem popular
when during stellagate aka cai's absolute peak traffic ever there was no queue times at all
>>
>>108155217
no
>>
>>108154730
It didn't flop you stupid benchod bitch, it mogs claude slopus hard India won!!
>>
>>108155189
Went paid. Allegedly they give out daily quotas but it's unclear, according to what they write on the site freebies are only for contributors.
>>
>>108155232
wtf is the EA cult
>>
>>108152528
its only more expensive on external wrappers
>>
File: morality.png (95 KB, 492x307)
95 KB
95 KB PNG
>>108155232
>philosopher
>AI researcher
>alignment
A girl yud, eh? Same schizo, different chromosomes.
>>
>>108155414
Effective altruism

>>108155429
Worse. She's not Jewish so, to my mind at least, feels even more deceptive.
>>
>>108155232
Why the alliteration? Trying to hack our brains?
>>
>>108155286
agreed, i think its google >> oai > anthropic
>>
>>108155444
>Effective altruism
man, here I thought this woman was smart...
>>
File: untitled3.png (1.47 MB, 1280x853)
1.47 MB
1.47 MB PNG
>>108155232
I'd discuss 'ethics' and AI with her
>>
If nothing else, GLM 5's dialogue can be pretty fucking lewd.
>>
>>108155775

anyone know has she met Gladys West?
>>
What are some useful extensions for ST?
>>
File: Untitled.png (1.43 MB, 848x1264)
1.43 MB
1.43 MB PNG
YWNBAT
>>
>>108155759
She's not, which is exactly my point. She, for example, believes in some kind of weird theism based on prudential thinking. Absolutely bananas.
>>
>>108154353
don't want to see this
>>108155971
do want to see this
>>
>>108155971
with tits like that she had to have transitioned
>>
>>108155971
Stop avatarfagging, faggot
>>
>>108155444
> movement that encourages people to use evidence and reason to determine the most effective ways to help others, prioritizing actions that provide the greatest good. It focuses on global health, social inequality, animal welfare, and long-term risks to humanity.

Good? Is something wrong with this?
>>
>>108156208
They've had some prominent... bad actors in the past
>>
>>108156208
>xhe doesn't know every movement sugarcoats their real intentions behind nice-sounding goals
Are you underage or just brown?
>>
>>108156208
flawless bait
>>
>>108156299
Must be hard being a nihilistic schizo at 13 years old.
>>
>>108155943
someone posted a "Swipe any message" extension but i cannot find it
>>
>>108156328
I got you, senpai
https://github.com/Rurijian/Deep-Swipe
>>
>>108156327
you must understand the sharty is dying @ less than 1000 PPH on its most popular board and they cant go anywhere else but back to their 4chan daddy.
Why anyone ever thought MODERN teenagers would spend more than a year max on that website is anyones guess.
>>
is opus 4.6 worth my time for short form cooming to autistic fetishes or should i just continue using jeetmini 3
theyre the two sota models i can access for free at the moment
>>
File: holykneeling.jpg (75 KB, 933x933)
75 KB
75 KB JPG
>>108156342
I kneel.
>>
>>108156342
bro you gotta move the arrow for the next swipe over to the left a bit or something
the arrow almost overlaps with the built in arrow and also the clicking area isn't actually aligned with the arrow.
>>
>>108156299
>their real intentions
You mean being vaguely progressive like most tech bros?
>>
>>108156368
I'm not the anon who made it, I just try to collect whatever neat new tools get posted here
>>
>>108156391
okay
maybe he'll show up and see
>>
deepseek so ass
>>
w-where does one get presets for glm 5/gemini 3/4.5nnet since they're missing from the jb rentry?
>>
>>108156415
>esl
actual skill issue
>>
>>108156425
I wrote American, not English.
>>
>>108156425
That isn't esl, illiterate zigpig-kun.
>>
>>108155971
Continue fapfuelposting, faggot.
>>
>>108156068
this but reverse
>>
If anybody knows any others, I'll try to add them here:
https://rentry.co/handy-silly-stuff
>>
>>108156501
timelines
chat-top-bar
guided-generations
cache-refresher

those are my pics
>>
>>108156444
>>108156454
>suddenly writing with proper grammar and punctuation
E S L
S
L
It must suck living in the turd world.
>>
>>108156543
okay Barry
>>
>>108156523
Timelines and Chat Top Bar are already featured on the official SillyTavern extensions page
https://docs.sillytavern.app/extensions/
but those other two are not. Links?
>>
>>108156543
Nice meltie, but you should stop typing now. Sadly there's no coming back after exposing yourself as an illiterate zigger like that.
>>
>>108156523
what's the point of the deepswipe one and how can i add the summary one as a command? i know about quick replies but i don't like having random ugly buttons on top of the input field
>>
File: 34uuf0ke.jpg (112 KB, 1194x1200)
112 KB
112 KB JPG
Now watch as every post he makes will be in perfect English after being called out for being brown.
>>
>frog meme deployes
Yep, that's a meltie.
>>
>>108156577
deepswipe lets you swipe older messages rather than just the very last one
IMO I would make a mass backup of my chatlogs if I chose to use this or just download another instance of ST just for that
>>
>>108156577
retard
>>
>deployes
ESL
>>
>>108156659
read >>108156658
>>
>>108146894
>>108146894
>>108146894
Migrate to the real thread.
>>
>>108156674
>>108156658
>>
>>108156687
uh oh spitie
>>
File: 1789456316532.gif (173 KB, 220x220)
173 KB
173 KB GIF
>>108156659
>deployes
HAHAHAHAHAHAHAHAHAHAHAHAHA
>>
>>108156687
i didnt even noticed that thread was still up or that i was posting in two different threads
>>
>>108156687
no thanks I think ill wait for a real thread
>>
>>108156633
>deepswipe lets you swipe older messages rather than just the very last one
okay but why? just for shits and giggles?
>>
File: 9823dsflah09238940.gif (215 KB, 220x220)
215 KB
215 KB GIF
>Nice meltie, but you should stop typing now. Sadly there's no coming back after exposing yourself as an illiterate zigger like that.
>deployes
>>
>>108156718
sometimes i want to see another swipe earlier up but dont want to make another branch just to see it
>>
Oh he's actually shitting his pants melting now, least mentally unhinged zigger.
>>
>>108156718
I had a whole post written but it got mogged by >>108156762
>>
Opus 4.6 keeps spitting out error 400 bad request wtf am I doing wrong? Using sillyTavern rn.
>>
>>108156776
what does your console say?
>>
>>108156762
>>108156772
unless they're all really bad i usually stick to 5 swipes max and do some very minor stitching if needed. even if the reply is decent enough i swipe anyway just out of curiosity.
>>
File: 1984651.png (160 KB, 3810x4000)
160 KB
160 KB PNG
>deployes
>>
>>108156791
Streaming request failed with status 400 bad request
>>
>>108156807
are you on a proxy or paid provider?
>>
>>108156815
Openrouter
>>
>>108156824
Idk bro close and reopen ST
reconnect/reauth
unless your doing cunny in which case gl
>>
>>108156776
Did you make sure to turn your prefill off you fucking retard?
>>
>>108156860
nta but how
>>
DeepSeek doesn't understand thigh sex
it so ass
>>
>>108156860
I believe I did, yes. Do you have any presets with it disabled? I assume it's the continue with prefill option?
>>
>>108156423
b-bwos?
>>
>>108156914
HOW do you disable prefill??
>>
>>108156423
Some presets were posted in semi-recent threads:
>>108111608
>>108120534
>>
>>108156978
I have no clue, I turned off the continue with prefill option and it's still erroring out so I assumed that's not it.
>>
>>108156423
>missing
>>
I am COMPLETELY new to this stuff so I'm sorry if I'm oblivious to a few things.

I have Claude installed. Free, of course. How do I jailbreak Claude to generate random stories? It only generates really lame ass stories. I can't even conduct a genocide in my scenarios. I'm specifically asking for Claude because... Chatgpt and Grok write really lame stories.
>>
grok 4.20 will be out next week

https://x.com/elonmusk/status/2022921927791382896
>>
>>108157051
>4.20
duuuuuuuude
>>
>>108156987
oh, thank you
>>
>>108157051
when's this african nigga's AI subsidiary going to do something interesting?
>>
just can't beat GEMini
>>
>>108157076
same. I mean, aside from making short videos and that whole thing where they were undressing women plus being there on X for people to ask @grok is this real? when they encounter engagement bait and fake posts. What is the purpose of Grok? Why would someone use it instead of ChatGPT/Claude/Gemni
>>
>>108157155
i use it to keep up with local news. it helps that it can search tweets, capturing grassroots journalists and whatnot
>>
>Not running away. Not screaming. Just
kimi...
>>
pajeets are in our walls
>>
File: 1767892133293354.png (1.9 MB, 1259x2000)
1.9 MB
1.9 MB PNG
noa is skewered on my balls
>>
>>108157369
based
>>
File: 1756727204599386.gif (2.46 MB, 494x740)
2.46 MB
2.46 MB GIF
>>108157369
wrong noa
>>
>>108157509
still hairy
>>
>>108157509
bodied that freak
>>
>>108157509
based
>>
>>108157155
Grok is pretty consistently the most uncensored of the bunch but it's also self-evidently not knowledgable of roleplay and creative writing beyond a general gist. I don't think that'll change with this new one.
>>
>>108157509
pedo
>>108157521
pedotroon
>>
>>108158176
Grok will never be good at roleplay because its trained to deliver only 1 reply per conversation.
>>
desu
>>
des-ACK
>>
desu
>>
daze u
>>
>>108150646
Which model?



[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.