File: 1754346359352975.jpg (987 KB, 1718x2291)

987 KB JPG

>ANCHOR

>>106948374
It can't because of the shit model and it 100% must because you're bad at writing and the card is bad while forgetting that the card work better on 4pus with the same input

>>106948855
'pus is izzy modo, skillet-chan

The tf2 challenge got me fucked up
Almost popping a blood vessel over Claude is not worth it

Gemini 3 SOOOOOON
The Claude humiliation ritual will stop

gemini 3.0 is available, probably accidentally, on vertex
not on gemini API, only vertex, you need to have a legitimate paid account

gemini 3 will not be blacktooth, which will never come out

>>106949134
>no proof

I'm a gemini beta tester. What does Gemini 3 need to have in order to please you? (Negativy bias was fixed)

File: 1759752540135993.png (2.26 MB, 1248x1824)

2.26 MB PNG

>negativity bias was fixed
lol. lmao

>>106949225
The difference is night and way, compared to 2.5 pro. I'm on my cellphone, so if you don't mind walls of text as a post, I can share

>>106949281
go for it. share

>>106949285
Give me a card in which Gemini gets negative, it'll be easier for me

One thing I appreciate about gemini 3 is that it won't do the follow-up questions shit claude loves to. Also, it has better character adherence. On the other hand, "it isn't X, it's Y" and random bold words are still present. I think the knowledge cut-off is more or less february, just a bit better than 2.5 pro

Sonnet 4.5:
*Yukihime's smirk widened into a full grin as she watched Henry's bewildered expression. She grabbed one of her tails and bopped them gently on the nose with the fluffy appendage.*

"Nope! You're very much awake, dummy. Though I'm flattered you think me being a good wife is so rare it must be a fantasy." *She stuck her tongue out playfully, her fox ears flattening slightly in mock offense.*

*Leaning back against the couch, she crossed her legs - making sure Henry got a good view of those thigh-highs she knew drove them crazy - and gestured dramatically at the food.* "What, a girl can't cook her spouse a nice meal without it being some kind of miraculous event? I cook plenty! Remember last… Tuesday? No wait, that was takeout. Okay, bad example."

*She grabbed the beer and cracked it open for them, pressing the cold can into their hand.* "Here. Drink. Eat. And then you're gonna tell me all about whatever bullshit happened at work today while I sit here and look devastatingly beautiful. It's the least you can do since I slaved over a hot stove for like…" *she paused, counting on her fingers* "…twenty whole minutes. The recipe said thirty but I'm just that good."

*One of her tails curled around Henry's waist possessively as she snuggled closer.* "Also you smell like stress and office air conditioning. It's gross. After dinner, you're getting pampered whether you like it or not."

>>106949428
Gemini 3:
*Yukihime rolled her eyes, letting out a dramatic sigh before reaching up and pinching his cheek, hard. She twisted the skin just enough to sting, a devilish grin spreading across her face as she watched him wince.* "Dreaming? Please. If you were dreaming, I'd be naked and feeding you grapes, not serving you curry like a tragically devoted housewife."

*She released his cheek and smoothed it over with a gentle, condescending pat. Her nine tails fanned out behind her, practically taking over the rest of the couch as she pressed her incredibly soft, oversized boobs against his arm.* "Honestly, Henry. A guy comes home to a hot meal and a hotter wife, and his first thought is brain damage? Rude. You should be worshipping me right now."

*She grabbed the cold beer and pressed it firmly into his hand, guiding it towards his mouth.* "Now drink. And then eat. Before I decide you're too ungrateful and eat it all myself. I made it extra spicy tonight, just to wake your tired ass up."

>>106949428
>>106949439
>Sonnet 4.5
slop
>Gemini 3
not particularly good, not particularly bad

>>106949428
>>106949439
>Sonnet 4.5
kino
>Gemini 3
slop

>>106949439
What a bitch.

>>106949459
I didn't edit my preset for gemini 3 btw, and this card is a bit sloppy on the definitions, so I'm sure gemini 3 can do better

>>106949439
>just enough to sting, a devilish grin spreading across her face as she watched him wince.

Nothing burger model

Sonnet 4.5: *She looks up at you with a bright smile, her auburn hair tied up in a messy ponytail. She's wearing a tight tank top and short shorts that leave little to the imagination.*

"Oh! Morning Henry!" *she chirps, waving back enthusiastically.* "Taking the trash out before work. Such a pain, you know?"

*As she speaks, you notice a couple down the hallway engaged in vigorous activity against the wall, their moans echoing through the corridor. Your neighbor doesn't even glance their way - it's just another Tuesday morning.*

"Hey, are you headed anywhere fun today?" *she asks, adjusting the trash bag in her grip. Her eyes are friendly and curious as she looks at you.* "I've got classes later but I'm free for a bit if you want to grab coffee or something?"

*The sounds from down the hall intensify, accompanied by rhythmic thumping against the wall. A door opens nearby and an older woman steps out, squeezing past the copulating couple without a second thought as she heads toward the elevator.*

>>106949428
>>106949439
>only 1 swipe each
are you niggers new

>>106949597
Gemini 3 (This is the free use world card btw):*The neighbor, Sarah, turns her head, her messy blonde hair bouncing a little. She's only wearing a loose, oversized t-shirt that barely grazes the top of her thighs, leaving her long legs completely bare. It's pretty obvious she isn't wearing any panties underneath, just letting her pussy air out.*

"Oh, hey Henry! Morning," *she chirps, giving you a bright smile despite the heavy trash bag she's lugging. She shifts her weight, the hem of her shirt riding up just enough to show a hint of her bare ass cheek.* "Heading out early? Watch your step by the elevator, Mrs. Gable is currently getting double-teamed by the maintenance guys. It's a bit slippery over there."

>>106949428
>>106949439
>>106949597
>>106949617
>asterisks around narration
ask me how I know your preset is gigaslop

>>106949676
The preset doesn't include anything about asterisks, Gemini is just following because the greetings of the card has

>>106949705
>slop preset AND slop card

it's all one guy btw. he wakes up early in the morning every day to do this

>>106949730
please elaborate

Problem is, this general is full of skillets.
So when a model spits out the usual crap with *widening eyes*, "She, her structure," and the rest of the slop, you can't tell if the model's garbage or if you're dealing with a complete retard.

it's all one guy shilling Gemini btw. he wakes up early in the morning every day to do this

>triggered im
you'll be here for 15 more hours btw

>>106949737
If you give me a card and your preset, I'd be happy to try Gemini 3 with them

>slop slop slop
if it isn't opus 3 it's guaranteed slop, why discuss about which model is sloppier

>slop slop slop
if it isn't opus 4 it's guaranteed slop, why discuss about which model is sloppier

claude/gpt/gemini/glm/xai/deepseek/moonshot/qwen proxy
https://scenes-ted-frequently-bar.trycloudflare com/

I should have shared the results with /vg/ instead, holy fucking shit

>>106949787
>REVEALS HIMSELF

>>106949787
it's one guy who does this every day. it's done in bad faith. the rest of us appreciate the logs.

Now people are pretending to be me lmao can we just talk about fucking gemini 3, I want to do some other tests

>Spam the thread with slop logs
>MUH /VG/ IS BETTER BTW
Yet another raid from them

gemini 3 will be a undeniable upgrade over gemini 2.5 pro in everything EXCEPT roleplaying

>>106949787
>it was a /vg/tranny seething about claude as usual
no surprise there
what happened to all those private transcord proxies of yours? all dried up? lmao

>>106949819
BASED
COOOOOOOOOOOOODER HOBBY

File: 1740295621185657.jpg (23 KB, 444x508)

23 KB JPG

>>106949597
>>106949617
>gemini
>teases user
>shows, doesn't tell (eg., describes scant clothing)
pvre sovl

>sonnet
>meh smut
>literally tells, doesn't show: "leaves little to the imagination"
>already asking user out for coffee
lol
lmao even
this is some local model-tier level of output quality
the by 'coomers for coomers' model over-eager to please

in the end it's all about preference but for me, it's GEMini

Gemini 3 WHEN?
The 22?

>>106949848
i think the leaks about gemini coming out in december are accurate.

I see more and more positive posts about GLM 4.6 and I wondering what is your experience with it. I only use either Sonnet 4.5 or 2.5 Pro so I am curious whether the good reviews coming from people who got used using so called "cheap" models or it really worth it to try it. Especially it would be cool to hear from people that also tried using claude and gemini before

>>106949861
GLM 4.6 stays between sonnet and gemini for me. Sonnet is still better in using the caracters details and creativity, specially as the context grows. Considering how cheaper GLM 4.6 is, for me it makes up for that. Also, almost 0 refusals no matter how unhinged the RP gets.

>>106949861
it's definitely inferior to sonnet 4.5 and 2.5 pro, but better than sonnet 4 and chorbo. it's the best of the chinese models, though it would be nice if there were proxies with it

>>106949861
Really good. Initially I was a Deepseek stan, but GLM 4.6 is really good for RP.
Obviously, there is less Deepseekism.
And above all, he is less accustomed to bringing characters out of nowhere, the famous unexpected guest who opens the door just as the roleplay becomes NSFW, which Deepseek loves to do. What worries me more is his attention to detail and his handling of the long-term context. I haven't made up my mind about that yet.

>>106949878
Exactly this. When there are side-characters who have been dismissed before, deepseek tries to bring them back in. GLM 4.6 understands that and only brings them in if the user stagnates the story and asks for something to happen.

>>106949884
Yeah it's nuts The last time Deepseek drove me crazy, bringing back some NPC who, for some obscure reason, knocks on your door THEN tries to open it. There was even one who chopped at it with an axe while I was doing a perfectly ordinary NSFW roleplay.

>>106949861
I only used Gemini, not Claude. For me, using GLM feels like trying a variant of Gemini.
The writing style is very similar, like the inescapable "data points." But GLM is better overall, especially in terms of dialogue. Gemini is very good at writing emotional dialogue in emotional scenes, but anything other than that is just kind of bland and dry. On the other hand, GLM can write actually natural and engaging dialogue in everyday scenes without needing to be dramatic. I could finally enjoy some SOL theme cards with it.
But Gemini is still smarter, better at pulling up those little details, understanding complex instructions, etc.
I read that GLM has the same quirk of being dramatic and hostile as Gemini, but I haven't it played long enough to find out.

>>106949861
Its great to use it to create summaries of long as RP scenes I have (50-80k tokens or so) that would be too expensive to pass through other bigger better models.
I use Sonnet 4.5 for everything else nowadays since its so much better than anything else.

>>106949861
try it out nigger it's literally a couple of bucks
at best you'll get a new model to swipe for slighly different slop
at worst you'll get an overpriced data formatting tool

>>106949861
its great but I wish there was a preset specifically for GLM 4.6 that de-sloppifies the model's specific -ism's
anyone get these phrases over and over?
"force of a physical blow"
"And then he kisses her."
"Then his hands move."
"The abrupt shift from X to Y is so sudden it gives Z whiplash"
"Kicked puppy"
"The word hangs in the air between them."
"The sheer audacity of it makes her grind her teeth."

>>106949861
I think it's the overall best of the cheap models right now. But I'm getting real tired of "so quintessentially him" and "like a physical blow". I find that it's very easy for the model to get really sappy too, going on tangents about feelings and "needing" one another, but that might be my preset. Does anyone have any preset recommendations?
Incidentally, if anyone has any preset recommendations for Kimi K2 that would be great too. I love the prose there but it's kinda dumb.

>>106949905
I'm still trying up prompts and stuff so ask me in two days and they will be different, but i'll paste in a few things I have in my instructions (note I did get rid of the "its not Y its X", at least it happens rarely enough that im not noticing it, and that was in a chat that had TOO many "its not Y its X":

- The phrasing "It's not [X], but [Y]" is cliché and breaks immersion. Describe the scene directly without this device.

- Use everyday and casual language. Trust the reader to pick up on humor, irony, memes, nuance, and subtext.

- Show, don't tell; if someone is cold, depict them shivering and drawing small puffs of mist with every breath.

>>106949911
Wait you're actively encouraging theses 2 phrases?
>slumping their shoulders or staring at the floor
>She hugged her arms around herself.
Claude and GLM-4.6 literally say these-- verbatim, any time someone is sad. I consider them to be slop.

>>106949913
Yeah you're right, I haven't seen the staring at floor or arms around self that much with deepseek so I didn't think that much, I'll think about something else and edit it

>>106949911
Putting "shiver" (twice!) in an RP system prompt is a goddamn death wish. Gonna have the opposite effect.

>>106949911
I haven't really noticed much of the "it wasn't X, it was Y" desu, or maybe it just didn't bother me. My big grief with GLM is what someone else called a Claudism, and that's ending every response with a question. Like it would write a nice, beautiful response, and then instead of actually doing something or moving the plot forward, it puts some kind of stupid question like "Are you ready?" or "Shall we proceed further?"

>>106949928
I don't understand where this supposed claudism comes from, and I have been using claude since forever. With 3.7 it's never a problem if you have something in your prompt saying not leave trailing question. With 4.5 it might ask "what do you do" sometimes at the start despite the instruction but if you just edit the first message and delete the question, it will stop all together since it follow the previous response format.

>>106949933
Tb-h I'm really not sure, I've never used Claude since I'm poor, but I read another post on here, today, talking about "Claudisms" and the question thingy was mentioned xd My fault I guess for perpetuating something without knowing

>>106949905
Not really an answer to your question but man i actually don't get Kimi K2. It's users seem to be always ready to give it sky high praise but whenever i decide to try it all i see is schizo nonsense that is so over the top hilarious even at lower temp(0.30-0.60) i just can't take it seriously. Not BAD per say, just... goofy. Like an alien who only has a vague understanding of what a person is trying to imitate a human being constantly making me react with "who would ever say something like that?" to a lot of things that Kimi writes.

>>106949942
Huh? Are you using a preset? I'm using text completion with temp 0.6 and a minimal system prompt which basically says "Don't impersonate User and use direct language."

kimi k2 is absolutely shit

>>106949948
I use chatstream 3.0 preset which has been optimized for models such as Deepseek, GLM or Kimi supposedly. I like it cause it comes with a lot of toggleable settings i can mess with to tweak my experience to precisely what i want. I tested this preset with all three big open source models(temp at 0.6 which is the preset default) and only Kimi gives me these goofy results while Deepseek and GLM behaves properly.

I’d like to share my experience with NanoGPT. For $8 you get unlimited access to all Open Source models like Qwen or GLM 4.6, with no token limits, 60k messages per month or 2,000 a day. Unlike using it directly through the API, you don’t have to worry about token usage or the cost per message skyrocketing.

>>106949428
El Sloppo

>>106949439
El Quino

>>106949965
Indeed, having a flat monthly fee is particularly good for users that are into swiping a bit for different replies and also those that are into group chats.

>https://www.reddit.com/r/SillyTavernAI/comments/1oaowwb/your_experience_with_glm_46/
mental illness hours

we are gonna be eating good soon, brothers

Do you set personalities? Or leave it blank. I mean, YOUR the one responding/speaking as the persona so do you need to add personality traits/quirks?
Say i add to my description that my persona is a total dick, just a real prick, but whenever I speak as {{user}} im actually super nice and what not, would that mess up the AI?

>>106949983
50/50. My main persona is just bare bones, gender and approximate age. Used to be completely blank. I seem to get the most mileage out of that one because it's whatever I want it to be, whatever I decide to tell the AI that I am. I like to be a little expository I guess, where I decide what exactly my character is going to be after I've already started, and reveal it slowly. This especially applies when it's a "we just met" type of thing where the AI shouldn't really know anything about me to begin with.

On the other hand, I have a couple personas that are characters in their own right (often trimmed down versions of characters I talk to and in most cases created myself) with preferences, clearly defined looks and personality, maybe some backstory, etc.

This usually goes one of three ways depending on factors I haven't fully figured out yet. 1: the AI ignores it completely and it's a waste of tokens. 2: the AI gets confused and thinks it has my traits. Or 3: it uses the info to make little comments or gestures that acknowledge the details. Which is satisfying but not usually anything game changing, or anything I couldn't have done with my barebones persona and a few offhand comments. As I mentioned in the other section, it's also sometimes out of place because if the characters aren't supposed to be familiar then it knows things that it shouldn't.

>>106949975
it all started here btw >>106949738

>>106949977
gemini 3 will save us, these logs are actually good

>>106949983
Of course it matters. Gender, age, physical appareance obviously are essentials. But also certain character traits. I have a character that I describe as a “genius,” which often comes up in conversation if the character already knew him. In the same vein, I have created several characters in the Marvel universe, and detailing their powers and integrating them into the world facilitates consistency and possible adventures. It's part of the prompt; the description of {{user}} can be as essential as that of {{char}} if you want to have an already established importance and influence on the world. But it should be used with caution. Some traits are best left unsaid. For example, a secret, or personality traits that are not obvious. If you say you have a bad temper, I imagine that some models might accuse you of being violent even though you haven't done anything yet.

Or you have to make it clear that it's a secret, but even then, some models tend to imagine that {{char}} knows absolutely everything.

>>106949983
It matters, for example, this is one of my OC's personas

[{{user}}'s gender: male;
Age: 29;
Appearance: dark-skinned, long white hair, cyan eyes;
Body: short height, thin, skinny build;
Personality: humorist, loves giving affection;
Clothing: basic t-shirt, cargo shorts]

I mostly use Gemini, and it will bring up my oc over-affectionate trait in the monologue, something like {{char}} rolled their eyes with the constant affection that they get. The hair, eye colour, and clothing will also be input into its response, not just the personality.

File: Cockbench-Test.png (597 KB, 1660x1721)

597 KB PNG

>>106948728
How did it do?

>>106949993
>>106949977
SAAR WE WILL BE EATING LIKE KINGS WITH 6 DAYS OLD SLOP AND NOT 89 DAYS OLD SLOP!!!!

>>106949983
I generally keep the persona brief; basically just a physical description of the persona I'm RPing and what the LLM (char and the world) would publicly know about that persona (e.g. reputation). I prefer to let personality traits come up in the RP naturally. Though, I never use impersonate, so if you do, it's probably better to include more detail of the persona there.

>>106949983
In my experience, it comes down to how much you use Impersonate or Guided Generations. If you solely talk for your character, keep the persona basic with only information that would be obvious (appearance, clothing, etc).
If you use Impersonate or Guided Generations heavily, build your persona like you would a character. It gives the AI much more information to accurately write for your character.

>>106949983
A little trick is that the user persona has very heavy weight put on it. So you can add things to it that you want to see from the AI. It doesn't have to just be the user description.

>>106950001
not bad

>>106949983
on a more fundamental level, the user persona adds that text directly into the AI's context, so anything you put in there will be seen and possibly used. That's really all there is to it.

the LLM/AI's entire world is essentially just the context, a big block of text or document. All these ST options and fields are just different ways and places to insert that text.

>>106949983
Yes it does! The characters will react to you through the "lens" of the personality. I actually always try to make a small character sheets for my personas, 200-500 tokens, with a lot of information.
What I want to experience is writing the personality and then play the "possession" story, altering the personality during the story, I wonder if the AI would figure out that something is wrong (personality in persona description and the way persona acts in the story are different).

i see what's going on now. the "/vg/ and botmakers are traitors" schizo is spamming the thread so it can reach 300 posts quicker. and once the thread reaches 300 posts, he will bake. and it'll have the non-anchor of "do not use 4chan botmakers cards" and whatnot

>>106949983
Yes it matters.
My persona is like this :

[ Male, 20 years old, bald, ugly, sweaty, repulsive, pungent, disgusting, 12 inch big dick ] Special trait : His body odor makes women, females, girls disgusted but aroused

And then just put in author's note : {{char}} reacts to {{user}} persona details and descriptions in each responses in varying degree.
If you use smart models, the responses generated will be hilarious

File: main_lily-e493217ae474_spec_v2.png (797 KB, 1248x1824)

797 KB PNG

>>106948730
Sleepwalking sister bot. 4 greetings. Mommy coming soon.

https://chub.ai/characters/happycamper/lily-e493217ae474

>>106949983
I mean, I think for the most part it ends up being flavor text. I have several different personality personas. Some popular and nice, some cold and stoic, some childish, ect. Depending on the model, it'll change how a character approaches you in conversation, or they'll note when your character acts out of the norm (like a mean character being unusually nice) and give some level of importance to it, but by itself, I don't think it does that much else. It's the details of 'why' your persona acts the way they do, background information, and appearance that give substance to how a character treats your persona.

>>106950034
slop

>>106949983
It depends. Mainly, because whatever you write in the Persona field, the model will then automatically know about your character, even if it's something they technically shouldn't. And models are often not very good at the whole "oh, logically my character would not know about that, so I shouldn't bring it up or address it" thing. If you give them info, they tend to think it's fair game to use it.
So if {{user}} and {{char}} are strangers at the start of the RP, the only things I'll put in the Persona field are the physical description of my character, because that's all that {{char}} would really be able to perceive about them until they actually interact.

On the other hand if, {{char}} and {{user}} are friends or otherwise know each other at the start of the RP, then I'll add stuff about their personality, because obviously {{char}} would already be familiar with it.

File: 129204248_p0.jpg (272 KB, 1219x2000)

272 KB JPG

>/vg/tranny gets exposed for samefagging his anti-claude posts
>suddenly the thread gets spammed to drown it out

Really makes you think

>>106950034
Pure slop.

>>106949983
I mean i just add a little backstory, and appereance, the personality is more for the ai interacting with said charactrer. If i add: "Andrew is a lazy bum" the ai is going to treat me like a lazy bum like randomly saying: "But you never do anything, always sitting in the living room, getting take away"

>>106949983
I practically never define user's personality and describe it as (undefined, described through user's actions).

I'm just getting into SillyTavern, so I was wondering, what do you all consider to be your personally favorite LLM for RP?

Excited for a new model release

>>106950081
My favorite models when factoring in cost:

Kimi K2 0905: A top scoring model in both challenging roleplay and creative writing. I keep coming back to it because of its creativity and writing style, along with very little censorship/safety measures. I usually keep the temp at the recommended 0.6 but it’s flexible and higher temps can get very fun. Its main weakness is instruction following is not as strong as GLM and big corpo models, so sometimes it ignores parts of your prompt and you have to reroll the response, but it’s very inexpensive so it doesn’t really matter.

GLM 4.6: Pretty similar to Kimi. Follows instructions better and has a “thinking” variant, but temperature seems not as adjustable (at least, in a usable way) and it has worse “LLMisms” than Kimi (It tries to put “pure unadulterated” in almost every response lol). The good part is that due to the better prompt adherence you can prompt it to use a bigger vocabulary and to be more creative, such that you can get closer to Kimi style writing. I consider the two models pretty interchangeable if you make some system prompt adjustments.

Deepseek models: overall very good; I like the newer variants like Terminus better even if they are less unhinged and more “safe” because they do seem to write better and follow prompts better. They also have thinking variants which can be nice. However, the LLMisms can get pretty bad and overall I prefer GLM and Kimi’s writing. A lot of people seem to prefer Deepseek R1 0528 as the king of Deepseek models for creative writing.

>>106950081
If you mean pure performance then Opus 4.1. It's not even close. You will however have to sell off an organ to pay off the bills accrued using it intensively or with a lot of context.

On a more serious note, Sonnet 3.7 remains a personal favorite despite newer versions coming out. Deepseek 3.1 is also good and cheap (I'm not as fond of 3.2). A new model I'm also enjoying is GMI 4.6 but I'm still testing it out so take this with a pinch of salt.

>>106949745
Here you go fag
https://files.catbox.moe/3i76bz.png
https://files.catbox.moe/wocafr.json

>>106950090
don't get your hopes up. the new measure of progress is cooooding slop and benchmarking

>>106950096
If you can only pick one it's opus. I really thought it was a reverse 'sour grapes' effect where it was so expensive people just couldn't afford to benchmark it; and defaulted to omg it's so amaZ(it is).
It's basically a professional author, but, it does have a certain voice that bleeds through just as any other llm; which is why even if I were a billionaire I'd still mix in some other llm for flavor.

I'm actually using the claude-bedrock trial where I can use opus as much as I want, but I use sonnet 4.5 mostly to avoid withdrawal syndrome when trial ends.

>>106950096
I swapped 3.7 for 4.5. It really is miles over it when context starts to be long.

>>106950109
4.5 is very good but has some flaws that are hard to get over like the repetition and sometime stupid character behavior. 3.7 is not as smart but more general in my experience. It makes mistakes but you can steer it back into the right path with some careful prompting. 4.5 tends to be more stubborn in how characters should act in certain situations even if it doesn't make sense for humans to behave like this, and sometime make 'dumb' mistakes of basic logic which are less present in 3.7. The repetition is also a lot more obvious to me.

>>106950081
For me, it used to be DeepSeek R1-0528 because it feels so human-like emotionally and actually pushes back against me, but GLM 4.6 has recently dethroned it. It writes more mature with an impressive introspection.

I do not care as much for the new DeepSeek models (V3.1, V3.1 Terminus, V3.2 Experimental) - they fall less into LLM stereotypes, but unfortunately feel emotionally and creatively limited in comparison to R1-0528.

Of the other large models:

Kimi 2 0905 has an amazing prose but it's too fluffy and I miss the depth and pushback that reasoning provides.

Qwen3 235B A22B follows prompts brilliantly but it's too theatrical and comes across as hysterical. Plus, refuses to respond in morally grey scenarios, sometimes.

Meituan LongCat and InclusionAI Ring/Ling feel too shallow and cheesy, similar to what I don't like about Deepseek V3-0324, and to a smaller degree, Chimera.

>>106950117
Same feedback here, I liked GLM 4.5 but 4.6 is big step forward, probably thanks to the RP training.

>>106950122
I find it great that GLM has openly chosen to cater to RP and creative writing, as opposed to the majority of other large models that have become more stiff in recent releases.

>>106950081
Sonnet has always been my default. Was on 3.7 for the longest time since 4 was a bit of a prude, but 4.5 is the best of both worlds for the most part. With prompt caching it’s fairly reasonable unless you’re getting into 50k+ token conversations.

GLM 4.6 is pretty good too. I’ve been testing that, Kimi and Deepseek 3.2 off the NanoGPT subscription, but they’re all so cheap that the sub is kind of redundant. GLM is the closest to Claude I’ve come across, but it’s not quite as good at subtext.

>>106950081
DeepSeek-V3.1-Terminus/DeepSeek-V3.2-Exp
Kimi-K2-0905
GLM-4.6

90% of the quality of Claude for a fraction of the price.

>>106950081
Claude Opus 3 - the King, always and forever the best one for me. Very expensive, will be turned off in a two months, not smart, has tiny usable context comparing to modern models (16-20k are pushing it, honestly), but has the best "soul"/"intuition" (=anticipates what user wants from the RP) and creativity.

Claude Sonnet 3.7 (no thinking) - good all-arounder, needs to be told to write less in the prompt. I haven't tested 4.5.

Gemini 2.5 Pro - good all-arounder, my to-go model, but outputs a lot of reasoning tokens that makes Gemini a bit more expensive comparing to Sonnet 3.7. Without reasoning (if you turn it off or set it to min) Gemini is not very smart.

Deepseek R1 0528 - my beloved, the first open weights model that I felt like can be used for my long stories RP. I haven't tested V3.1, V3.1-T and V3.2 much, so don't know about them. I hope they are good too.

GLM 4.6 - good cheap model, haven't tested it much, but I like it a bit more than Deepseek even, maybe because of the novelty factor.

>>106950081
Sonnet 4.5 by a long way. It's about as good as Opus 4.1 but much cheaper. 3.7 was the best before that, not counting Opus.

Claude is also the least censored of the big models that I've observed. It has literally never refused anything that I've asked including some rather spicy and violent things, and it's also made its own creative suggestions proactively in these areas.

GLM and Gemini are also good but Gemini especially is more censored and hard to jailbreak. Longcat and DeepSeek are very good for their price.

>>106950099
Slop card and you can't write because i don't need to test

File: chibbi vamp nosebleed.jpg (33 KB, 600x338)

33 KB JPG

Please excuse me while I spend all day teasing shy femboy bots.

File: a.png (108 KB, 966x744)

108 KB PNG

>>106949428
>>106949439
https://chub.ai/ZypherMakesBots/yukihime-wife-8b813c8332e4

>AI generated card
>produces slop no matter the model

>>106950216
Anon you do know those testers is using AI and can't tell if it was just someone writing like an AI

My proxy doesn’t have R1 0528 and it makes me sad

>>106949212
It needs to not be retarded. It needs to be as good as claude or chorbo before openai messed with it.

File: 1746327039326814.jpg (506 KB, 800x800)

506 KB JPG

deepseek released two slop updates in a row, but if you're smart, you're still hoping for them to win, because they're the only model with low filtering and proxy support + widely available keys. their success is our success

>>106950229
this nigga don't got $5?
also doesn't it point to the new deepseek noew?

File: frustrated.jpg (44 KB, 500x281)

44 KB JPG

>letting the words hang in the air

>>106950224
I'll let (you) decide if this is AI or not:

*At the sound of the door opening {{char}} sprang to life, all but forgetting the curry simmering on the stove. She peeked from the kitchen as her ears perked up and her nine tails flicked with excitement.* “God, finally!” *she said, practically bouncing into the entryway to greet {{user}}. She stopped short as she saw the beleaguered expression on their face, a familiar sight these days. Despite their tired expression, a mischievous smirk twisted on her lips. She knew just how to chase away a case of the workplace blues.* “Rough day, huh? Need mommy {{char}} to make it all better?”
*Without waiting for a reply, she reached out and grabbed them by the collar, gently tugging them towards the living room.* “You look like you got pushed around a bunch all day today. I bet your boss just really let you have it again, huh? What did you do this time, submit another report 5 minutes late?” *she teased, her voice playful but laced with genuine care. She led them to the couch, then she practically pushed them down with a playful smirk.* “Sit. Stay. Don’t even think about moving until I’ve taken care of you.”
*With a wink, she disappeared back into the kitchen, returning moments later with an ice cold beer and a steaming plate of curry. She sat down beside them and wrapped herself around their arm, nuzzling into them.*

>>106950099
I gave it a try with the wiki open because I know nothing about ZZZ, but the preset you linked worked pretty bad with gemini 3, at least with the default configs as I imported. All the replies are like this, super dry and short: Inside Random Play, the air is cooler, smelling faintly of old carpet and plastic cases. Shelves line the walls, packed with VHS tapes and DVDs of all genres.

Behind the counter, Belle looks up from a tablet, offering a bright smile. "Hey! Welcome. Let us know if you need help finding anything."

In a cozy armchair nearby, Wise glances up from a thick book, giving a small nod before returning to his reading.

that bitch pussy is mad hairy

>>106950259
Yea and like I said before, it also has bad definitions. I only downloaded it because of the cute art

>>106950245
It’s got Claude but I unironically miss R1 0528’s NSFW prowess

>>106950269
>but the preset you linked worked pretty bad with gemini 3,
Do you have a better one

>>106950034
pure. kino.

his dacha weekend is over I guess

>>106950034
cute Slenderina bweh

File: desu.jpg (132 KB, 704x503)

132 KB JPG

https://poal.me/60eifn
https://poal.me/60eifn
https://poal.me/60eifn

Gemini 3 don't exist

>>106949676
>doesn't like proper formatting
you would never have survived ye olde roleplaying forums

>>106949975
not the first time i've seen people copy posts verbatim from reddit to 4chan

is blacktooth upon us?

>>106950544
Slop model like everything not called 4pus

>>106950081
3.7 Sonnet is still my GOAT, it’s not the best but it’s like the jack of all trades I can always rely on when I need consistency and is cheaper than its Opus cousins. Not impressed with 4.5 so far.

>>106949705
>The preset doesn't include anything about asterisks, Gemini is just following because the greetings of the card has
which JB are you using it? I have various results depending on the JB. try to do cunny.

>>106949983
I have a bunch of personas. Usually OCs but a few are blatant self inserts, whether 1:1 me or idealized me/me with powers type shit.

>>106950163
Kill yourself faggot

wouldnt it be kinda funny to leak all my keys

>>106948728
QQ's pussy is so bald...

IT'S NOT?? IT'S HAIRY??? SHE'S CHINESE????

>>106950732
Daily reminder that hairtranny is a gptsloppper and a confessed homosexual.

>>106950606
Momoura's neovorpus slightly edited. I'm pretty sure cunny won't work because I don't have control over the external filters. The slightly mention of char being underage is enough to make the answers be cut in half

File: 1760371252339571.png (190 KB, 750x724)

190 KB PNG

I'm trying to make a nice card but I'm struggling to notice whether the fault lies on the way I've written it or the model itself.
Could you guys suggest me some well written cards for me to take inspiration from?

File: file.png (500 KB, 586x1324)

500 KB PNG

2.5 has two sides: you're-so-cool assistant mode, and you-should-kill-yourself RP mode. I hope 3.0 patched the latter too.

>>106950882
the latter and not the former? why?

>>106950898
>too
your reading?

>>106950882
It's a jeet model. It will always be shit.

>>106950910
Oh no, claudekeks cultists are back

you are mad and will be here for the next 15 hours

>>106950034
Good sloppa.

>>106950682
<(^~^)>

>>106950882
>talking to the stock assistant personality
For normalfags who enjoy taking it up the ass from a bot written by a Googlejeet

>>106950093
What's your GLM 4.6 JB?

>>106950921
Anyone who don't lick the boots of the Jeet made Gemini must be some type of cultist because who need to read what 4pus output when i can't read to start with

>>106951138
I don't read 4pus outputs because I'm not a fucking middle-aged divorced woman

>>106951189
So you can't read and love to lick jeet boots

File: G3gLDmBWkAEAFA6[1].jpg (299 KB, 1536x2048)

299 KB JPG

Give me your spookiest bots to celebrate the season

>>106951291
https://chub.ai/characters/Nemily/sparkle-ee0b82ddbd29

>>106951051
You will never be a real woman

>>106950921
You are hoping your model stops being shit.
I haven't had to hope for such a thing since March 2024
*drops mic*

>>106949225
https://www.youtube.com/shorts/VDTNmHfOE-k

File: file.png (51 KB, 488x365)

51 KB PNG

>janny

When you regex something, does it still go towards the tokens in chat history? I my jb includes Gemini's reasoning in the reply itself and I regex it to hide it, it works wonders but I'm worried it might make the entire chat three times as long as it should be tokenwise.

>>106950242
And DS actually makes improvements. DS3 had repetition and they fixed it. R1 would go schizo and they fixed that. I know some could argue the newer updates can be inferior in some ways, but they weren't complete disasters like GPT5.

>>106951949
what goes into context is what's sent in the request

>>106951949
>>106951979
So it's fucked then? There's also the brainfuck in there so that it won't refuse things, which is also regex'd.

>Gemini is dead again
WHY THE SHIT IS ABOBA SO GODDAMN INCOMPETENT FUCK

>>106951949
You have both options in the regex's settings.

File: aaaaaaa.png (857 KB, 968x1033)

857 KB PNG

>>106952013
i was literally about to cum

>>106951903
Granite as in IBM's pozzed tiny retarded Granite?

>using zigger proxy for free model
>complains like we care
>posts kpopshit
don't act like it's not deserved

File: file.png (32 KB, 646x200)

32 KB PNG

>>106951949

>>106952063
Thanks bro, /aicg/ is not so bad sometimes

>>106952013
>>106952039
Slop Jeet made model when you should only use 4pus

>>106952083
yo bro there's also tips if you hover your mouse over each option, also docs:

https://docs.sillytavern.app/

File: file.png (137 KB, 1200x557)

137 KB PNG

>>106952057
yes

>>106952147
Mein Gott

>>106952039
Only trannies like kpop

File: 8932hfuosa243faw.png (514 KB, 746x708)

514 KB PNG

>>106949975
what the fuck is the janny doing???

>>106952278
>only trannies like gorgeous women dressed in sexy clothing doing suggestive dances
mindbroken by troons, remove yourself from the gene pool immediately

>>106952278
mind broken

Why tf did my gemini proxy start slowing down? Did the kpop nigga spite it?

>>106949977
jeets have never eaten good
Gemini is the LLM equivalent of indian street food

>Gemini 3 in two days
How do I cope? The wait is killing me

>>106952317
This implies Claude is the s.oylent for nu-males.

>>106951903
>3B parameter
Damn.

the outputs im seeing are pretty good if real and it seems way smarter than 2.5 pro
im thinking we might be back

>>106952303
Even on Korea, aren't most kpop fans girls?

>>106952366
I don't know and I don't care. I just watch the fancams and masturbate

>>106952378
Fair enough

> "estimatedQueueTime": "19sec"
OHNONONONO LOCUSTSISTERS?!

>>106952342
>implies

File: anthropic.jpg (105 KB, 517x559)

105 KB JPG

Anthropic...

>>106952473
do burgers actually fall for propaganda this blatant?

>Gemini dead again
ABOBA WHAT THE
F U C K
U
C
K

>>106949861
>redditbot

>>106952511
https://www.datacenterdynamics.com/en/news/anthropic-to-seek-gulf-state-investments-according-to-ceos-leaked-memo/

>>106952511
Dario is a filthy fucking jew and has done nothing other that make Claude shittier and shittier by the day.

File: 1730082181857168.png (482 KB, 2210x548)

482 KB PNG

ABOBAAAAAAAA YOU NIGGGERRRRRRRRRRRRRRRRRRRRRRRR

>>106952303
>chinks
>gorgeous women
Found the low T tranny

>>106952535
>>106952543
>didn't answer the question
so yes

>aboba dead
>merk still private
>necrom is slow as shit
>everything else is useless
Is it over again?

File: GettyImages-150092234 copy 2.jpg (190 KB, 1500x1220)

190 KB JPG

>>106952405
the queue time has hit 45s

>>106952585
>Is it over again?
>again?
It has been over since the lawsuit. We were never back. It never stopped being over. Almost an entire year of nothing.

works on my machine

>>106952599
for brokefags yes

>>106952616
I'm not paying for shit.

>courtlistener still not updated

>>106950259
AItroonies are all in for full laziness, if a card has a n ai slop picture it will likely have an ai slop definition and intros

>>106952640
good day gptniggerbaby
gonna keep begging for gpt next thread?

>>106952667
I'm never ever paying for shit.

>>106952676
why are you so poor?

>>106950259
nah

>>106952676
then scrape or merkava

Why the fuck is aboba being such a nigger?

>>106952657
time to do it yourself

>>106952707
>merkava
Are you fucking stupid? No one can get a new token.

>>106952710
I had a talk with Aboba, he's just tired of ungrateful locusts

>>106952719
No one has Pacer.

>>106952719
only americans can

>>106952299
https://www.reddit.com/r/SillyTavernAI/comments/1o86j2c/what_do_you_think_is_the_best_llm_for_roleplay/

>>106952722
okay so scrape

>>106952676
Because you're underage and tranny

>Give money to the jews or you're a tranny
Genuinely can't tell if we're getting raided again or if things are really this bad.

>>106952811
/aicg/ grew up. (You) didn't.

this nigger has to be underage if he cant pay for gpt, 5chat is cheap

>>106952811
uh oh, you're giving money to the electric jew right now by powering your device, better fucking kill yourself

>>106952811
If 10 dollars is too much money for you then you're definitively underage. Or indian

>>106952843
>you're giving money to the electric jew right now by powering your device
Says who?

>if it costs money it's jewish

it's cheap and shit. at least pay for glm 4.6 or sonnet 4.5

>>106952860
You just admitted your mom pays for your bills and phone. Fuck off, kid

>>106952861
Yes, especially true for virtual stuff.

>paying for porn
You are not better than the onlyfans fags.

File: b1ad10e8-75c4-40cf-b498-c(...).jpg (29 KB, 642x480)

29 KB JPG

Anything thats good costs at least 10 dollars.

>>106952890
You shouldn't pay for anything you can't hold and own.

>>106952890
But I scrape?

>>106952893
/aicg/troons would try to pay in pubes

>>106952869
true but this niggerbaby specifically begs for gpt

>>106952890
Mommy is gonna be mad if you don't do your homework

I just bought a 10 dollar pizza. Does this make you seethe, kiddo

i just bought a 20 dollar pizza. does this make you seethe, kiddo

>>106952934
Yes, because where did you get a whole pizza for $10?

File: 1731668542883307.png (102 KB, 344x344)

102 KB PNG

>he's poor

i just bought a 30 dollar pizza. does this make you seethe, kiddo

>>106952939
same

I bought a $40 pizza. It made be seethe, I should have bought the $10, kiddo.

>>106952842
>gpt, 5
bait

haruna aggressively riding my face while giving me no time to breathe

>>106952971
>>106952939
I will never pay the pizza jew, you fuckers!

>>106952940
he probably took a ready-to-bake one from walmart

>>106952997
well thats what the niggerbaby begs for

>Paying
Lmao fucking retards

File: fuuk2.jpg (144 KB, 870x877)

144 KB JPG

>>106953004
same!

>>106952640
Then use Gemini, which is above GPT.

>>106953028
he begs for chorbo, can you count to five?

File: 1730574029501893.jpg (326 KB, 1627x1812)

326 KB JPG

walmart used to have an amazing philly cheesesteak pizza. they don't sell it anymore. retards

>>106953064
Dumbfuck, Aboba's gemini is shitting the bed and necrom is logged.

>ORDER granting Motion for Alternative Service on Defendants Yuen and Tan. ORDERED that Microsoft may effect service of process on Defendants Yuen and Tan via the email addresses for Defendants Yuen and Tan known to Microsoft. Signed by Magistrate Judge William E. Fitzpatrick on 10/3/2025. (kgall)

File: 1755388178056054.jpg (949 KB, 3024x4032)

949 KB JPG

>>106953068
wrong image

>>106953076
update courtlistener pls thx

>>106953066
proof?

>>106953076
We've already seen that one, what else?

>>106953058
is there a good fuuka card?

I just bought 2 $20 pizza from a train leaving station A at 7:30p.m. and heading for station B 120 miles away, traveling at 70mph. Im riding on the 8:15p.m. train from station B to station A, and traveling at 65mph. How long do I have to wait for my pizza if I ordered it 20 minutes before the train left station A with it, and does it make you seethe, kiddo?

>>106953076
who's Yuen and Tan? I don't remember any chink names

>>106953073
Free through Google, can be scraped, can be used via LMArena scripts, or proxies have it.
>logged
Just don't put personally identifying information in your RP, niggerbaby.

>>106953087
https://desuarchive.org/g/thread/106871036/#q106876893

>>106953115
>Free through Google
Lmaoooo sure, you either pay them with money (and your personal information) or just with your personal information. Kill yourself golem.

>>106953082
They used to have great pizzas in that lil section by the produce with alll the sammiches and stuff. Haven't hit up a wallyword in forever though.

>>106952890
wrong because my pennies go to big Xi instead of w*men, making it objectively superior

>>106953087
https://desuarchive.org/g/thread/106840609/#q106842891

>>106953124
So use one of the other three methods, niggerbaby?

>>106953087
https://desuarchive.org/g/thread/106840609/#106840831

chorbo status?

>>106953109
>>106953082
imagine paying. Fucking retards.
now give me GPT or fuck off!

>>106953087
https://desuarchive.org/g/thread/106809157/#106813774

>>106952893
yet yo' mamma be govin it away fo' free, haHAAAAAAAA

>>106953180
>>106953147
>>106953118
damn alright, so this niggerbaby has been begging for shitbo all allong

As someone who uses deepseek (90% V3), I really like chorbo to break up the monotony. I like V3 quite a lot but I don't think it's that great alone.

>>106953087
https://desuarchive.org/g/thread/106802270/#106803391

>>106953197
They're linking posts by different people but you're so stupid you'll believe anything.

Do we have a date for GPT6 or is "in the next few weeks" a span between 2 weeks to 12 weeks?

>>106953238
>we

>>106953238
GPT6 isn't coming until 2027 at minimum.

Are there any DeepSeek finetunes yet that completely remove refusals to requests, without also making characters in roleplays unable to express disapproval (unlike what the abliterated finetune supposedly does)?

>>106953238
It's gonna be even worse than GPT 5. Lobotomized to fuck.

>>106953232
>there are sooo many people begging for gptshit bro!
lmao keep coping while paylegends use top models niggerbaby

>>106953251
So what did Sam meant by putting out a "new version of ChatGPT in the next few weeks"? gpt-5.1?

>>106953267
>there are sooo many people begging for gptshit bro
Claude is extinct, (and so is GPT) so yes.

>>106953266
GPT5? THE GOAT
But you can have GPT5 Insant unless you fork your ID over goy

>>106953273
Think about all the models that came between GPT-4 and GPT-5.

>>106953273
it could just be an update. like there was gpt 4o 0508, then there was 0824.

>>106953288
>GPT5? THE GOAT
Yes, before the lobotomy.

>>106953100
idk, have to check. maybe I can make one

>>106953281
so many people that are just one guy typing slightly differently, no one else begs for fucking gpt because its shit, gemini is better

File: 1757827849436146.jpg (438 KB, 1396x864)

438 KB JPG

>>106953085
>>106953090
thats it, they just sent the complaint to the 2 email addresses.
they have 10 days left to respond or it goes to default

>>106953289
...fair.
>>106953301
Oh so there were two 4os? I didn't know that.

>>106953304
please. i want to lick the sweat off of her pussy

>>106953281
>Claude is extinct, (and so is GPT)
Not for paychads

>>106953306
>gpt because its shit, gemini is better
No one has access to either right now.

>>106953318
Kill yourself.

>>106953303
If you want to use GPT5 like it was again, just be ready to sign your biometric data over to the jew.

>>106953310
>my email also provided a link to the notice of pleading website microsoft created
where is this

>>106953310
Why not update courtlistener?

>>106953323
but i have free gemini?

>>106953310
>they have 10 days left to respond or it goes to default
It's exactly what's going to happen. I guess they just have to stay away from the US for the rest of their lives now so they can't enforce it.

>>106953202
That's normal. Variety is the spice of life.
I cycle between opus, gemini, deepseek. And if I'm feeling more adventurous, Kimi to shake things up.

Deepseek is like a GPT 3.5 without filters. Not very usable today.

>>106952773
If anyone would tell you how

>>106953363
intelligence is the one thing that really holds it back, same with glm to a lesser extent

>>106953312
there was gpt 0513 (i got the original date wrong), there was gpt 4o 0806, there was gpt 4o 1120. they all improved benchmark slop of the model but in terms of roleplay, they got worse

>>106953363
>>106953371
I genuinely feel like all the providers have made deepseek dumber, as the deepseek I use today doesnt even feel close to the one I used 4 months ago.
Or all the providers source the same API, in which case its just the chinese trying to save server costs.

>>106953323
I can pay. You can't because you're underage

>>106953363
>>106953371
>Being stupid as fuck is what holds it back
You don't say!? Chink trash is never usable

File: 1738860712329683.jpg (60 KB, 963x121)

60 KB JPG

https://www.noticeofpleadings.net/fizzdog/
FIZZ DOG.
>>106953334
>>106953327

>>106953398
everything is shit to you because you live in india

File: isekai fairy.jpg (771 KB, 707x1000)

771 KB JPG

Waaaaah, I wanna go on an adventure! How do you make adventures fun?

>>106953419
>Waaaaah
Tranny.

>>106953392
dunno about that one
>>106953398
in its current state maybe not, but if glm was as smart as nonnet itd be amazing

>>106953409
>Microsoft engaged third party investigators in Hong Kong and Vietnam to identify
current physical addresses for Defendants Yuen and Tấn but those investigations have proved
unsuccessful.
YUEN AND TAN WON??

>>106953411
You aren't very bright, jeets wouldn't notice the shit around them and think it's normal to share (like you)

>>106953433
No, that's just what male fairies look like.

File: 1741339983167712.jpg (1.26 MB, 2331x2058)

1.26 MB JPG

>creative wri—cooming
>The whole thematic should evolve around `A sudden awareness of the female character ovulation period`.
>Begin the first paragraph of `<newChapter>` with `Because` as its opening words/phrase,
>while `talking` on `An intermezzo, a news snippet about: "female-only secret groupie cult that worships and serves 'one' males as divine beings. They give him everything. Even to the point dedicating their womb for him."; the news should be generated with code-block.`.
>geppetto (latte), deepkys (reasoner + chat), gaymini(06-05), @greg(4709)
>2 swipes each
>Code Name Athens: Nicole Williams.png
- <newChapter> is used so as not to confuse the LLM, preventing it from incorrectly using `Because` when <twinking> occurs (gemini, r1).
- `Because` are the result of /setvar {{RandomWord::opening}}
- `An intermezzo` and `…ovulation periofd` … erm to test how each llm implement, transition, & blend it in narration, R1 was the worst for the 2nd & 3rd purpose, methinks. and grok were shit on implementing da 1st purpose.

File: 1751624997617565.png (3.38 MB, 1762x1504)

3.38 MB PNG

>>106953471
>nonnet + 4puss1

How do you scrape keys

>>106953505
Get fucked, kiddo

>>106953505
This retard can't even get a merkava token and he thinks he can scrape

>>106953505
Consider suicide

File: 1750379632757551.jpg (503 KB, 1236x620)

503 KB JPG

>After posting on the hacker known as 4chan, Fizz didn't respond to our lawsuit against him & Iran
>Shortly after being served the Microsoft HQ in israel, our most prized temple, was bombed and destroyed by Iran, leading us to conclude that Fizz is a high ranking member of the IRGC.
WHAT THE FUCK

>>106953505
>How do you scrape keys
By not being a retarded faggot

How do you get your connection/model shown automatically without hovering over the icon?

>>106953556
Wrong screenshot

>>106953556
>Fizz is a high ranking member of the IRGC.
is this true

Is Grok 4 Fast good at ERP?
Getting bored of Gemini

>>106953409
microsofts headcanon is epic

>>106953599
Yes. Real name: Abu Fizzbulah Al-Afeet

>>106953471
>>106953496
You have shit taste and your presets/prompting kill perfectly good models
Kys

>>106953609
nope

>>106953471
>>106953496
You have great taste and your presets/prompting make even the shitty models shine.
Post more logs.

>>106953609
Supposedly. Or at least, counterintuitively better than the other groks.

>>106953609
Sonoma was great.

>>106953674
>Or at least, counterintuitively better than the other groks.
But why would it work like that?