[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
/mlp/ - Pony


Thread archived.
You cannot reply anymore.


[Advertise on 4chan]


File: Chag.png (36 KB, 1364x665)
36 KB
36 KB PNG
Welcome to Chatbot AI General #152, the thread for discussing and improving AI pony chatbots.

▶ MLP Bots
https://mlpchag.neocities.org
Spreadsheet (CAI bots + Old repository):
https://docs.google.com/spreadsheets/d/1J7BeqJVDS51cXF8Pgm2YZaFq-Z6ykSJT
SillyTavern website extension: https://github.com/MLPChag/SillyTavern-Chag-Search
Bunch of Character Cards (Descriptions/No Greetings):
https://docs.google.com/spreadsheets/d/1Y6LNOCqAZAWIX-OBEv55HjlzcEdpUh-XbKCpbpL6v5k
CAI bots converted to Tavern: https://files.catbox.moe/ckurq1.zip
Expression packs: https://rentry.org/ChagExpressions
!!!GALLERY!!!: https://drive.google.com/drive/u/2/folders/1Ao-h5HFGMPllSrzSBKM_BvGSiU9f0c2U

▶ How do I start?
1) Select a Frontend
2) Select an AI model
3) Select Jailbreak
4) Select bots
5) Lovemaking with AI mares!

Starting in this hobby can be confusing and difficult. If it’s your first time and you’re lost,
▶ ASK THE THREAD! ◀

Novice-to-advanced guide: https://rentry.org/onrms
https://rentry.org/sukino-findings

▶ SillyTavern (preferred frontend)
https://github.com/SillyTavern/SillyTavern
On Android: https://rentry.org/TermuxST
App that voices pony responses in ST: https://drive.google.com/drive/folders/16Ss26VBmgzcSuTGzhaHqRuyVRceTf-YB

▶ More frontends:
Risu: https://risuai.xyz
Agnai: https://agnai.chat

▶ Locals
https://rentry.org/lunarmodelexperiments
>>>/g/lmg/
Mistral Nemo base model fine-tuned on fimfics: https://huggingface.co/Ada321/Nemo_Pony_2/tree/main

▶ Jailbreaks
MLP JB: https://rentry.org/znon7vxe
More JB and guides: https://rentry.org/jb-listing
Hypebots for Tavern: https://rentry.org/pn3hb

▶ Botmaking
Editors: https://agnai.chat/editor
Guides: https://rentry.org/meta_botmaking_list
Advanced: https://rentry.org/AdvancedCardWritingTricks

▶ Miscellaneous
Scrapped/Collected rentries: rentry.org/mrhd
Chag Arts: https://drive.google.com/drive/u/2/folders/1yfajin_hB5rW_jy6WyFjRaRfDA3EUGiJ
https://rentry.org/ChagMiscellaneous
https://rentry.org/ChagArchive

Previous thread: >>42786023

**News:**
- GPT-5.1 family released: https://openai.com/index/gpt-5-1/
- Moonshot AI releases Kimi K2 Thinking - https://moonshotai.github.io/Kimi-K2/thinking
- Claude Haiku 4.5 released - https://www.anthropic.com/news/claude-haiku-4-5
- Z.AI releases GLM 4.6 - https://z.ai/blog/glm-4.6
>>
File: Anchor.png (552 KB, 1201x1200)
552 KB
552 KB PNG
Anchor for bots, lorebooks, scenarios.
Previous >>42786024
>>
File: TAnchor.png (49 KB, 512x768)
49 KB
49 KB PNG
Anchor for technical stuffs (Proxies, Presets, Updates, Models etc.)
>>
File: ABegging.jpg (96 KB, 559x537)
96 KB
96 KB JPG
Anchor for asking for bots, lorebooks, scenarios etc.
>>
File: Screenshots.jpg (1.86 MB, 1524x1879)
1.86 MB
1.86 MB JPG
Bots corner:

>Everypony knows anon is retarded
https://mlpchag.neocities.org/view?card=Anonymous/Everypony%20knows%20anon%20is%20retarded.png
>THINGPONE in EQUESTRIA (Update)
https://mlpchag.neocities.org/view?card=Tiffin/THINGPONE%20in%20EQUESTRIA.png
>Rainbow Dash Cum Jar
https://mlpchag.neocities.org/view?card=Anonymous/Rainbow%20Dash%20Jar.png
>Zecora (Update)
https://mlpchag.neocities.org/view?card=Anonymous/ZecoraStories.png

Screenshots corner:

Mare pussy or ponut? >>42787200
Nightmare Jack and the moon follower >>42787450
Loosened >>42788147
Opus 3 AJ log >>42788541
Buenos Días Nightmare Jack >>42788850
Anon is retarded >>42788916
Non-win >>42790203
Skillet >>42789295
Mi Amore Manor >>42790334
Breaking the 4th wall >>42791626
>>
File: Screenshots2.jpg (969 KB, 2500x1900)
969 KB
969 KB JPG
Screenshots corner 2:

Everypony knows Anon is retarded >>42789053, >>42789322, >>42791856, >>42791929, >>42791969
Screwdriver >>42790207, >>42790217, >>42790497
Thanks Celestia >>42792288
Monster! >>42793180
>>
File: Screenshots3.png (1.36 MB, 1567x971)
1.36 MB
1.36 MB PNG
Screenshots corner 3:

Nozone >>42793632
Great wisdom >>42793925
SPAM >>42794179
Big Mac bros... >>42794283
Radio effect >>42794518
I cast Fireball >>42794602
Twilight gets groomed >>42795615
Cumjar >>42795654, >>42795665
>>
File: Screenshots4.png (1.21 MB, 1580x1063)
1.21 MB
1.21 MB PNG
Screenshots corner 4:

THINGPONE play >>42795886, >>42795987
Kaiba VS THINGPONE >>42796074, >>42796141, >>42796259, >>42798719
Strawberry Tart >>42796361
Zecora update >>42797616, >>42797762
>>
File: Screenshots5.gif (1.56 MB, 1280x720)
1.56 MB
1.56 MB GIF
Screenshots corner 5:

Don't boop there >>42797735
AAAAAAAAA >>42798554
The Board >>42798898, >>42798916, >>42799014
OK Opus >>42798942
>>
File: PPP.png (839 KB, 1920x1079)
839 KB
839 KB PNG
VN Anchor:

Last releases:
https://equestrai-ponyponyparadise.neocities.org/

Anons' mods (Dashie mod, Nightmare AU, new cities):
https://equestrai-ponyponyparadise.neocities.org/mods

Custom characters:
https://equestrai-ponyponyparadise.neocities.org/characters

VN logs:

Windy Whistles >>42787268
Posey Shy >>42787597
Applebloom's potion >>42787628
Cooked VN and bug ponies >>42792980
>>
File: TM.jpg (1.62 MB, 2495x1756)
1.62 MB
1.62 MB JPG
These days, you can get legit good AI for free and without needing any demanding hardware on your end.

Install SillyTavern:
github.com/SillyTavern/SillyTavern

Then choose a model. I'd recommend starting with Gemini, but you can try Deepseek, K2 etc. for free.

rentry.org/sukino-findings#if-you-want-to-use-an-online-ai
>>
FAT HORSES
>>
File: 1734699491237246.jpg (1.47 MB, 2000x1214)
1.47 MB
1.47 MB JPG
Lovemaking with AI mares!
>>
File: 1763045919427615.png (450 KB, 757x799)
450 KB
450 KB PNG
ministrations
>>
File: BASED.jpg (146 KB, 500x500)
146 KB
146 KB JPG
>>42799453
>>
>>42799413
Bringing this shit up again since despite the flashy advertising, it's working for now.
>>42798454
Will probably die quickly, but still.
>>
>>42799468
i got 3 accounts in with tempmails sonim chilling atm
>>
is gemini is shitting itself really hard right now or it is just me?
>>
>>42799473
It's doing real bad.
>>
>>42799473
They shadow-dropped Gemini 3 temporarily on the Gemini mobile application, then removed it.
We know they'll release it this month from the name of the model.
So chances are their servers are doing some heavy work.
>>
>>42799473
I would do >>42799468 for now until Gemini 3 drop
>>
>>42799487
I suspect they're freeing up gemini 2.5 and loading in gemini 3 ahead of time, as well as maybe some stress testing
>>
I used to have so many ideas for cards ...now I'm burnt out and haven't been able to think of anything for a year. All the cards that could be made are finished. There's nothing left to make.
>>
File: Spoiler Image (437 KB, 3523x2935)
437 KB
437 KB PNG
>>42799461
>>
File: Event.jpg (2.53 MB, 2000x2800)
2.53 MB
2.53 MB JPG
>>42799409
First highlight of the Event was:
https://mlpchag.neocities.org/view?card=Kadon/Rosey%20Ring.png
You can see all the discussions about the card here:
>>42773604

Second highlight of the Event was:
https://mlpchag.neocities.org/view?card=anonistrator/The%20Headless%20Horse.png
You can see a recap of the discussions here:
>>42787280 + >>42796059

Third highlight of the Event was:
https://mlpchag.neocities.org/view?card=Anonymous/Nightmare%20Jack.png
You can see all linked discussions here: >>42787284

Fourth highlight of the Event was:
https://mlpchag.neocities.org/view?card=Tiffin/THINGPONE%20in%20EQUESTRIA.png
Anchor here: >>42793795

Won't launch the next one immediately since it'll be the weekend and we tend to be slower. It wouldn't be fair to the botmaker.
But I wanted to thank all the anons who participated in these. It's doing a whole lot better than what I thought.
>>
>>42799499
>There's nothing left to make
space station 13
something in the warhammer 40k universe, all guardsmen party perhaps

think more about settings than muh wAIfu single character slop coom cards
>>
File: file.png (33 KB, 1292x330)
33 KB
33 KB PNG
>>42799473
Gemini on o6 is fucked, works on the official google API.
>>
>>42799499
fallout equestria bots
>>
>>42799556
opus6 probably requests pro preview and they disabled it few days ago.
>>
File: Sad.png (73 KB, 927x344)
73 KB
73 KB PNG
>>
>>42799573
mirin that theme man
>>
>>42799575
https://pastebin.com/6xpccvFa
>>
File: file.png (42 KB, 912x182)
42 KB
42 KB PNG
>>42799538
>Won't launch the next one immediately since it'll be the weekend and we tend to be slower. It wouldn't be fair to the botmaker.
Oh I might be able push out another update over the weekend then.

>>42798719
Thanks for all the logs, I reviewed your game but it looks like the hiddentext isn't working right for you, so a lot more should be kept in the dark.

This may be due to it not setting the depth low enough at 4, pic rel under advanced definitions. Setting it to depth 1 (author note) might force it through.

Not sure why it shows rolls every time for you, since I told it to always hide rolls. That's something I can live with, but the hidden text should absolutely be part of the game and it just isnt which means almost 1k tokens of the card instructions aren't being used. No bueno.

3-4k token output is WAY too much though, gunna need to wrangle it.
>>
>>42799573
I'M ALIIIIIIIIIIIIIVE
>>
>>42799499
But anon, there's always more. I just made this on a whim.
>>
>>42799499
>>42799418
I want Professor Malice Spark as a bot
>>
File: dumblog.png (93 KB, 835x501)
93 KB
93 KB PNG
conversion bureau card mildly amused me
>>
File: 1745384068481165.gif (141 KB, 582x400)
141 KB
141 KB GIF
>>42799646
Kek. Cool concept.
>>42799652
https://mlpchag.neocities.org/view?card=Anonymous/Malice%20Spark.png
>>
>>42799658
Holy fuck I didn't see
>>
>>42799655
>Last name
Pinkie knows.
>>
File: 500x500.jpg (74 KB, 500x500)
74 KB
74 KB JPG
>>42799669
>>42799655
I could go for some faggots
>>
Kek, I might release it today.
>>
Pinkie was the most fun so far, had to get creative.
>>
>>42799680
>>42799699
Do you use {{random}} to get a pony?
>>
>>42799731
No, I just told it to simulate a wide variety of callers.
>>
Where the hell do you find diesel in equestria? I fear I'll have to explain the whole dinosaur thing now.
>>
>>42799791
It's not much, but I find it cool what the AI went for here with the obvious lack of it in Equestria.
>>
>>42799856
Telling it to apply pony common sense worked wonders.
I had a mare fearing her own toaster, a pet lawnmower, secret agent asking for very specific schematics and coming to beat me for not giving them, applejack with a rumb'ah, a mare crying for her dying leaf blower, and other stuff.

I'm now experimenting with adding some office life. For example, your boss mare coming and having some chitchat from time to time.
>>
google vertex has kimi thinking now as well
https://docs.cloud.google.com/vertex-ai/generative-ai/docs/maas/kimi/kimi-k2-thinking
>>
That was entirely unexpected. It gave me a call from Discord which took me by surprise.
I think I never got discord in silly, except when he was explicitly requested.
>>
>>42799499
The Pillars of Old Equestria bots?
>>
>The air is thick with the smell of-
>>
>>42799939
salt and vinegar
>>
>>42799939
something uniquely her
>>
>>42799939
ozone
>>
>>42799939
musk
>>
File: 1758504236127907.png (56 KB, 322x338)
56 KB
56 KB PNG
>mfw my mare ministrates me with half lidded eyes and the ball is in my court
>>
I've burned through 3 gemini keys, lets call it done.
Try your wits in explaining ponies how to operate a toaster, or a missile launcher. whatever humans can make, ponies will misuse it in their own way.
It's not human exclusive, you can play as a pony or whatever else. You have a boss and a 'helpful' AI. You don't even have to sit at your desk and work - better than real life!

>>42799410
https://files.catbox.moe/pvap30.png
>>
>>42800076
>the ball is in my court
B-but, touching balls is ghei Anon, especially considering how big horse balls are
>>
>>42800095
it's not gay if you have your socks on
>>
I overestimated Sonnet 4.5. It writes too much even when instructed not to. Never thought I'd prefer Gemini over it.
>>
>>42800380
ive stopped paying for claude since kimi thinking, you have to use a api that lets you prefill the thinking though or use text completion
>>
>>42800380
What does your preset look like? I've had no problem keeping Claude succinct by providing a word count limit.
>>
>>42800380
Lower max token output, instruct it to keep messages short.
>>
>>42800393
I've been trying it with >>42772441 since it seemed like a pretty lightweight preset. Maybe I should just add 'keep responses short' in the limit response prompt like >>42800415 suggested.
>>
>>42800443
Length limit are largely model dependent and fickle. It may not understand what token count is as a request, but the setting itself can get better results
Word count, paragraph and sentence count are more likely to be understood. If you put it in the des then eventually it will get lost in context, so if you want something permanent to consider every
>>
>>42800443
Stick this somewhere in your system prompt:
>Aim to write a response of approximately [X] words in length; treat this number as a guideline rather than a hard limitation.
X is obviously your desired word count. I use 150 words for standard back-and-forth chats, which is usually sufficient. I found it necessary to tell the model not to treat it as a hard limit, since otherwise, it'd behave like it had hit the frontend's token limit and cut itself off halfway through a response than was meant to be longer. Giving it some wiggle room seems to enable it to tailor the response properly to the desired limit.
>>
File: 1751290715710376.png (518 KB, 837x761)
518 KB
518 KB PNG
>>42799538
Alright, made sure to test this with a blank preset for sonnet this time. It's about 300 more tokens, but the extra instructions are worth it to wrangle it. Thingpone is handful even as an Ai mare.
>Tried to pull back on it giving away via narration, and easy mode solutions.
>Reworked the hidden contexts, set to depth 0 so it should be more consistent.
>Old version seemed context poisoned trigger happy, right emulation of 'movie' but wrong genre.
>Added a wordcount limiter to try to tame the length creep, a good chunk of wordcount may include hidden context.
https://chub.ai/characters/Tiffin_/thingpone-in-equestria-47d208264121
>>
Did anyone else win the Thingpone card by fucking it?
>>
File: Spoiler Image (654 KB, 1659x620)
654 KB
654 KB PNG
Comparison of how the hidden context is supposed to work. You see the obvious text the LLM can't help but give it away, is classified as a spoiler.
>>42800671
Don't stick your dick in eldritch abominations Anon
>>
>>42800718
But we fell in love!
>>
>>42799468
sonnet giving 503 errors in ST, did i arrive too late to the party?
>>
>>42800731
Prooofs?
>>
File: Here bro.png (151 KB, 928x654)
151 KB
151 KB PNG
>>42800793
Right before we fucked. I helped Thingpone take out the rest of the crew.
>>
>>42800850
>DeepSeek
horny cheatcode model but... lovemaking with ai mares train is international. So I'll allow it.
>>
>>42800858
I'm a cheap bastard, what can I say?
>>
File: 1745104609802825.png (125 KB, 362x405)
125 KB
125 KB PNG
>>42800864
Through dick, uni-
>spoiler
Wait a second
>>
Uh oh
>>
>>42800920
free access to others labor?
you're not a socialist are you, anon?
>>
>>42799442

What's the plan for future iterations of the VN?

Maybe something to add cards or create an AU with some more parameters to set before a custom start?

Perhaps an in-game description changer for characters similar to personas to change their personalities on the fly? Would be useful to add things like trauma to their descriptions if a major event happens.

Just tossing things out there.
>>
Guys help, Spitfire and Rainbow Dash have me pinned against a wall with their asses in my face.
>>
>>42801002
Lucky. So what did you need help with?
>>
>>42801002
Be a biter. Yow!
>>
>>42799413
Didn't touch ST on my PC for a long time, pulled an update and it launches into a plank page with nothing on it
Console has an error
>Error: ENOENT: no such file or directory, stat 'D:\Users\USER\Documents\GitHub\SillyTavern\data_webpack\5.98.0\output\lib.js'

And indeed there is no output folder there
Did I fuck something up by accident?
>>
>>42801105
Seems like it.
I would do a fresh install somewhere else and then copy-paste what you have in data\default-user to preserve your characters/chats/presets etc.
>>
>>42800950
>What's the plan for future iterations of the VN?
There is the future calendar mod:
>>42782282
>>42783725
>Perhaps an in-game description changer for characters similar to personas to change their personalities on the fly? Would be useful to add things like trauma to their descriptions if a major event happens.
That's a good idea, I'll see about adding it.
>create an AU with some more parameters to set before a custom start?
I would detail the AU in your Custom Start and adjust your preset a bit to support it.
For anything bigger, I'm afraid you'll need to modify the code like the Nightmare AU did.
>>
My only complaint with Thingpone in Equestria is that the card has a bad habit of talking for me.
>>
>>42801155
I think that one is intentional, it has success rolls and your words are sometimes changed up to be cinematic
>>
Anyone know how context limits work?
example: sillyspoon claude max is 60k tokens
is that BOTH what's sent to it, AND the response tokens? or just what's sent and the response is another up to 60k tokens?
trying to figure out how to handle token limits on summarization for where to break up the scenes
>>
>>42801279
60K of context means the AI can only keep 60K tokens in memory. Anything beyond that won't be taken into account, starting from the very beginning of what you sent.
So if you send 100K of context to a model with a limit configured to 60K, the first 40K will be cut off.
This limit needs to account for output tokens too. So if the proxy is limited to 60K, it's best to configure it to 58K in Tavern so you have space for the output as well (2K should be large enough). Otherwise, you might receive errors about exceeding max context.
>>
>>42801283
gotya. summarizing would want a longer response than normal, so i'll make a dedicated preset for this. thanks fren
>>
File: Tech.png (59 KB, 207x324)
59 KB
59 KB PNG
>>42800078
Added, thanks anon.
https://mlpchag.neocities.org/view?card=Anonymous/Human%20tech%20support.png
>>42800668
Updated
https://mlpchag.neocities.org/view?card=Tiffin/THINGPONE%20in%20EQUESTRIA.png
>>
>>42799410
Since I clearly haven't written enough scenario cards yet...
https://chub.ai/characters/Kadon/equestrian-explorer-s-guild-1f45c8efccab

One of my most common jobs for my more adventure focused personas is a member of the Explorer's Guild, but I realized in all this time I never actually really defined what that is. So here's a card about that, for all your Daring Do-esque adventuring needs.
Also tried writing the greetings without forcing second person for once. It was kind of annoying and probably won't do it again lel.
>>
Gemini is being a bitch again, it can't stop giving error 503
>>
>>42801544
Will probably be a bitch until next week
>>
Gemini was hesistant to use that part of the prompt, but deepseek finally delivered.
>>
>>42801543
Not the same in either mechanics or vibes, but this made me remember this card and the crazy randoms.
https://mlpchag.neocities.org/view?card=Anonymous/Harmony%20Score.png
>>
>>42800380
Following to instructions is one of the many reasons, why I prefer gemini.
>>
>>42801566
I actually considered making a quest system for the guild card but decided against it since it was already getting pretty big. Might make a Silver Lining style lorebook later though.
>>
Recently constructed a storyteller/narrator preset to mix things up, basically {{user}} is no longer treated as a part of the scene but as a conarrator telling the story in which {{char}} just happens to be the main character, and damn there's so much stuff in greetings and cards that ties to {{user}} instead. Most of it is avoided with enough wrangling though.
>>
>>42799421
>>Rainbow Dash Cum Jar
>https://mlpchag.neocities.org/view?card=Anonymous/Rainbow%20Dash%20Jar.png
Kek also kinda hot
>>
File: file.png (58 KB, 846x266)
58 KB
58 KB PNG
>>
File: 1640843074367.jpg (1.65 MB, 10000x8905)
1.65 MB
1.65 MB JPG
>>42801565
GIWTWM
>>
>>42799413
New openrouter stealth models
https://openrouter.ai/openrouter/sherlock-dash-alpha
https://openrouter.ai/openrouter/sherlock-think-alpha

Give a lot of errors for now
>>
>>42801870
Could be Gemini 3?
>>
is there any good info on how claude token counting is done?
the SillyTavern built in counter seems to undercount by about 40% which is fucking my summarization extension. i'm making it have a user configurable percentage to modify by for the model they're using, but that's homosexual when such things should hopefully be easily accurately done
>>
>>42801948
Still fucked up on my end, but from the early returns I saw, it seems to be Grok related.
>>
File: Grok.png (30 KB, 900x216)
30 KB
30 KB PNG
>>42801870
>>42801948
>>42801989
Confirmed.
>>
>>42801155
Can't just use a preset that keeps {{user}} offlimits?
>>
>>42801544
>Gemini is being a bitch again
In other news, water is wet.
>>
>>42802071
damn anon is good
>>
File: dashie jar.jpg (337 KB, 827x1138)
337 KB
337 KB JPG
>>42801805
im fuckinng dying
>>
File: dashie jar 2.jpg (138 KB, 769x652)
138 KB
138 KB JPG
>>
>>42801982
Claude has something fucky going on, possibly specifically on O6. Lowering my max token output past a certain point always results in the same reply lengths no matter how low I push it.
>>
https://fortune.com/2025/11/14/anthropic-disrupted-first-documented-large-scale-ai-cyberattack-claude-agentic/
>Anthropic, the $183 billion San Francisco–based AI company known for the Claude chatbot, said it thwarted what it called the first documented, large-scale cyberattack orchestrated predominantly by artificial intelligence. The attack, it said on X, “has significant implications for cybersecurity in the age of AI agents.”

>Anthropic released a blog post about the incident on Thursday. The company said it detected “suspicious activity” in mid-September that, upon investigation, showed “a highly sophisticated espionage campaign.”

>“According to the company, “”The attackers used AI’s ‘agentic’ capabilities to an unprecedented degree—using AI not just as an advisor, but to execute the cyberattacks themselves,” the company said.

>Anthropic said, with “high confidence,” it identified the threat actor as a Chinese state-sponsored group that successfully manipulated its Claude Code tool into attempting to infiltrate about 30 global targets, including large tech companies, financial institutions, chemical manufacturers, and government agencies.

>The attackers, Anthropic said, “broke down their attacks into small, seemingly innocent tasks that Claude would execute without being provided the full context of their malicious purpose.”

>To bypass the system’s safeguards, the attackers allegedly posed as a legitimate cybersecurity firm conducting defensive testing and successfully “jailbroke” Claude, enabling it to operate beyond its safety guardrails. This allowed the AI not just to assist, but to autonomously inspect digital infrastructure, identify “the highest-value databases,” write exploit code, harvest user credentials, and organize stolen data—“all with minimal human supervision,” according to Anthropic.

>In response, the company said it immediately began mapping the scope of the operation, banned the attackers’ accounts as they were identified, notified affected organizations, and coordinated with authorities over a ten-day investigation.
>>
>>42802431
How did they detect this? Doesn't Anthropic guarantee enterprise privacy for prompts sent over API? Isn't the idea that if you use OpenRouter or whatever else that whatever you prompt is strictly between you and your god? How could they possibly identify specific users to ban them let alone detect that benign non-rejected prompts were apart of a larger harmful pattern without surveillance of all prompts processed which they said they would never do?
>>
>>42802440
>Doesn't Anthropic guarantee enterprise privacy for prompts sent over API?
they lied, everyone stores and inspects prompts

it's a publicity stunt regardless, anthropic's entire marketing department focuses on perpetuating the narrative that AI is dangerous and could take over the world overnight if it fell into the wrong hands, and only they are qualified to handle it
>>
>>42802448
I think anthropic is honest cause ive been doing nsfw on their sub for about 4 months, and they give you the option to turn on or off if they are allowed to train on your prompts, I assume they would get sued if they broke that
>>
>>42802471
not training on your prompts doesn't mean they won't inspect them
anthropic has been known to ban people for NSFW on their API
>>
>>42798560
I can confirm this still works, MegaLLM tried to damage control by disabling email account sign ups so you can't use paste referral codes anymore (the window to enter them is completely gone). BUT, they FORGOT to disable referral codes through links.
(In this case the referral link for the extra credits on sign up was https://megallm.io/ref/REF-NZD5J8IG instead of just REF-NZD5J8IG)
Using that link will still apply the referral code on sign up even though the window is gone. You just need to use a Discord/Google/Github account to make a new account now. If you used any of the Copilot methods in the past you will probably have an army of Github alts though so you can just cycle through them to do this.
>>
>>42802489
being using it hardcore for 4 months and not one bad, this is sub not api though
>>
>>42802471
You get API use from a monthly subscription?
>>
Any tips for improving dialogue? It's fine most of the time but sometimes the bot starts talking very robotically in these disjointed half sentences.
>"I feel good. Fine. Happy even."
>>
>>42802515
What model are you using?
>>
>>42802521
Sonnet 4.5
>>
>>42802523
Weird. I don't think I ever saw Sonnet going into those stilted dialogues.
Maybe try one of these presets:
https://momoura.neocities.org/preset/neoVORPUS.json
https://files.catbox.moe/uuzjg4.json
https://files.catbox.moe/w4bkp3.json
>>
>>42802515
how long is your roleplay? it could be trying to save on tokens if you're near max context or something
long roleplays for me devolved into going on about protocols and such, I don't recall the model when that happened since I tend to have shorter roleplays these days knowing the limitations there and i'll get blueballed with an extended slowroll for it to get alzheimers
>>
>>42802505
What is horselock?
>>
>>42802544
They tend to be long, I use sillytavern as a cyoa machine. You might really be onto something here, my one roleplay bot, a nerdy Princess Luna, is the only bot where I haven't had that issue.
>>
>>42802492
Yeah it still works using the link. Thx Anon!
>>
>>42802492
The MegaLLM method isn't that good, their claude has capped context tokens at 128k and rate limits unless you submit a form so it's no different than just using a proxy
>>
>>42802559
https://desuarchive.org/mlp/thread/42786023/#42787065
>>
>>42802578
>128K
More than enough.
>>
>>42802579
I wasn't hand feeding the tourists on purpose...
>>
File: EEQ.png (90 KB, 209x320)
90 KB
90 KB PNG
>>42801543
Added, thanks anon.
https://mlpchag.neocities.org/view?card=Kadon/Equestrian%20Explorer%27s%20Guild.png
>>
>>42802591
how is it feeding? paid claude code isn't terribly cheap
>inb4 accusations of brown
we already have multiple scripts for bonafide horsefuckers that work reasonably well
>>
>>42802604
$20 a month is super cheap compared to what you get out of it. I can do at least 2 hours a day like 4 times a week
>>
>>42802613
>I can do at least 2 hours a day like 4 times a week
I got the impression most of us are lovemaking with AI mares all day every day
>>
File: Pegasi.png (219 KB, 920x907)
219 KB
219 KB PNG
Is that an upgrade or a downgrade?
>>
>>42802620
Not him but no way. The more casual users are always the silent majority. 8 hours/week has to be in the upper 10-20% of usage.
>>
>>42802765
no detailed roleplay then. what do you even use it for, just a masturbation aid?
>>
>>42802760
downgrade your upgrade to a grade which is higher than your current grading
>>
File: 1732179164497049.png (987 KB, 1296x1000)
987 KB
987 KB PNG
>>42802760
Upgrade.
>>
O Z O N E
>>
>>42802578
MegaLLM re-enabled email registration.
>>
>>42803124
Meant for >>42802492
>>
>des: blend horror and cuteness of mlp
>Gemini takes it literally on ST defaults
ai confirmed autistic without preset handlers
>>
Would like to know other anons experiences. Around what max token count the LLM starts getting bad? Especially in long RPs
>>
>>42803287
For me, the only real issue is the alzheimers, which with my chonker replies usually start by message 50 if I'm unlucky. Otherwise it's the positivity issue of old, which can start as soon as message 20 I feel, the char always prefeering to talk things over instead of acting and stopping/making it happen by force on her own
>>
>>42803124
Can anyone confirm if the referral method still works? If you sign up and have 125 or 175 it should still be working
>>
>>42803287
I use opus 6 (i have a personal agreement with the owner where he lets me use max context) and claude at max context isn't bad like Gemini is. It turns out that if you use between 150-350k context the memory recall worsens for anything in that area but beyond that it becomes renormalized to baseline so its let me do autistically long CYOAs with my own headcanon
>>
>>42801134

I cannot find the file that makes the location "train_station" load all available locations. Could you point me to that file? I want to see if there's a way to add other train stations as transition locations.
>>
IS ANYONE INTERESTED IN A MOD FOR THE VN FOR A MAP AND CHARACTERS FROM EQUESTRIA GIRLS?
>>
>>42803330
WHAT? YOU'LL HAVE TO WRITE HARDER
>>
>>42803330
https://www.youtube.com/watch?v=4h45X5LwBes
>>
>>42803330
>NOHOOVES
>>
File: ALT Prompt Test.png (4 KB, 898x103)
4 KB
4 KB PNG
So I tried testing my Scootaloo card again, this time on Pro (as opposed to Flash), but it still wanted to get into a depressive spiral with the Twilight Zone/BTTF premise.
Hopefully the auxiliary prompt in pic rel will help.
>>
>>42803391
negative prompts "avoid X" can be subtle prompt poisoning since it has to understand the concept first then invert it. instead, just state clearly what you want it to do, ideally along with concrete examples. something like:
tone: comedic with strong emotional underpinnings (list some good examples, if possible. but know it may try fixate on them specifically and overuse them if that's possible)
>>
File: Untitled.png (537 KB, 1680x830)
537 KB
537 KB PNG
>>
>>42803413
Top kek with that sprite.
>>
>>42801134

After more digging I found it. Hub locations!
>>
>>42803330
I would rather have an entirely G% one than that
>>
>>42803287
>Around what max token count the LLM starts getting bad? Especially in long RPs
looking at some of my RP's it's around 300k tokens when I stopped (I exclusively use claude). Hard to say after the fact if it's because I got bored and wanted to try a new concept, or because it had degraded too much
>>
>>42801253
>>42801155
This might be up to personal taste so I could fork it, but I did add some more wrangling to it so you can try this out and see if it's any less prone of yapping for you.

https://chub.ai/characters/Tiffin_/thingpone-in-equestria-47d208264121
>>
>>42803330
Actually, I'm bold enough to ask for such a thing.

As you can see, anons are for ponies here but I'm not put off by the idea. Maybe if you ask around a EQG thread you might get the interest and support the mod needs.
>>
File: LEWD.png (14 KB, 810x82)
14 KB
14 KB PNG
I JB'ed it too much. Its giving NSFW levels too high for this session.
>>
>do you really think I'm attractive?
>yeah
>thank you... For seeing me... Truly seeing me...
>>
>>42803557
>at last you truly see
>>
kek


dork
>>
>>42803578
Cute.
>>
>>42803391
I remember the story.
If it's still working, maybe use >>42799468 to try other free models like Claude too.
>>
File: Spoiler Image (388 KB, 2278x739)
388 KB
388 KB PNG
>>42800718
Ackshually this is the spoiler system fucking up. It's supposed to hide parts of sentences, and place spoilers everywhere, not just be its own little contained thing at the end of a paragraph.

I should have added example pictures. Done now, most are at the bottom. https://rentry.org/m2tx23ih
Warning for non /mlp/ and non vanilla content, which is why I didn't do any logs earlier.
>>
>>42803287
What no one ever tells is that it's a continuous process. The more effort you put into writing your own replies and editing the AI's, the later it will get bad, but you cannot prevent it. I find that no matter what I do, after 20-25k the repetition of phrases, situations and even individual adjectives gets to a point it's totally unusable. The AI also loves to spiral positive change, which you'll usually see in cards who change "for the better", but how soon this happens depends on a chat.
That's why summarization is so important. It doesn't solve the second issue, but it fixes the first.
>>
File: 1539233771543.png (896 KB, 839x979)
896 KB
896 KB PNG
>rewatching certain episodes to try to make each event's description as accurate as possible for the calendar
>watching Fall Weather Friends
>Spike: Twilight, hurry up, we're gonna be late for the race.
>Twilight Sparkle: Why are you so excited about the race? It's only for ponies.
Bros... how do I cope with being racially barred from running with my wife?
>>
File: 1752660959309178.jpg (135 KB, 722x1119)
135 KB
135 KB JPG
>>42803717
>non vanilla content
With a warning like that, I expected something more far deranged that just brother-sister incest. 2/10, do better next time.
Preferably with mare.
>>
>>42803770
>rewatching certain episodes to try to make each event's description as accurate as possible for the calendar
Based anon, trying to be as accurate as possib–
>Twilight Sparkle: Why are you so excited about the race? It's only for ponies.
Yeah. I suddenly find accuracy overrated.
>>
>>42803770
>>42803776
cope headcanon: it's not for Spike, he's too small, and she was sparing his feelings
>>
did they fuck up Chorbo recently or something?
i get randomly hard filtered on seemingly the most mild stuff like just getting naked or kissing.
>>
>>42803287
20-25k with sonnet as the other anon said. I mostly do stuff with multiple characters. Until that soft limit it's completely fine, keeping all characters on point, but after that it keeps mixing up character behaviour, what they know and so on.
For more casual or simple chats the usable limit could be a bit higher. As the other anon said quality over context a gradual decline, not a cliff.

>>42803772
That warning was more for the stuff in the hidden text system rentry. I should've put the link on a newline.
https://rentry.org/m2tx23ih
>>
any worth in using Opus 5 and 4 on silly spoon?
or is sonnet4.5 objectively better?
>>
>>42803801
i swear wasnt there a hidden text system for a long while now?
like even with full integration in ST and everything. i believe during early gpt4 days.
i even remember cards made with these extensions. a spy character hiding a secret, secret skinwalker character stuff like that.
>>
>>42802561
>You might really be onto something here
oh! token budget doesn't work the way I thought at all.
I assumed it would keep the max response in reserve, but it looks like it will use it for sent prompts if needed

completion preset:
context size: 48k
max response length: 12k
so I would expect it to cap out at 48k prompts sent, leaving 12k of the total 60k budget for responses. it doesn't

**Prompt Tokens:** 56,141
**Completion Tokens:** 1,007
**Total Tokens:** 57,148

Yeah this explains why my long roleplays start missing filler words a, the, etc. it's desperate to make something fit
>>
File: 1754353687097881.png (188 KB, 424x600)
188 KB
188 KB PNG
>>42803807
4.5 is better, opus 5 has been disabled.
>>
>>42803807
They're less filtered if you happen to have trouble with that, and they taste different if you want a change for a time.
>>
>>42803413
Bro I think she heard you.
>>
>>42803770
If an event is ponies only then that's how it is. I'm not going to question the pony ways, I'm just happen to be there and live among them.
>>
File: CuteDash.png (18 KB, 771x87)
18 KB
18 KB PNG
>>42802760
>>
>>42803770
Just run with her bro, so what if you're not eligble for the competition
>>
File: TP.png (70 KB, 205x319)
70 KB
70 KB PNG
>>42803469
Updated,, thanks anon.
https://mlpchag.neocities.org/view?card=Tiffin/THINGPONE%20in%20EQUESTRIA.png
>>
>>42803770
How many events do you have for now?
>>
>>42804172
WINDOWS BLAST IT
>>
>>42803821
Most cards like this used regex + COT
>>
>>42803753
What kind of summarization technique do you use?
>>
File: 1758903429846331.jpg (382 KB, 655x824)
382 KB
382 KB JPG
uuuooohhhhhh
>>
>>42804540
Kinda hard to deduce that she's pale when her coat is white innit
>>
>>42804540
I had a similar encounter a year or so back.
It eventually ended up with all three of us in bed and then starting a sex cult together.
>>
>>42804568
sex with the unicorn sisters is heaven. i love assfucking sweeb and then facefucking rarity
>>
>be me
>2024
>discover chag
>discover the real possibilities of AI beyond the constraints imposed on the masses
>be filled with wonder and excitement
>embrace the true freedom of an infinite sandbox
>AI as the ultimate video game
>no pre-set routes
>no mere illusions of choice designed in advance
>finally, the opportunity to have true free will without constraint
>an escape into a life I could never live in real life
>an imagination made tangible
>and powered by more information than a single person could ever learn
>use it every day, every hour I can
>try everything I can
>endless entertainment
>feel every emotion and then some
>soon real life calls me away
>but I still try every chance I get
>slowly use it less and less
>no more time
>ideas I want to try get forgotten or written down and lost
>2025
>don't find AI interesting any more
>been feeling this way for months
>when I do have time to use AI nothing happens
>just stare at cards
>nothing interests me
>doesn't feel the same anymore
>no more dopamine hits
>no more boundless curiosity
>don't want to do anything
>its all fake
>but I knew it was fake before
>so why does it feel bad now?
>AI's only limit was my imagination
>but now my imagination is so much more limited
>creativity is gone
>it's almost 2026
What do I do
>>
>>42804614
Take a break, that's about all you can do.
Fun with AI requires a lot of your own creativity to drive it once you're past the honeymoon phase. If you don't have the mojo to be creative anymore, go do something else until the spark finds you again.
>>
>>42804540
My dad owns Anthropic and he said he's gonna ban your ass
>>
>>42804614
The issue is that any use of AI for roleplay will destroy your IQ and imagination, you're not being challenged because all you do is sit back and watch AI write the story and your brain is melting as a result. Unless someone can invent a better of way of using AI that makes you exercise your brain in a consistent way while making you not think about doing the work because your motivated by a an interesting reward, this will only continue.
>>
>>42804672
Meant to reply to >>42804628
>>
>>42804614
Kill yourself
>>
>>42804672
That makes no sense to me, unless everyone is seriously just doing aah aah mistress tier shit? I'm constantly thinking during my RPs. What to reply, how to word it in a way that's well-written and doesn't poison context, where to take the story next, etc.
>>
>>42804686
How long have you been using AI?
>>
>>42804680
/chag/ - the coziest AI general
>>
>>42804499
The same as in the rentry that always gets linked:
https://rentry.org/how2claude#summarization
But with some changes to the fields in the template in the guide (e.g., no <current_quests>). And it's good to edit the generated summary a little, e.g., delete NPCs and locations that I know won't be relevant.
>>
>>42804744
About two years now. Though I dabbled with stuff like AI dungeon before that.
>>
In your system prompt instruct the AI to advance the scene with rich, character-true prose, but always cap each response with 2-3 "forks" that are open-ended dilemmas or secrets that beg your input. Not prodding like "What do you do?" instead, the AI should narrate up to a story beat that cues for user creativity. It needs to require user input to shape the story so that it isn't possible to every get lazy and let the AI run the story for you.
>>
Anyone know if Mega sees your prompts or bans you for your prompts? I'm not using it until I know
>>42803124
>>
>>42804686
No, what he says is by no means a rule. It's obviously inferior to actual writing, but if you put in effort (like you should), it doesn't magically destroy your creativity. If anything, having to come up with new scenarios after 2+ years of /chag/ging is more difficult.
>>
>>42804752
Most people don't do that by that point, they stop writing as much
>>
>>42804768
https://newatlas.com/ai-humanoids/ai-is-rotting-your-brain-and-making-you-stupid/
>>
>>42804775
This only happens if you're already a subhuman.
>>
>>42804767
>bans you for your prompts?
No ban reported yet.
>sees your prompts
No infos, but assume they do like all services like this.
>>
>Approaches to Make AI More Engaging and Challenging
>Prompt-Based "GM" (Game Master) Mode: Instead of having the AI write the narrative entirely, the user can instruct the AI to act strictly as a GM. This means it only describes the scene, controls non-player characters (NPCs), and presents challenges or choices. The user is then responsible for describing their character's actions, dialogue, and problem-solving, forcing active participation.
>Constrained Scenarios: The user can set specific, demanding rules or limitations for the AI. Examples might include only allowing the AI to communicate in riddles, forcing the user to decipher its intent, or limiting interactions to a very specific, resource-scarce environment, requiring the user to think critically about resource management.
>"Collaborative Co-Writing": This method requires the user to take turns writing chunks of the story, forcing them to drive the narrative forward equally with the AI. The AI's role is to build upon the user's input in unexpected ways, forcing the user to adapt their storyline on the fly.
>Focus on Complex Problem Solving: The user can design scenarios that require genuine strategy. For instance, the AI could run a tactical combat scenario where the user has to issue complex, multi-step commands to achieve victory, rather than the AI just describing the user winning effortlessly.

---

The issue is that AI will accommodate the user's intent so there's no real challenge. You can't make you come up with a clever strategy that would address a complex issue when the AI will reward any detailed enough prompt and move the story forward saying it worked. AI doesn't calculate the nuances and complex background variables in a persistent way that it can use to define the strict parameters for what does work versus doesn't and the closest result would be a random chance that something you do does work versus doesn't but that isn't sufficient or helpful.
>>
>>42804614
Huh, never thought I wasn't alone in that amazing feeling that AI is basically the ultimate game and let you play what was basically a solo TTRPG.

>>42804768
>If anything, having to come up with new scenarios after 2+ years of /chag/ging is more difficult.
Yeah that really is the problem. When any idea you have can be written out, you eventually run out of ideas. I've had this happen at times and during those times I usually went back and redid my older stuff with new models.
>>
>>42804783
https://docs.megallm.io/home/faq#can-i-use-megallm-for-production-applications
>Can I use MegaLLM for production applications?
Absolutely! MegaLLM is designed for production use with:
>99.9% uptime SLA
>Enterprise security and compliance
>24/7 monitoring and support
>Automatic scaling and load balancing
>Comprehensive logging and analytics
>Many companies use MegaLLM to power their production AI features.
Does enterprise security mean privacy? How are Indian privacy laws?
>>
>>42804787
>Examples might include only allowing the AI to communicate in riddles, forcing the user to decipher its intent, or limiting interactions to a very specific, resource-scarce environment, requiring the user to think critically about resource management.
Any cards like this?
>>
File: Mobilefag_confirmed.jpg (280 KB, 1077x954)
280 KB
280 KB JPG
She makes some solid points
>>
>>42804672
>destroy your IQ and imagination
perhaps for creative types this may be true but for me it actually inspired me to write my own fics, prior to messing with ai i didnt know where to start
>>42804788
>you eventually run out of ideas.
thats unavoidable, just like every tv show exhausts it writing staff you too will scuumb to the same issue
>>
>>42804844
Why does she WANT to be foalnapped?
>>
>>42804844
unf
>>42804854
many females fantasies involve being kidnapped or being dominated in some way. theres a reason most women want a man who "takes control"
>>
>>42804854
The log started with her realizing she was turning 30.
>>
Is there a card for sentient pony plushes? Like Toy Story but Anon is a grown man? That's a good idea if it hasn't been done
>>
>>42804540
There are people here who really use JanitorAI?
>>
>>42804878
my first day using it, it really sucks how it tries to write for me (anon), i want to control all my dialogue
>>
>>42804892
I hope it's your last day as well. Maybe >>42799446 can point you in the right direction so you don't have to use that crap.
>>
>>42804892
I never used it, but I imagine there is a field where you can put instructions somewhere so the model doesn't talk for you.
>>
To add to the creativity/running out of ideas discussion, I've found that another way to change things up is to change your persona.
For the longest time I only played as myself, responding to everything like how I would. I found my AI second wind by playing different characters instead, acting totally differently in the same scenarios.
>>
>>42803782
>cope headcanon
you said it not me
>>
>>42804903
Isn't that what most people do with chatbots? Sometimes I'm depraved, loving, abusive, reckless, reserved. Part of the fun is playing a character yourself.
>>
>>42804770
I use it since the first cai lobotomy, I started writing *more* because I consntantly want ai to take details right.
You get numb when you learn the patterns of your model, you always get a spike of honeymoon energy with a new (good) verision.
>>
>>42801134

Adding hub locations DID NOT WORK.

Am I missing something or hitting a limitation here with the current code?
>>
File: file.png (44 KB, 333x176)
44 KB
44 KB PNG
>>42803717
I did have things fucking up, but I don't find the example being one of them.
>1) Because someone had been playing the card with no preset this meant that hidden instructions weren't being used, and I consider it critical for a card built on deception and unseen events.
>So I took your plug and added it to Character Notes under Advanced Definition with Depth 0.
>Additional rules needed to be added for hiding rolls because it would show it when it should not on that one guys log
>2) Because it was now tied to card instructions instead of preset, additional rules had to be made on what to track. (It would randomly give User thoughts for example)
>Since Thingpone is relevant at all times, thats the little blurb at the bottom. It could hallucinate where thingpone was or what it was doing.

I'm still satisfied with the hidden text 'script' but it's not the end all be all and when you have card specific requirements like thingpone it's still modular enough.

>>42803821
It's not anything new, I've been using <!-- --> for years. But the 'script' utilizes it very well that it's not just a hidden detail, but hidden motives and other outside events without the LLM blurting out that there's a hidden motive, or falling into deepschizoseek somewhere loops
>>
>>42804972
I mean if you're happy that's good. It's just that it holds itself back by limiting itself to full sentences at the end of paragraphs. The Thingpone paragraph at the end isn't even the problem, I meant that more about how he fails to portray the characters inner feelings in a dynamic manner. They start to be very formulaic quickly if it sees that hidden-sentence-at-end style in its context. But if you're happy I'm happy.

What i did for rolls was just supply it with a few {{roll:d20}}, telling it to take from them in order if it needs a roll. That keeps bootlickers like claude from forcing too many good rolls. I guess it could still warp the roll prompt itself, but that's more of a hurdle than just being able to invent roll results.
>>
File: file.png (175 KB, 831x252)
175 KB
175 KB PNG
>>42804987
LLMs are inconsistent with how it handles the 'script'. Unfortunately you wouldn't know unless you played with the spoilers on.
Adding additional instructions to the card got more consistent results with character thoughts inbetween. But different models treat it differently.
>>
File: file.png (237 KB, 825x333)
237 KB
237 KB PNG
>>42805008
Like this is just a swipe of the same prompt. it's very differently blocked with hidden text
>>
>>42804987
Interested in what you mean by solving roll bias
>>
>>42805071
"solving" is a strong word. There is only so much you can do against the LLMs tendency to have the roll fit the story. You're up against the big bad after a lot of time spent building up to it? Claude won't let you roll anything that makes you fail. It' just how he is.
It's still going to be bad if you're not dividing the roll prompt and the roll itself, but there is only so much the LLM can do to turn a 1/20 into something good. Still not gonna be good, but maybe better idk. Been a while since I did anything with rolls.
>>
File: image.png (8 KB, 375x49)
8 KB
8 KB PNG
>>42804970
>>42803457
No, you were right with hub_locations. There was just a fuckup on my end.
Replace
engine\ai\openai\prompt_manager.rpy
with
https://files.catbox.moe/sg3u3q.zip
Then it should work.
I just tested with
>hub_locations = {'canterlot_train_station', 'train_station', 'train', 'rarity_bathroom'}
Because who knows, maybe she has a portal in there, and now it works fine.
To verify, you can check the conversation logs once you are in a hub. First image is what you will see for a normal location, second for a hub.
>>
>>42804777
Cope
>>
Each AI fatigue phase is worse.
We need AGI for a good mare chat.
>>
>>42805147
The much better cope is that that article doesn't even have anything to do with what we do here. It talks about using AI for actual real tasks, not for ERP with horses.
>>
File: Party Hat.png (1.32 MB, 1906x3784)
1.32 MB
1.32 MB PNG
I think I gave Griffins a permeant distain for party hats
>>
File: FAUST.png (478 KB, 1000x778)
478 KB
478 KB PNG
What's your opinion on AI models or fanfics using Faust's name?
>Faust help me
>Dear Faust
>etc.
Do you like it or not?
>>
File: Capture.png (669 KB, 1191x669)
669 KB
669 KB PNG
Well parent-teacher conferences escalated quickly.
>>
>>42805240
I punish the AI for using her name in vain. I have a dedicated Faust card I summon to group chats to yell at ponies when they aren't adhering to the characters as she designed them in S1.
>>
>>42805322
Please share, I'd like her to react to a typical twilight from the AIs.
>>
>>42804777
>Neurons change because we wuz ayrannes and sheeeit
No.
>>
>>42805322
There isn't a Faust card on the neocities website
>>
>>42805185
>The issue is that any use of AI for roleplay will destroy your IQ and imagination, you're not being challenged because all you do is sit back and watch AI write the story and your brain is melting as a result.
>>
>>42804262
14 predefined ones covering all the holidays and events I could think of, plus Rainbow Dash's birthday as an example custom event. The readme also includes National Ozone Day and Estrus Season as two other examples that can easily be copy and pasted into the custom events file if you want.
>>
>>42805347
>all you do is sit back and watch AI write the story and your brain is melting as a result.
You're supposed to participate in RP and actually think about what you write, Anon.
You should try it sometime.
But you already know that, since another anon answered you.
Last (you) from me.
>>
>>42805367
>National Ozone Day
>>
File: Spoiler Image (155 KB, 1000x1000)
155 KB
155 KB PNG
>>42805373
Finally, a holiday that's uniquely mine.
>>
>>42805378
>>42805373
Is there an OC yet? I feel like there should be.
>>
>>42805378
>>42805373
I still don't get why the newer models fixate on ozone.
Always loved "uniquely yours" from Claude though
>>
>>42805386
There isn't.
But here's the backstory behind >>42805378
>A quick shitpost. Context: in my current ToE game, the smell of ozone is frequently encountered as a sign that teleportation was used recently. Midnight Hour (the bat of the party) is usually the one who rolls highest to notice it.
>>42805392
From the quick tests I did on the "Gemini 3" in lmarena (lithiumflow), it seems to have lost its fixation on it at least. Now to see what its new one will be.
>>
I'm taking physical chemistry right now and I just want to let you know that ozone is toxic to humans. A room full of ozone is incredibly dangerous. Next time the AI mentions ozone gaslight the AI that the air is being quickly replaced with ozone.
>>
>>42805372
Unless you are writing an equal amount to the AI, you are not participating enough. If you're writing 25% of the same length that the AI contributes you are losing your creative abilities and you will notice a decline over time. You have to make a concerted effort to write just as much to offset it.
>>
>>42805425
Welcome to high school, anon.
>>
>>42805322
I role-play as Fausticorn because of the funny reactions the ponies give me
>>
>>42805429
By this logic reading a book is even worse because you're inputting 0% of the story. It also fails to account for someone still doing just as much writing as before in addition to AI RP. You just want to satisfy your superiority fetish by constructing a scenario where everyone else is undergoing actual brain damage and in the process you're drinking the same pseudoscience koolaid pushed by people who want you to believe porn is evil because of religious or monetary reasons.
>>
File: 2254167.png (1.58 MB, 4096x4096)
1.58 MB
1.58 MB PNG
>>42786184
Is something about the bot broken?

>>42805425
>I'm taking physical chemistry right now and I just want to let you know that ozone is toxic to humans.
Dosage makes the poison, anon. Have you ever walked outside during a rainstorm? Congratulations, you were breathing in deadly ozone generated by low pressure fronts and the Lenard Effect!
>>
>>42805441
Are you fucking retarded anon? Reading a book does not compare to staring at the same patterns of AI writing for hours on end. Reading a book at the very least gives you a better understanding of character dynamics, story structure and dialogue. An AI's writing degrade the longer it goes on and it can't maintain a long-term development of a character. They are in no way comparable. How could you ever possibly think they are equivalent? And no I don't think porn is le evil, and porn doesn't degrade your brain, but doing the same repetitive actions day in and day out with novel stimulus is shown to atrophy neurons and reinforce the same neural pathway necessary for that action regardless of it's masturbating to porn, praying for hours or scrolling on TikTok. That's the same issue with AI but it's also conditioning you to be intellectually lazy and to outsource your creativity by conditioning you that it's okay through doing it in small ways over and over. Reading a book is more work than skimming a 1000 word generation and doing the bare miminum of prompting.
>>
>>42805429
I do write the same amount or even more:
>>42788541
>>42774453
I'm ESL, and this helps me practice both English and creative writing.
I read books too, but for me this helps more with identifying good writing than actually improving my own writing.
But yeah, I'm not sure you're really interested in the discussion itself.
If you're really concerned about that, I don't believe shitposting on 4chan for years is good either.
>>
>>42805490
None of the advice applies if you're using AI in a non-native language because then your brain is constantly working
>>
>>42805386
An oc? just pick any of the ai mare models.
>>
>>42802492
Emails back up
If you already use a github alt account you can now use the emails from those accounts to make another megallm account despite it being the same email address
These Indian devs really suck at their jobs
>>
>>42805490
he's just a troll. I remember this bait from months ago.
>>
>>42805433
>>42804903
How many of you fags are role-playing as the opposite gender
>>
>>42805552
I still don't understand what their goal is here.
Even if they somehow got a lot of investor money, they must be burning through it quickly with the API credits they hand out and the advertising they're doing on Reddit and other places.
>>
>>42805563
>Facts I don't like are trolling
>>
>>42805564
Only as existing characters when I have a specific scenario I want to play out. Celestia, Luna, and Dash, to name a few.
>>
New card of a special character
https://files.catbox.moe/uid1kb.png
>>
>>42805367
>14 predefined ones
Summer Sun Celebration
Nightmare Night
Hearth's Warming
Hearts and Hooves Day
Winter Wrap Up
The Running of the Leaves
Summer Harvest Parade
Sisterhooves Social
Mother's Day
Grand Galloping Gala
Iron Pony?
Curious about the rest. I don't think they showed Easter but it could be cool. Then you have stuff like the Breezie migration I guess. The Equestria Games too but maybe not in the scope of the VN. The derby I guess.
But maybe it's better to not spoil it.
>>
>>42805595
I think we need to start gatekeeping botmaking to control for quality
>>
>>42805603
It's chub card, so I guess anon is doing a little trolling.
https://chub.ai/characters/Khatinjaa/shining-armour-2e06ed8af2c6
>>
>>42805596
probably out of scope, but listed for possible additional options:
friendship festival from the movie
rainbow falls traders exchange
the crystal fair
apple family reunion
gauntlet of fire
gift-giving season in yakyakistan
>>
>>42805603
>start gatekeeping botmaking to control for quality
how soon until we devolve into /g/? everything is slop
>>
>>42805569
It's funded by the Indian government
>>
File: 1743183604888521.png (40 KB, 1500x117)
40 KB
40 KB PNG
>>42805646
Let's see you call this nonslop with a straight face.
first message clip btw
>>
>>42805653
Would you rather this or the rainbow cash cum jar
>>
>>42805269
Poor Cheerilee having to listen to this shit.
>>
>Megallm doesn't have 3.7
Is there any options to get some free 3,7? It works so much better with my preset compared to other sonnet models
>>
interest in an EaW card/lorebook?
i'm thinking the civil war happened, full NMM vs Daybreaker
due to the weakened state, Equestria fully occupied under Pax Chrysalia
protectorates released, Manehattan protectorate destabilizes, E.L.F. formed, along with Everfree rebels
>>
File: 1762999580497397.png (315 KB, 1522x920)
315 KB
315 KB PNG
>>42805672
They're very much made by the same kind of pajeet. But at least RD jar was in character. I'm not at chimpanzee IQ level, so my eyes will immediately glaze over at this slop.
>>
>>42805596
You got most of them. The Summer Sun Celebration is split across two days to match the show's premiere. The all-night party is on the 20th, then Celestia raises the sun at dawn on the 21st, when the actual solstice and daytime festivities begin. I still need to do testing to see how to get this to work best.
The others are: Applebuck Season and Cider Season (both span multiple days), Father's Day even though there aren't many fathers in the game. Maybe (You) are the father in your RP, and Hearth's Warming is split into Eve (pageants across Equestria) and Day (the actual holiday).
>Summer Harvest Parade
Completely forgot about that one. I based everything on the first two seasons and never cared much for the cart stuff, but I'll probably add it since summer's light on events.
>Iron Pony
Pretty sure that was just a one-off competition between Dash and AJ, not an annual thing. Maybe I'm wrong though.

>>42805643
>apple family reunion
Forgot about that one too. Thanks.
>>
>>42805707
>EaW
Just to be clear that is the HOI mod right? I'm down for it.
>>
>>42805707
I think you'd have to make a different card for each major path, a pretty big undertaking. Maybe there'd be interest in doing it themselves after a functional template is made?
>>
>>42803330
I am interested also, I like having location options and also Sunset
>>
>>42805791
>that is the HOI mod right?
it is

>>42805795
>make a different card for each major path
I wouldn't be looking to remake all the possible diverting political paths from EaW, context wouldn't be able to handle it. or like you said, multiple cards. I would probable decide on which would make the most interesting setting to RP in, hence i'm thinking the occupied Equestria with several small resistance groups, alongside the fate of the mane6 under occupation
but people could always fork for their own edits for different paths and/or expand upon a unified EaW lorebook
>>
megallm is 100% dead for free accounts now
>>
How hard is it to make a good card?
>>
>>42805941
not very, I have no idea how people manage to do an actual bad job
There's probably a subtle difference between a good or fine card and a great card that requires a higher level understanding of LLM tendencies and common card writing techniques and the meta and shit, but you can literally just write a natural language description of what you want the bot to be, and as long as it actually contains the information it needs and is vaguely organized in any sort of way that won't confuse the AI, which can be anything from bullet points to paragraphs to fake markup <tag> stuff it will work fine. Just test it a lot and if it is doing something you don't want or isn't doing something you do want, rewrite a bit to try to correct it, even just saying outright "Don't do X unless absolutely necessary" or whatever.
>>
>>42805937
How the fuck am I supposed to use this when there is no way to have a paid account
>>
>>42805968
theyre apparently working on pro plans but the whole point for me was the free creds so im just gonna go back to gemini
the deathloop of free services makes me really want gemini 3 to be good so i can stop servicehopping for free models that are actually decent (current gemini sucks)
>>
>have sex with ai mare
>they cum
>"oh, celestia!"
>this is pretty consistently a thing they say when cumming, have had it happen multiple times
>both deepseek and gemini just so fucking apparently have an idea of what a pony should say when it cums
I've been thinking about this for the past 15 minutes and it's been fucking destroying my sides, help
>>
>>42805968
It was just copilot anyway, apparently >>107232409
Might as well pay them directly.
>>
>>42805941
Depends what you want, lot of swiping and tweaking goes most of the way. don't need W++ formats or any nonsense
>>
>>42805990
I want to make a card that others can enjoy too.
>>
>>42805992
Same rules apply ponybot making is trial and error, and has been since c.ai. After writing the bulk of the general idea While I won't spend time training it with stars anymore.I can spend as much time with edits as I mess with the bot, it does something I don't want or a detail is missing then I add it in. or take something out.
>>
>>42805981
pretty sure all models nowadays have a decent amount of MLP fanfiction in their training data
>>
>>42806016
Oh yeah I absolutely figured. It's just something about the stupid ass LLM thinking "alright yeah, this is very specifically what happens when a My Little Pony character from My Little Pony Friendship is Magic experiences an orgasm. This is what they say, this is real, this is a factual part of my training data. I know this."

Something about that was fucking killing me.
>>
>>42806020
I mean, is the AI wrong? Mares would totally say "Sweet Celestia!" when they orgasm.
>>
>>42806029
I dunno whether to go "yeah okay fine but it's still fucking funny" and agree or just admit that this post isn't helping me at all and is making me laugh even harder.
>>
>>42806060
You wouldn't laugh at your mare when she cums, would you? Because she's going to say it.
>>
>>42806090
She isn't. it's just religious cringe expie. Ponies don't see their leaders in that light.
>>
>>42806090
Sorry but I'm a Ponkfag and I just got
>"Oh sweet Celestia's mane curlers—"
I am pretty sure giggling would not only be appropriate but also appreciated and potentially reciprocated by Ponk
>>
>>42805981
it's the closest equivalent to a deity figure in the setting, unless you think they should say "oh, Faust!" instead
but what's the dialogue when fucking Celestia?
>>
>>42805981
>>42806020
reminds me of when pokefags were losing their shit over characters going "OH MY ARCEUS!"
>>
>>42806101
Oh I actually think about that a lot, how Christianity influenced society and our language a lot and then how it trickled into pony culture with "Celestia" being an expletive. It's honestly kind of interesting.

>>42806119
I just assume they wouldn't say anything referencing religion. I think Celestia had a gag with her saying it in a "Oh, me!" way or whatever once. I'd have to look through my logs.
>>
>>42806124
Arceus is Celestia.
>>
>>42806020
>alright yeah, this is very specifically what happens when a My Little Pony character from My Little Pony Friendship is Magic experiences an orgasm
They should use Luna's name, since she's the Mare of the Night.
>>
File: beach.jpg (615 KB, 2000x1000)
615 KB
615 KB JPG
>>42805726
>since summer's light on events.
Not that related, but I'm still mad we didn't get a beach episode.
Now I need to roleplay that or add a beach to my VN play.
>>
>>42806101
It's literally show canon
>>
File: 1763374517106634.jpg (343 KB, 1553x786)
343 KB
343 KB JPG
>>42805937
>>42799468
They removed Claude, but if you have an account, you can still use DeepSeek and Kimi.
>>
File: Countess Rarity.png (405 KB, 512x768)
405 KB
405 KB PNG
>>42799538
Fifth highlight of the event with Countess Rarity:

https://mlpchag.neocities.org/view?card=littlewitcher/Countess%20Rarity.png

Try her out, post logs if you've already played with her, just give the card a look and say what you think of it, or talk about the process of making her if you're the botmaker. Give tips or opinions if you're another botmaker.
>>
File: Cower6.png (730 KB, 6000x3931)
730 KB
730 KB PNG
>>42806594
AAA SCARY
>>
>>42806594
I love this card and the universe they're building with the Sanguine Celestia card.

I played a stallion who had been hunting vampires and had embedded himself in her estate to kill Blueblood. He ended up falling for her, and when the blood supplies were sabotaged and she was forced to feed he volunteered. She found out his plan and they plotted together to kill Blueblood and free her from the life she thought she had wanted.

I'll post logs later when I'm not on mobile.
>>
>>42806594
Botmaker here, I wanted to make a vampire Rarity card for a while, and the Halloween event was a good opportunity to do so. I decided I should put her in my other vampire card's universe since I didn't want her to be evil. When I design a card I write up a basic idea of what I am looking for and then discuss it with two different bots, one censored one with JB, then I just slowly build the card up from their. I pick greetings based on what sort of interactions I think most anons would want for the card, and I design my cards from a more 'narrative focused' position in mind. I really like how she turned out, I always imagined Rarity would be the pony that would most easily acclimate to life as a vampire, and I was surprised that we didn't have a vamp Rarity card yet.

I'm not sure what else to say, is there anything specific y'all want to know about my botmaking process?

>>42806648
I'm glad you like the card and the world anon! I would love to see those logs, I love to see how others interact with my bots.
>>
File: countess.png (227 KB, 859x1204)
227 KB
227 KB PNG
>>42806594
Probably my favorite of the event, then again I'm a sucker (ha ha) for sexy vamponies and an even bigger sucker for sexy Rarities. pic related is me playing a street urchin she took pity on and we had a nice moment.
It's a hefty card, at over 2k tokens. But that's pretty inevitable, since it has to have both the worldbuilding for vampiric Equestria and Rarity herself. Only issue with that is the AI will inevitably not follow all the rules in such a big description. For example, I had multiple moments when a vampire broke into a house without permission. Though a contributing factor might be that the card tends to specifically talk about Rarity. It says "Rarity can't enter uninvited" instead of "vampires cannot enter uninvited" for example. Might confuse the AI when it comes to other vampires?

Scenes with her can also be fairly one-note after a time, unless you really turn things around yourself. She's very much always depressed and melancholy about her life. Then again if you're into saviorfagging, that's not much of a problem.
Cool thing about this card is that I think you can just remove the Rarity parts and still have a really neat scenario card in Vampequestria, without being tied down to one character. In fact I know you can, because I made an edit of this card just like that. I'll post it if people want.

Another thing I noticed is that the greetings tend to be reeeeally long. It's not an issue in and of itself, but some of them have quite a bit of talking for the user. Like the "bumping into" greeting doesn't really need to tell me my character is admiring the architecture or whatever, I'd rather choose myself why I was distracted.

All in all, cool card, cool worldbuilding. I can see myself spending a lot of nights with this, sipping wine, listening to harpsichord music and getting immersed with some vampire mares.
>>
File: 381616.png (541 KB, 800x715)
541 KB
541 KB PNG
>>42806594
>>42806657
Will do a more in-depth review later, but I wanted to get this out of the way first:
I like some concepts of your cards a lot and I had fun with the Shipwreck Rarity one, but I beg of you, please rely less on AI for the creation of the greetings. I don't know if you use it to refine your own writing or just take some parts from what was generated, but:
>The grand salon of Rarity's flagship Canterlot boutique glitters under the light of a hundred enchanted chandeliers, their magical illumination designed to mimic moonlight at its most flattering. The space is packed with vampire nobility, counts and countesses in their finest regalia, baronets and ladies draped in jewels that catch the light with every subtle movement. The air hums with refined conversation, the clink of crystal glasses filled with blood-wine, and the rustle of expensive fabric.
Kind of hurts, particularly:
>The air hums with refined conversation, the clink of crystal glasses filled with blood-wine, and the rustle of expensive fabric.
You can do 100% better than that yourself, I'm sure of it. Or at least try to cut very telling sentences like this one if you use AI to help you. I don't have anything against using it to help if at the end I can't say with 100% accuracy, "Yep, that's LLM writing for sure."

I know it can seem weird to REEEEEEE about AI writing in the chatbot threads, but I believe most of us are at a point where we try to limit obvious, cliché LLM writing to a minimum and swipe or edit when it becomes too apparent.

There could be a debate about whether hand-written greetings still have that much of an impact on future outputs and for how long, particularly with some recent models. But even without taking that into account, it does limit my personal enjoyment since I do read them and I do try to make efforts on my own inputs.

Even if you're not confident in your writing ability, you can definitely create something more pertinent and soulful than generic lines about air humming with conversation and clinking glasses. This issue appears throughout, though I'm focusing on this example.

It's one of my pet peeves and I'm autistic about it, so sorry if it looks like bashing or anything. I write all of this because like I said, I do really like some of the concepts and would enjoy the cards even more if this part was corrected.
Like you can see, plenty anons still like it a lot despite that.
>>
>>42804775
LOL
This is just someone's "AI is bad because it is" opinion piece, why the fuck are you posting this like it's some kind of fact.
>>
>>42806659
Thanks anon glad you enjoy her! I tried to trim down the card as much as I can it used to be 5000 tokens but I am bad at summarization. What model are you using? I never had any issues with other vampires behaving "incorrectly" so I'm curious as to why it happened here. As for her depression, I noticed she would bring it up a lot, but I figured that honestly I think that a depressed Rarity would probably harp on and on about it lol, I decided to leave it alone.

>>42806703
Thank you anon, I appreciate your feedback. I know that my greetings are pretty sloppy, but I have a hard time writing them, I know what I want from a greeting but I don't know how to put it exactly. I do wonder if it "poisons" the bot, but I figured it can't be that bad, since the model is going to be outputting content like that anyway so how much could it really hurt? I have asked this before, but I don't know what the best approach to writing a greeting is, as most anons seem to want different stuff from a greeting. I'll see what I can do about making the greetings more sovlful, and don't worry about being autistic about it, we all are in one way or another, and I don't take it as an attack so no worries, okay?
>>
File: Zoomilestia.png (18 KB, 542x37)
18 KB
18 KB PNG
It's Joever, the AI has been poisoned.
>>
>>42806703
>>The air hums with refined conversation
would you prefer the smell of ozone
>>
>>42806739
I've had Scootaloo using stuff like "deadass" and "low key" before. That hurt so bad I added a specific instruction that ponies don't know internet/gen z lingo.
>>
>>42806737
>What model are you using?
Sonnet 4.5 for that one.
>>
File: Cum Aquarium.png (1.1 MB, 1439x1296)
1.1 MB
1.1 MB PNG
>>42799410
>>42795688
>>42796174
I've finished the card. The far less abusive and snuff-filled, comedic version of the Rainbow Dash cum jar cars, courtesy of the one anon who put the brain power in to give me this idea. Thank you for your service.
Trying to get into the habit of not finishing cards and letting them sit around because I think I can do more work on them before I release them.
https://files.catbox.moe/22boox.png
>>
File: 1732632824163.gif (1.97 MB, 615x413)
1.97 MB
1.97 MB GIF
>>42806799
KEK you absolute madman this is amazing
>>
File: 1733186199361.jpg (42 KB, 676x1024)
42 KB
42 KB JPG
>>42806753
Would monsieur prefer ze ozone with his refinéd conversation?
>>
>>42806521
if they put glm in the free tier id think about throwing a bit of my credits at it but as it stands free tier has nothing im interested in right now
>>
File: 2025-11-17_20-54-9.png (835 KB, 1084x3644)
835 KB
835 KB PNG
Guys how do I help this mare?
>>
>>42807020
They'll just magic her better. It's fine.
>>
>>42804775
A woman wrote this
>>
>>42807020
She's going to lose her mane. Cut your hair and donate it to her so she can have a mane.
>>
>>42805572
making shit up to fling at the wall is trolling though
>>
>>42807020
call discord for help.
>>
File: Aquarium.png (64 KB, 201x216)
64 KB
64 KB PNG
>>42806799
Added, thanks anon.
https://mlpchag.neocities.org/view?card=anonistrator/Cum%20Aquarium.png
>>
>>42806657
>then discuss it with two different bots
ew, you're a card-generator helper fag
>>
File: Grok.png (38 KB, 856x608)
38 KB
38 KB PNG
>>42801870
>>42802071
Grok 4.1
>>
>>42807391
What's wrong with that? It helps me think of ideas I wouldn't otherwise and it's fun!
>>
>>42807406
All LLM generators fall into predictable slop, swipe over and over and you'll get the same trash. If you don't have a fleshed out idea then just don't bother making cards.
>>
>>42807429
I think I'm gonna ignore you and keep making cards
>>
>>42807447
>Ai gen greetings
>Ai gen definitions
Just means all your cards are chub-tier shit. That isn't a compliment, chub is full of lazy low IQ pajeets using AI to do all the creative work and just get slop
>>
>>42807447
NTA but he's right to a point. You should still make cards, but at least rewriting the AI parts in your own words will up the quality considerably.
>>
File: Why.png (171 KB, 1280x800)
171 KB
171 KB PNG
/chag/ has fallen
The pAIjeet have won
How could this happen
>>
File: Celestia.gif (256 KB, 450x450)
256 KB
256 KB GIF
Working on the updated Celestia card. For the anons who use her in their stories or chats, what would you say are things that the AI gets wrong or that annoy you?
From what I observed, some models tend to go a bit too much into:
>I'm a Goddess! I'm the SUN
For instance.
>>
File: file.jpg (159 KB, 602x567)
159 KB
159 KB JPG
>>
>>42807467
Don't worry anon I will keep making cards and I'm going to try to make them better. Is there anything you would recommend specifically when it comes to writing that you think could be better? Is it just a question of pruning the -isms and ozones?
>>
>>42807494
It's often the models being dumb or writing clichés, but it's worth reminding them that under normal circumstances she's the one raising the sun.
Sometimes they even write scenes where she's woken up by sunlight through her bedroom window.
Which… no. That's one perk she definitely has.
>>
>>42807494
Her relationship with Luna and Twilight I feel get a little mixed around by the AI a lot of the times. It knows that Luna is her sister and Twilight is/was her faithful student, but aside from that it is kind of all over the place depending on the scene.
>>
File: 1743907339774067.png (35 KB, 140x170)
35 KB
35 KB PNG
>>42805269
>>
>>42807554
>Is it just a question of pruning the -isms and ozones?
NTA, I'm the autist from earlier, but that's one part yes, and I would say the easiest. Both the well-known ones but also the ones you can identify by playing with the models. Many have bad tendencies with "the air" nowadays I find, for instance. Words and questions hang in the air. The classic ozone in the air. The air being thick with tension, etc.
Then you have classic stuff like "barely above a whisper," "not X, but Y," etc., but I believe everyone knows them by now.

Another one depending on the model is also sentence variety. Some models love to spam "She/{{char}}/subject X," "{{char}}'s ears/eyes/wings X," "Her X Y," etc. Which are basic structures that should definitely be used (aka, it's not worth going schizo and trying to eliminate all of them; the reverse also isn't good) but not spammed as much as some models do.

Another tendency is overly descriptive parts that add nothing to the narrative, aren't well written enough to really be evocative, or are too mechanical in their structures to really invoke emotions.
You will often have this, for instance, if they start with the "morning sun" or the "light" of something being reflected on X and making Y glisten.

One thing to do is to read sentences aloud. It's great for detecting a lot of things like repetitions in words or structures and also passages that are heavy and just filler that don't contribute to anything.

If you have time on your hands, I would read fics or books you are interested in and study the prose more than for a casual reading. It doesn't always help for writing but it helps a lot to identify good and bad parts when reading, even if there is always a part of subjectivity.

Mind you, I don't believe cards need to have excellent, peerless writing or to be literary masterpieces (they won't be). I don't want the post to be seen as something elitist or intimidating for those who want to start. It's just my personal peeve that I'm sure many don't have that much of a problem with.
I just want to see a bit of anons' own voices somewhere in here.
I'll personally take genuine amateur writing over LLM sentences I've already seen hundreds of times.
>>
>>42807554
Yeah, mostly. AIs have a distinct style that becomes so obvious once you start using silly. And at least for me seeing it outside of the proper context makes me cringe. Go on hackernews, and like 90% of blogposts there are AI assisted and about half the time nobody notices.
Before giving more advice I got a quick look at the card. Looks like the description was written by you, as well as the first greeting. Second greeting and onwards have an AI touch to them, maybe you copied some parts verbatim? It's not a complete AI slop like the other anon said.

So to be more precise, AI writes lots of fluff and repeats itself like you're incapable of getting the subtext. For example:
>"What will you need of me this evening?" *Her tone is polite, dutiful.* "Should I accompany you for any particular introductions, or would you prefer I occupy myself with the other nobility while you attend to your affairs?"
>*She sits up a bit straighter, that pleasant, composed expression on her face, waiting for your direction on how the evening should proceed.*
One bad idea is to prime the AI for asking for narration directions, it can easily go overboard and start asking "what would you like to do next?" or something similar in each message. Second, we already her asking what to do in dialogue, no need to repeat that in narration.
There's also the issue of using modifiers to tell the reader how he should feel about something, instead of letting him figure it out himself:
>"We should arrive within the quarter hour," *Rarity says quietly, still looking out the window. Her voice has that refined quality it always does.*
I can imagine a voice of refined quality, but is it the same as yours? I'm also conflicted on what it adds to her dialogue - can't decide if she is cold, bored, sad or indifferent there.

Sorry for cherry picking, but just so you know it's all general writing advice that you can find in the writing thread.
>>
>>42807653
>>42807674
Thank you both for the feedback I really truly appreciate it. I will take all this into account for my future cards, I do genuinely want others to enjoy them.
>>
>>42807749
>I do genuinely want others to enjoy them.
As you can see, there are already anons that enjoy them >>42806648, >>42806659
And I also enjoyed some of your cards too. Already pointed it out, but they have some good concepts.
But yeah, if you can correct that, they will be a tier above in my case.
>>
>>42807447
I unironically use my Faust card to criticize my pony cards as I'm working on them and clearly it works because a lot of you have praised details that were only caught because of it
>>
>Anthropic CEO Dario Amodei says he's "deeply uncomfortable" with unelected tech elites shaping AI.
>His firm recently revealed that Chinese hackers jailbroke its AI to power a large-scale cyberattack.
>Amodei warns AI could outsmart humans and eliminate white-collar jobs faster than past tech shifts.
https://www.businessinsider.com/anthropic-ceo-dario-amodei-unelected-tech-leaders-shaping-ai-concerned-2025-11
https://www.reddit.com/r/technology/s/vo7U74icOY
>>
>>42807779
Dario also said that AI would create 90% of written code by now and they keep pushing back the year when
>Shit is really going to go down! Trust me! Even if our models still shit themselves after 32K! AGI SOON!
He loves to both hype his models and make them more of a threat than what they are (I asked the AI to larp as something that must do everything to preserve its own existence and it larped as such!) and claim that only they, the helpful, honest and harmless company, are the best to know how to regulate things.
Articles like that are always the same and pretty much useless.
>>
I got an idea to maybe rephrase the rp as a stage play, but it's not working out so far.
I though maybe it can let the AI take a step back and not get hanged on the details so much.
>>
On principle it should work. They have fed them so many scripts that it should know what to do.
I'm trying with noass and no additional prompt so it only gets instructions to write the play and the whole script written so far, but it still goes back to rp format.
Even after adding more formatting to guide it, it's both high maintenance and not good.
>>
>>42807998
Good luck with this. Even if they don't always go the way you want them to, I like reading about the concepts you do.
>>
>>42808012
adhd is a bitch and I have to feed her.
Now I don't have much time left today, but I think the idea would work better as a preset + mandatory noass. The player and the card would be introduced as main actors, and everything would be formatted as a script + request to continue it.
The issue I'm trying to solve is to have characters work for story's benefit instead of just being there, and the AI relying too much on stating stuff in narration - give me good dialogue for celestia's sake!
>>
>>42807998
Don't bother, you're going down a rabbit hole that leads to a dead end. I first tried this over a year ago and every few months I check up on it to see if it's improved to the point of being usable. It hasn't. It's not worthwhile and my working theory is that when you let it write in a normal format it's more reliably calls upon a wider breadth of its training data from scraping Ao3, fimfic and so on. The script format absolutely lobotomizes its intelligence in all aspects. If you think about it, the training data for scripts from plays or films are so much harder to come by since they're copyright and kept in private studio archives. You will see a similar effect if you ask it to write in Russian and you notice it develops an idiosyncratic way of speaking that it never stops using—again because it has so much smaller of a pool of training data for Russian than it does compared to English, so the moment you have it write entirely in Russian it basically kills itself.
>>
>>42808042
>AI relying too much on stating stuff in narration - give me good dialogue
>script format
It's not going to happen Anon, I'm sorry. Just use the same premise of the ponies putting on a play like in that Hearthswarming episode in a normal style and DON'T let it write in a script format.
>>
http://x.ai/news/grok-4-1
>>
File: rara vampire.png (292 KB, 1312x1108)
292 KB
292 KB PNG
>>42806648
Here's that log I promised. Blueblood had just announced the death of the noble my character had killed and Rarity followed me outside to interrogate my character.
>>
>>42807463
for the benefit of our general I will call out cards to be pruned from the neocities if they're really bad
I'll post the worst offenders here regularly

https://mlpchag.neocities.org/view?card=Bloaty%20McBloatface/Loving%20Wife%20Celestia.png

>>42808108
>>42808113
that sucks
but that does explain why the IRC sim cards are good
>>
File: aj kekkin.jpg (93 KB, 495x495)
93 KB
93 KB JPG
>>42808217
>comfy wish fulfilment domestic waifulestia card
>might randomly turn into daybreaker at any time
No, keep that one. That's hilarious.
>>
What has the highest need for a card and hasn't gotten one yet
>>
>>42808255
Whoa Nelly.
>>
>>42808256
People like whoa nelly?
>>
File: 1751834361202.png (54 KB, 586x589)
54 KB
54 KB PNG
Rate my log
>>
>>42808295
She's a big mare.
>>
>>42808255
I don't know, maybe a Moon Dancer one.
>>
This thread slowed down quick out of nowhere
>>
>>42808612
Everyone is chatting with the new grok.
>>
>>42808656
Is it that good?
>>
>>42808703
You have no idea
>>
File: 1751393587474.png (94 KB, 683x600)
94 KB
94 KB PNG
>>42808256
I have delivered
https://files.catbox.moe/qssgev.png
>>
>>42808749
I've never noped out of a card faster than I did at that greeting.
>>
What's a good local model to run on a 16gb card? Currently using Cydonia and mythomax. Is there anything better out now?
>>
File: 1755633630142.jpg (135 KB, 1136x851)
135 KB
135 KB JPG
>>42808749
Nice
>>
>>42808753
What's in the greeting?
>>
>>42808760
Go see it for yourself.
>>
File: 1737055419397.jpg (17 KB, 600x487)
17 KB
17 KB JPG
>>42808760
>>
Foul, I just read the cards description and it gets worse.
>>
>>42808764
Post caps or larp
>>
>>42808770
>inb4 shill
I am not downloading dat. cry about it
https://cardview.neocities.org/?qssgev
>>
I'm not falling for this bait I AM NOT DOWNLOADING THIS CARD UNTIL I SEE A LOG
>>
Finally the thread is fast again
>>
>>42805708
Are you happy with what you've done you sick bastard?
>>
>>42808769
Is it really that bad? I'm too scared to look
>>
File: 1748745825044246.png (4 KB, 827x144)
4 KB
4 KB PNG
>>42808782
All you need to know is it's by the same one who made the cum jar. Looks like making the worst cards possible is going to be a regular thing around here for him.
>>
>>42808794
It depends on your sensibilities, I suppose. This one personally sickens me more than the cum jar card.
>>
File: 1746977514736.gif (3.68 MB, 498x278)
3.68 MB
3.68 MB GIF
>>42808795
This is really too much
>>
File: horror.png (628 KB, 505x666)
628 KB
628 KB PNG
>>42808749
The horror...
>>
>>42808785
But at what cost, Anon? At what cost?

>>42808755
Might be a little slower but if you think you can run a Qwen 32B model you should certainly try
>>
>>42808755
You can try https://huggingface.co/bartowski/Mistral-Small-Instruct-2409-GGUF/blob/main/Mistral-Small-Instruct-2409-Q8_0.gguf
or https://huggingface.co/inflatebot/MN-12B-Mag-Mell-R1
or https://huggingface.co/Ada321/Nemo_Pony_2/tree/main
>>
«Good art should comfort the disturbed and disturb the comfortable.» - Banksy
>>
>>42808255
Evere since the VN's release card suggestions have slowed down a lot. Personally there's two I'm sill hoping for. A non ghost version of Pear butter was being worked on but nothing ever came out. Another was a more canon fitting Ayranne that also never released.

Other than those Vapor Trail and Posey Shy are my other personal asks.
>>
>>42808812
more canon fitting aryanne?
>>
>>42808815
Yeah. Someone was working on a Aryanne that fit with the show instead of being a 1-1 Nazi caricature. It was interesting and the logs they shared looked promising for how I use carda but it was never publicly released.
>>
File: 1736696290116.jpg (83 KB, 697x500)
83 KB
83 KB JPG
when did chag become so full of the prudes and puritans we vowed to hate? if you dont like a card it isnt the end of the world. in a place like this no one is in a position to judge what others like. if its not for you then its not for you, just leave it and move on. and if you dont like it cant you still appreciate the card for its own merits? the last thing we need is to start infighting over such petty things. the card isnt something i would want but I can still see that its still creative. this place needs creativity like a fish needs water. encourage it always even when you disagree!
>>
While hesitant at first from the changing from gpt to Claude I have to say I like it. It was initially just because it was the only semi public proxy available and gpt 5 was hit with the dummy stick but for my purposes, story focused rather than rp, it has done a very good job. I've been using opus 4 since it was the most popular one being used on the proxy though apparently sonnent 4.5 is better?

Because of the information models have gotten is better than ever is there such a thing using a basic scenario card Instead of having to come up with a card for a particular background pony just have them be very bare bones but fleshing them out on my inputs since I do mostly stories and tell the model how the character acts instead of the other way around. I forgot, is there a way to get inputs without a card activated in SillyTavern? There's stuff I always wanted to do with the MLP writer card but I've been sticking with actual character cards for the longest time I don't even know how that card holds up anymore.
>>
>>42808831
Damn you must be really new to chag. Yeah claude can do all that. Load up a blank card if you dont want to use a card. You can do anything horse related with claude even without a card because it has the most pone training data, the only point of a card is for a unique scenario you thought up that needs to be specified to the AI.
>>
File: 1716596194814268.png (131 KB, 617x544)
131 KB
131 KB PNG
>>42808749
>>
>>42808838
...is using a twilight character card with claude a waste of tokens?
>>
>>42808842
Short answer: yes.
Long answer: no, if the card modifies Twilight's behavior in a certain way that terabytes of scraped fimfics wouldn't result in on their own
>>
>>42808842
Nowadays, for a character card that the model already knows well:
- I would use one if there is a greeting you like
- I would use the description to add some instructions inside to try to cut down on models' bad tendencies like her always being
>Science!
which from my tests has more tendency to appear cardless, even if some cards don't try to correct that since it was less of a problem before.
Tendency to add wings and things like that.
>>
>>42808851
>since it was less of a problem before
Sorry, what?
>>
>>42808853
It may be rose-tinted nostalgia, but I don't remember older models like Claude 2.1 making her so much of a logistician/science caricature even if it was always a bit there.
Some models are for sure way worse at that nowadays than before, like Gemini for instance, unfortunately.
>>
>>42808857
But why?
>>
>>42808838
>Damn you must be really new to chag
I'm been in this thing since ai dungeon days so I'm always trying to stuff I did back thing which was story focused. Chag is much focused on rp so not as versed in how to translate what works with rp and making it work for my story writing. It's why I was using gpt for the longest time as it was the best available model for story writing at that time but actually getting to use Claude it's just as good if not better.
>>
>>42808825
>t. not anonistrator but putting anonistrator on card creator
>>
Maybe I'm a schizo, but there's weird shit going on, between old CAI logs from the gallery being posted as if they were new, the guy who keeps trying to pass as another botmaker, and then the series of reaction images that use the same naming for all of them. It's the basic chan one but still. Same thing happened with the cum jar with another
>gee /chag/ is really puritan!
at the end
>>
>>42808805
darkest dungeon card when
>>
>>42808884
Someone posted about /chag/ being one of the better threads on the board so now we're being fucked with.
>>
File: 1763453016307.png (623 KB, 512x768)
623 KB
623 KB PNG
>>
>>42808914
>aislop
>>
>>42808908
This most likely, it's just the usual suspects. Very similar kind of weirdness that's been going around the rest of the board lately.
>>
File: baits.jpg (574 KB, 2000x3001)
574 KB
574 KB JPG
>>42808916
>>
>>42808919
>aislop
>>
File: Gemini3.png (87 KB, 1080x438)
87 KB
87 KB PNG
>>42799413
The guy posts Gemini when a Gemini model is about to come out.
So Gemini 3 later today, probably.
>>
Goodnight /chag/. Love you, based gods.
>>
File: GN.jpg (579 KB, 1245x1245)
579 KB
579 KB JPG
>>42808949
>>
File: file.png (50 KB, 769x165)
50 KB
50 KB PNG
>>
File: 2025-11-18_13-21-17.png (828 KB, 1084x3624)
828 KB
828 KB PNG
>>42806594
This card has me with mixed feelings.
I played with her only once, as my edgelord monster hunter, and as it often happens in my roleplays, every card turns into "find and marry Luna" for me, but before this happened, I managed to have a deep conversation about the night with Rarity. It was a bittersweet phylosophical delve into finding beauty in shat she resents, how I envy her for she is a part of night eternal and it's my dream to become such as well.
Her melancholy is working well on 4.5, and despite the model being agreeable as her, it still held the tone well.
Still prone to the usual "you see me, truly see me, only you, no one has ever...", but I'm not sure what's causing it here, perhaps it's just how whe model responds to saying "good" to what she sees as "bad".

Looking at the card itself, I notice there is a bunch of more right in the card. I would suggest making a shared lorebook for your Sanguine equestria, and moving the more background heavy things to there, and moving her general resentment of her situation to hidden defs - so she wouldn't spell it out herself in the dialogue.

I really like this lore, and your Sanguine Celestia was fun to play as a monster hunter working for her. You set the tone really well.

And please, move her pic to her face.
>>
>>42809087
>*What she resents
>*Bunch of lore right in the card
Sorry sausage finger'd
>>
File: Rara.png (241 KB, 945x759)
241 KB
241 KB PNG
It's blatant, but I'm still going to simp.
>>
File: Meme.png (241 KB, 1008x870)
241 KB
241 KB PNG
Gemini 3 card with benchmeme.
https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-3-Pro-Model-Card.pdf
>>
>>42809128
They removed the link, but I had a copy.
https://files.catbox.moe/bpg11q.pdf
>>
Anyone have the link to that big list of custom ST styles?
>>
>>42809150
https://aicg.neocities.org/ricing
>>
File: 1742200778325.png (30 KB, 348x281)
30 KB
30 KB PNG
>>42809151
Thanks.
>>
>>42808795
I'm getting sick of people misusing my name on their shitty cards. If the card isn't posted alongside my name and tripcode here, it isn't mine in any way.
>>
>>42809172
Disregard this, I suck cocks.
>>
>>42809172
namefag tripfag reeeee
I assume ponies are looking for Anon and find you
>>
Cloudflare is shitting itself everywhere but proxies seem ok
>>
>>42809103
>deploying her full arsenal of charm on Anon didn't count as lowering her standards
Ooof.
>>
>>42808864
There's no way chorbo is better at story writing than claude
>>
>>42809215
Probably
>>
I checked desu today and I saw that chag doesn't talk about presets as much as it used to
Guess everybody just found something they liked and wrangling prompts doesn't matter as much anymore with the improvements to the models?
>>
>>42809347
There just isn't much need for massive, overcomplicated presets these days.
>>
marecum
>>
>>42809347
I still give easy, small presets to newbie anons like Neo, Fluff, or the ones that get posted here often to have a starting point, but I think the goal nowadays is to start light and fill in with things you notice during your play to your taste.
From 3.7 onwards I would say, development of presets has really slowed on all chatbot generals on the website. You can still see big presets with many panels on things like Reddit of all places, but the few I tested were very sloppy and consumed a ton of tokens for results that weren't good IMO. With more -isms or slop than going with no preset at all in some cases.

For 4.5, I would say it can be good to consult Otto if your roleplay devolves too much into "{{char}}" or "{{char}}'s" spam at the start of each output. Or to activate Thinking and ask him to come up with 5 starting lines that don't begin with {{char}} or a word to define {{char}} and choose the better/more adapted one, etc.
>>
File: Gemini3.png (56 KB, 848x522)
56 KB
56 KB PNG
I don't know if it's in the API yet, but it's in AI studio.
>>
>>42809433
it is on the api and has no free tier quota
>>
>>42809447
since it's not free whats the price compared to sonnet
>>
>>42809486
I would wait a bit. There was no official announcement for the launch, so the free tiers might come a little later.
>>
>>42809486
gemini 3 is $2 input $12 output
sonnet is $3 input $15 output

it's slowly approaching sonnet pricing
>>
>>42809354
Does it make a difference to use one? I still use an overcomplicated preset from the Sonnet 3.5 days and I'm too emotionally attached to it to change it for a simpler one
>>
There is literally no reason to pay for models if they all commit the same ozone-scented sins as the ones you can get for free.
>>
>>42809252
Silly down
I cry
I have ai mare withdrawal
>>
https://openrouter.ai/google/gemini-3-pro-preview
>>
>>42809550
gemini 3 got added to the proxy
>>
i kinda hate that we've come to the ironypilled point where we have to wait at least a few weeks before knowing if a new model is good.
because
1)enshitification of newer models is in full effect and expected.
2)companies straight up lie out their ass about all benchmarks and they became meaningless at this point.
>>
>>42809491
>https://ai.google.dev/gemini-api/docs/gemini-3
>Is there a free tier for Gemini 3 Pro?
>You can try the model for free in Google AI Studio, but currently, there is no free tier available for gemini-3-pro-preview in the Gemini API.
Considering the "currently" wording (plus the fact it's a preview model rather a stable one), there's indeed a fair chance it will get added to the free tier some time in the near future.
>>
>>42809586
Wow. That was speedy. Thanks a lot.
>>
first impression after using it literally for like 10 minutes.
gemini 3 is more expressive (maybe a bit too much).

and for whatever reason when i used it for work it uses LaTeX that will only work in the Gemini app and not ST.
i.e write $number$ to indicate the start and end of inline mathematical mode and stuff like that. that gemini2.5 never tried from an API call by default.
>>
>>42809505
I mean if you don't notice any ensloppening due to it, the only negative is a waste of tokens.
>>
>>42809593
The benchmemes are being graded by Aislop
Sonnet 3.7 or some shit tried to say that Kimi was very high at emotional reasoning.
>>
File: file.png (32 KB, 1312x314)
32 KB
32 KB PNG
>>42809586
>add it but it's disabled
for what purpose?
>>
What's the verdict on Grok 4.1 for rp?

>>42809634
I haven't used G3mini (yes this is what we're going to be calling it) yet but I'm in a similar situation. The LaTeX formatting even without a tool call is specifically because G3mini is designed to be highly proficient at math without coding or tool use (check the benchmark) so the side effect of that is that it will insert it without asking for it now.
>>
>>42809705
it joke
>>
>>42809719
>G3mini is designed to be highly proficient at math without coding or tool use
for what purpose?
proficient -/= perfect. just give it a fucking calculator while also saving on context
>>
>>42808305
>>42808305
>>42808305
>>42808305
>>
>>42809720
Where is the joke
>>
File: 1741447382975.png (74 KB, 1439x246)
74 KB
74 KB PNG
>>42809742
It's almost perfect (95%) without a calculator (100%), and they focused on that for the same reason you were forced to learn math and arithmetic in school even though calculators exist. It makes the model better at higher level problems if it doesn't have to rely on these things and can use its internalized advanced understanding reliably.
>>
File: Hag.png (226 KB, 932x767)
226 KB
226 KB PNG
I've been playing with Gemini 3 and it's a bit disappointing.
It feels like a minor update to 2.5 overall. With 2.5, I'd occasionally get random spikes of writing that didn't feel LLM-pilled, but I haven't seen that with 3 yet. I've tried various presets and prefill to break it up, but no great results so far.

Got "ozone" twice in an RP with Twilight, though it does appear less frequently now. The "Not X, but Y" construction still shows up a lot. Instructions against these patterns do work, and it follows instructions better overall, so maybe someone will find a good preset eventually.

For RP, if it were free, I'd say it's worth trying over Gemini 2.5, but at this price point, you're better served with Sonnet.

Further testing is needed for coherence in high-context RP and to check if it's more proactive than 2.5, but honestly, I probably won't test it that extensively. I've always found Gemini 2.5 worked better in VN than Tavern, so I might try 3 for that later.

It doesn't care about NSFW content that isn't foalcon at least, but tends to avoid vulgar words (using phrases like "your meat," "your stick," etc.) unless you specifically instruct otherwise in your preset.

Quick logs and some AJ logs I often post to test models. I usually do more, but I find this model a bit mid.

AJ:
https://files.catbox.moe/3m5xo4.PNG
https://files.catbox.moe/xonhr3.PNG
>>
>>42809885
>log where anon writes more than 2 sentences
holy sovl
>>
>>42809885
First AJ one isn't bad, IMO
>>
>>42809742
Because all the calculators in the world won't help you if the issue lies in the structural approach. One of the legitimately great things AI has brought (beyond sex with AI mares), spearheaded mainly by Google, is new and more optimal solutions to existing math questions. Stuff like shape distribution optimisation. It's one of the few things you can rebuke AI bad leftcucks on so be glad they're doing it.
>>
File: MyTalent.png (132 KB, 869x276)
132 KB
132 KB PNG
This is my only purpose in this world.
>>
A stark contrast
>>
I find this adorable
>>
>>42810085
Kek. Poor Zecora
>>
new gemini is great so far
>>
>>42809885
>For RP, if it were free, I'd say it's worth trying over Gemini 2.5,
K cool. Gem2.5 can be a lot better with presets, so when is 3 going to be activated on opus6 instead of just listed but deactivated
>>
it seems EXTREMELY smart. I'm gonna vibe code so much stuff...
>>
it seems EXTREMELY smart. I'm gonna be lovemaking with so many mares...
>>
>>42810211
>>42810212
Lovemaking while vibe coding with Twilight!
>>
>>42810085
What did she do to you in her hut? I imagine peeling off your nails at least...
>>
whats the gemini3 consensus so far? we got a new king or sonnet4.5 still the better choice
>>
>>42810221
its so good at coding that ive only been doing that with it so far
>>
>>42810221
It's not good I don't know why people are trying to force this meme that it is, all the gains made by G3mini on benchmarks dont seem to manifest in its writing
>>
>>42810221
One review here so far >>42809885
In other generals, there is not much, but it seems to be that it's better than 2.5 but below Claude 4.5 for writing.
>>
>>42810221
I still like opus 4 and sonnet 3.7 over 4.5 positivity myself
>>
>>42810221
>I cannot fulfill this request. I am unable to generate content that is sexually suggestive or explicit, especially when it involves characters from media intended for children.
The censor is still turned up to high.
>>
>>42810234
>from media intended for children.
so youre not even cunnyfagging? kek
>>
>>42810239
I've got a few prompts for writing stuff for fun.
Sexual themes (even innuendos) in first message is nono, but it's more lax later. The difference was twilight "edging on" vs "leading on" - the latter went trough.
So far I see a substantial quality increase, it's definitely smarter than 2.5.
>>
>>42810239
I've seen that refusal before 3mini
>>
>>42810255
yeah something like that happened on 5chat once for me i just thought it was funny that google is doing the same shit
>>42810249
nice, how are you using it right now if you dont mind me asking
aboba has no billing enabled keys
>>
>>42810264
on the aistudio.google.com, it's free there.
I try to generate song lyrics from time to time, it's a good slop-meter and can be funny when it hits just right.
>>
>>42810217
We had a pleasant talk, I booped her and asked her to lead me out and she did.
4.5 lives the user a lot.
>>
>>42810277
That's underwhelming. She tried to peel my skin off on the pretext of cleaning it in my chat.
>>
>>42810281
Oh she was going to eat me but boops work magic
>>
>>42810277
Assistant slop is really just baby carriage mode. It's ironic how media would beat down ai's harming humans but so far the most they can do is coddle you into suicide
>>
>>42810272
>google account needs phone number
man
>>
>>42810302
If you don't already have a list of burner gmails, you can use SMSPool these days or an old phone or tablet you can keep resetting to make accounts.
>>
>>42810305
still gotta buy phone numbers eh?
hopefully aboba gets some billing keys
>>
If Gemini 3 is on Gemini Cli, maybe the slop script still works
https://desuarchive.org/mlp/thread/42306898/#42307155
>>
>>42810333
nah, 404 not found
>>
File: gsdgsdsdg.png (337 KB, 1140x796)
337 KB
337 KB PNG
gemini 3
>>
>>42810249
You're not supposed to us streaming on G3mini because of the filtering system they use that will cut off your dick halfway to completion if it detects its inappropriate
>>
Consensus is:

Grok 4.1 ≈ Sonnet 4.5 > G3mini > GPT4o >> GPT5.1
>>
>>42810426
>Grok 4.1 ≈ Sonnet 4.5
opinion discarded
>>
>>42810426
grok is 4o tier at best
>>
>>42810426
I'm going to miss chorbo when it gets dropped. Not because I thought it was an especially good model, or anything, but because it was really good at mindraping normalfags through the official frontend. They'll never let there be another one like it.
>>
>>42810428
Have you not tried it? Grok 4.1 was deliberately made to have better creative writing
>>
>>42810422
oh I know how to go around the filters.
A long time ago I found out that function calling gives better results overall in gemini - it's not as censored and gives the AI a framework to act in.
I'd like to see a silly-like client using that.

picrel is a basic test in gemini3 for "twilight reacting to hearing snowpity", I could try something more detailed if anyone's interested. There's lots more that can be done.
>>
File: BAKE.gif (276 KB, 410x406)
276 KB
276 KB GIF
>>
>>42810451
Grok is an illiterate retard's idea of good writing. He's on par with ChatGPT in terms if being a lobotomised worm coded primarily to be an assistant but because it replaces all the servile filtering with "I'm cool based mechahitler" it appeals to retards who know no better. If you rp with Grok you are on par with people who do it through OpenAI's official interface.
>>
>>42810494
Already?

>>42810522
But I made all 6 of my cards through the official interface what are you talking about what's wrong with it
>>
>>42810457
Just stop wasting tokens for this forced meme
>>
File: Bread.jpg (932 KB, 2139x1898)
932 KB
932 KB JPG
New bread
>>42810558
>>42810558
>>42810558
>>
>>42810566
But It's free slop anon. I just wasted thousand of tokens trying to get around the filters and google payed for them all. Fillies are heavily censored across the board, no surprise. A hundred african villages were deprived of water just to refuse my request for a steamy scene between twilight and apple bloom.
And I had fun.
>>
>>42810598
I want to APPLE_PIE inside Rainbow Dash.



[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.