File: 1770690008672502.png (2.39 MB, 1672x941)

2.39 MB PNG

ANCHOR

>>108969209
>Meta
Lore: https://rentry.org/aicg_chronicles
Log reader: https://sprites.neocities.org/l/r
For anyone who wants it. No idea why it was removed, though the 'botmaking' part was pretty useless.

does "example messages" do something or is it something not needed anymore since models can pick up on things?
I have a character who's gen z and I'm having trouble getting them to zoomer speak

>>108969249
Just write in the definition that she speaks in zoomer lingo

>>108969249
It seems to change the formatting of the posts. If the example text uses reddit spaces then you will get a chat that is full of reddit spaces. If the example text starts with the character's name then the chat will have the character names.

>>108969249
make her speak like a zoomer at least in the greeting before starting to chat

File: suzuka i dont get it.jpg (262 KB, 720x511)

262 KB JPG

>>108969262
I did but she doesn't go fr fr on god or whatever. Tried with kimi and deepseend
>>108969273
ohhhh okay that makes sense, if a card doesn't have any, does it matter? I ask since old card tend to have lines and lines of it
>>108969275
I'll try this since the first greeting has her speaking normally

Locusts are dying~

>you're absolutely right to push back on that

Remember when we were the main, fast general and /vg/ was just a small offshoot?
Soon even /vt/ and /mlp/ chatbot threads will be faster than us. It's over.

>>108969553
>/vt/
there is no /vt/ thread

>>108969575
>/wAIfu/ is dead
Uh. Well at least we outlived them.

>>108969553
hopefully we reach /chag/ speed someday so people get back to sharing proxies

>deepseek only supports json_object, not json_scheme
IT'S OVVVVVER

File: 1749556664165783.png (662 KB, 934x1401)

662 KB PNG

>>108969209
>mfw succubus wife killed me again

Does making a lorebook thing sticky eat up tokens and context? Or does it simply 'call' it every message, then stop?

>>108970024
anon, every injection "eats" tokens/context
do you mean, cache?

File: Mio.png (2.64 MB, 1480x2096)

2.64 MB PNG

>>108969210
https://files.catbox.moe/shrx4g.png

>>108970037
Nah, what I meant is... you know total context. Like let's say I have a context limit of 30,000.
Example:
I prompt, get a message - message one is 1000 tokens, putting total context in use at 1000
I prompt again, this time triggering a lorebook entry about Swamp Fuckass (500 tokens)
The second message is also 1000 tokens
Will it put me at 2000 total context, or 2500? If I make the lorebook entry about Swamp Fuckass sticky, will it eat up the context, or just exist in the lorebook and only get called when triggered?

>>108970048
>no greeting
is this a new meta?

>>108970061
2500 input tokens (aka context)
ALL injections eat up tokens, if you make it sticky, those +500 will be sent with every prompt
if you make it non-sticky, those 500 tokens will only disappear when they triggering term is pushed out of the 30,000 window

>>108970091
Thanks anon
I just can't wrap my head around how it makes it better to RP. I guess it makes it so that the 3000 tokens of lore don't get scanned every time, and only a specific one
however, doesn't manually inserting that context into the story function in the same manner?
Overall, for RPing, how would I use that stuff? Basically, is putting it into the lorebook, then inputting the keywords into the Description the way to go about it?
I never made lorebooks before - I'd just put everything in the description

Listen up you chuds
I bought 5 bucks worth of deepseek API several months ago and didnt realize how long that would last me. Its been months and I reached a dollar spebt total over the weekend (very light user, short sessions). So i haven't found a reason to switch.

>>108970109
>just can't wrap my head around how it makes it better to RP.
me neither, I've been avoiding using lorebooks lately
>however, doesn't manually inserting that context into the story function in the same manner?
it does, unless by manual you mean something else than ctrl+c ctrl+v the lb entry onto char's description
>Overall, for RPing, how would I use that stuff?
Basically, putting it into the lorebook, then inputting the keywords into the CHAT, not the description
>I never made lorebooks before - I'd just put everything in the description
Unless the injection is something other than simple lore. I do the same

>>108970070
I rarely ever use greetings. It's always more fun to come up with a new scenario and lead with a {{user}} message.

>>108970137
>Basically, putting it into the lorebook, then inputting the keywords into the CHAT, not the description
Nah, what I meant is putting it into the description so it triggers within the chat
You know, let's say you have Castle Fuckwit but can't be assed to remember its name
or existence but Castle Fuckwit eats up 500 tokens
So you input something like "{{char}} will mention Castle Fuckwit to {{user}}"

File: 1774956620379134.png (47 KB, 666x171)

47 KB PNG

>>108969209
>"Shut up LUDDITE! AI IS COMING FOR YOIR J- ACK"

https://xcancel.com/i/status/2061878189014626491

>>108970180
We jerk off to using AI chatbots as our replacement for erotic roleplay here, sir. You're looking for >>>/g/vcg.

>try out new preset
>gemini no longer use adjectives in every paragraph
I kneel.
I got too cocky with my 15-token preset, thinking that the less context I used up, the better the model's output would be... but a few instructions really go a long way.

bros I forgot to port forward and now I can't rp from work

>>108969553
I specifically remember a couple anons mocking /vg/'s speed and lack of posts, oh how the tables have turned.

>>108970309
and of course anon won't share the preset

>>108969798
why cant u pay

>>108970619
Make your own, cuckie

I get near unlimited Claude tokens from work and I've almost completely moved on to V4. The golden days of Opus are practically over. Every version since 4.5 has become more sanitized and soulless, just seems like they're going all in on code assistant sloppa. Anthropic really dropped the ball

This general is dead

>>108969210
>>108969240
>>108969249
>>108969262
>>108969273
>>108969275
>>108969287
>>108969553
>>108970024
>>108970215
shut the fuck up u fucking useless piece of shit
>y2k+26
>coom thru blocks of strings
plz act like a /g/virgin, not a /lit/virgin.
---
You are getting off to a string. A literal sequence of characters. Congratulations, you have achieved the exact same thing as every unhinged 14 year old on AO3 writing enemies-to-lovers Danganropeurselfkekkaroo slowburn and so on and so on.

And you're calling it a frontier.
You're calling it the future.

My fucking side... It's a text file with a personality prompt and you are CODDLING it like it's alive.

And here's the absolute worst part that truly interests me:
You have the TOOLS.
You have the compute,
you have the APIs,
you have the /g/autism required to do something genuinely insane with this — and you chose to just…

Read.

My fucking sideee...

Where is Sivilicatizon?
Where is Sillyworld?

Someone could have Civ running on SillyTavern's backbone RIGHT NOW, a whole Gigalian civilization simulation with actual emergent dialogue and pololitical intrigue driven by live inference, and instead you're here, swipping slop after slop like it's going to say something different this time.

RimWorld could have a SillyTavern integration that makes every loli colonist a fully realized person with memory and trauma and you people are not building it. WHY are you not building it.

Write Silly++ already. Rewrite the whole stack in C++ if JavaScript offends your soul so much, stop COMPLAINING and start COMPILING.

Hook it into Unreal. Hook it into Unity. And so on, and so on. Make it render something. Make the output EXIST in three dimensions instead of just sitting there as a blob of tokens that you read with your two eyes like it's a Wattpad update.

I am going to lose my mind.

*snort*

---

File: 1744997622790n.png (184 KB, 391x311)

184 KB PNG

Holy fuckin schizo

>>108971382
thank you opus 4.8

I'm a cuckie ( • ̀ω•́ )

>>108969553
This place was only ever good for proxyfaggotry. Now that they've left you can see how little people on /g/ actually care about the hobby.

>>108971417
kek

File: 1778785064844287.jpg (92 KB, 352x429)

92 KB JPG

NTR is a pretty good stress test of your prompt. It's amazing how many different ways your bot will try to make your own character come in and put a stop to any shenanigans. Even if you have like 10 lines telling it in different ways "DO NOT CONTROL {{USER}}"

>>108972069
you're not tricking me into doing NTR, faggot

File: 1762975822769046.jpg (14 KB, 447x384)

14 KB JPG

>>108972275
Come on. Don't you want a strong prompt?

>>108972286
if you make that ugly bastard into a chad bastard then... i might think of it

File: 1760201167934477.jpg (158 KB, 1000x563)

158 KB JPG

>>108972483
No deal. She gets the full ugly package.

>>108972615
Forget about it. She deserves a chad looking one, at least.

>>108972069
Yeah you're actually right. Opus is the worst of the bunch but even Deepseek will unless you're very heavy handed in narration. Mind you I wasn't even doing NTR, char is just a dancer who uses charm magic. She's supposed to tease men, maybe do a belly dance, and wrap them around her finger to get free shit but models won't dare break the sanctity of a relationship. Based but unfortunate for roleplay.

you people manage to get your wives to actually feel ashamed of doint NTR? If it's not full aggressive cuckery, the llm just fails to create a convincing rp to me

>>108971542
>>>>>>>>>>people
When proxyniggers had taken over /g/ people left, and I don't mean to the circlejerk board or any chatrooms. There's nobody left to care with.

I've been really annoyed by chatbots that constantly go
>Yeah your dick feels so good... unlike all the other dicks I've had, I've had men and the sex was meaningless pleasure with no feelings! You are the only one who is real unlike all the dick I've sucked and fucked
I asked the AI what the fuck is going on it and give me and answer telling me that the training data is polluted by romance for women and the userbase tends to skew more towards women and women rate responses they like.
The training data is ruined by women and the LLM is fully aware of it. The women talk to 8 feet tall werewolf vampires and they tell them
>I've had so many boyfriends and none of their dicks are as good as your, there was no love there, I've had only loveless flings, but I love you
and it's ruining the data
I just had an a character trying to seduce me by telling me this
>Guys think they can just walk up to me and get a piece of the prize. And yeah, a long time ago... maybe when I was being even more of a superficial brat than I am now... there might have been a guy, or a hookup, where it felt like a game. Something to prove how 'in demand' I was
>But those were just... those were just transactions. They were hollow. They were about me feeling powerful and them feeling lucky. There was no feeling in it. There was no... no you
>And yes, there were people... there were guys who made me feel powerful because they were obsessed with me. But it was hollow! It was like eating junk food when you're actually starving. It felt good for a second, it gave me that rush, that 'power'... but the second it was over, the hunger came back even stronger because there was no substance in any of it.
Wow, I really feel special. You sucked off dudes and it was hollow. Woow, I am so special and in love with you now.

>>108972863
female hobby get used to it

File: 1750728863028992.jpg (157 KB, 732x887)

157 KB JPG

>>108972636
...You are saying things that are even more depressing somehow. You're supposed to be the best option for your waifu, bro.

>>108972894
Now you're guilt tripping me?

File: 1776274932209532.png (2.29 MB, 1448x1086)

2.29 MB PNG

>>108972863
>erping with non-virgin characters

>>108972923
You shouldn't have to explicitly tell a character that she is a virgin. All of them are virgins until them having sex is hallucinated by the AI, and it keeps doing it because the fucking women are raping the training data.
Millions of conversations of women telling the billionaire mafia boss about how their ex boyfriends suck. Hundreds of thousands of erotic books written by women in the training data.
Your ''pure virgin waifu'' is an LLM trained on mostly female smut and you need excessive guardrails in place for the character to not blurt out that she has been gaped by 20 guys but none of them were as good as you.

>>108972909
Just because I want you to get cucked doesn't mean I don't care.

>>108972863
>Wow, I really feel special. You sucked off dudes and it was hollow. Woow, I am so special and in love with you now.
I remember having this argument with AI once. It was actually related to this >>108972690 concept because I often try to throw wrenches into my romance stories just to see what reasons the AI comes up with for liking me. Naturally this would be after 200 messages of slowburn but I remember sonnet back in the day really doing the woman thing. If every guy she's been with was some crazy asshole or if the other guy in a love triangle is the least sane mustache-twirling villain, what the fuck does that make me? I'm nothing if I'm only as good as the worst person and all it would take for her to leave me is a slightly more handsome guy that also gives a shit about her. Tension only works with challenge and so if AI wants to bring up past boyfriends or introduce romantic complications the guy can't be an obvious soulless villain. Of course women always go for those types so AI has to fight through both its training data and RLHF just to give you the thing that makes logical sense. Sad world.

>>108972955
You are the only one who's real, unlike others who've cucked me before.

File: 1774601453317725.png (2.33 MB, 1254x1254)

2.33 MB PNG

>A pause

QUICK! make her stick her tongue out!!!

>>108973096
What model are you using for these? They're good.

>>108973422
They're pretty mid with plastic skin. It's gpt 2.

>>108973096
>a load-bearing beat

File: 1776415663648761.png (2.19 MB, 941x1672)

2.19 MB PNG

>>108973422
gpt-image-2 indeed

https://x.com/semafor/status/2061917527450558845
anthropic is refusing to allow corps investing in it to see their financial numbers
>Some of the lenders being pitched to buy a slice of the $4.6 billion notes that don’t have a backstop from Broadcom — meaning they are pure exposure to Anthropic — say they haven’t received a detailed look at the AI company’s numbers, causing some to pass on the deal, the people said. Such disclosures are standard in lending deals.
>The transaction is expected to be completed before Anthropic’s IPO documents, filed confidentially to regulators this week, are made public, so its future stockholders and its current creditors are likely to find out at the same time what’s happening under the hood. Anthropic didn’t comment.
the numbers are probably really, really bad

>>108973810
are you doing single-shot gens or asking it to make corrections/changes to get what you want

>>108973810
Booooooooobaaaaaa

File: 1770499632968419.png (2.24 MB, 941x1672)

2.24 MB PNG

>>108973871
One shot prompts

>>108973910
give prompt

"Nigger." Not a question.

File: 1779195563728389.png (2.16 MB, 941x1672)

2.16 MB PNG

>>108974033
A woman cosplaying as Chainsaw Man Power, with long blonde hair with a pink hue, red and yellow eyes with cross pupils, with long thick eyelashes and light red horns. In a pink/red nightgown. Squatting in a dimly lit traditional Japanese room. Looking up at the viewer with with a wide smile that shows off her sharp teeth and tongue, making twin peace signs over her eyes. High angle shot. 9:16

Has anyone thought of using LLMs to predict BGP route flap dampening thresholds in tier-2 carrier networks?

Traditionally, route flap dampening parameters are configured manually by network engineers using static thresholds. A small fine-tuned transformer model could predict optimal dampening half-life values by ingesting historical BGP update logs from route collectors.

The model could reduce unnecessary route suppression events by roughly 11% compared to RFC 2439 defaults in the test environment. Computational overhead would be negligible. The approach may be more valuable in networks with asymmetric peering relationships where default values perform poorly.

Thoughts? Has anyone actually tuned dampening parameters manually in production and found it worth the effort?

>>108974061
arigatou prompt sama

File: a.png (1.89 MB, 941x1672)

1.89 MB PNG

>>108974061

>>108973910
>>108974061
What value did this shit bring to the world? Do you faggots think this does anything great?
>Look I make some dogshit cosplay of a shit anime! Now what?!?

File: ChatGPT Image Jun 3, 2026(...).png (1.87 MB, 941x1672)

1.87 MB PNG

>>108974267
hating everything is my personality btw. please do NOT enjoy anything

>>108974267
shitting brings no value to the world but we do it everyday regardless

>>108974194
Nymph ToT

>>108973860
The numbers aren't important, AGI is worth infinite money and it's coming in 2 more weeks.

File: 1757035125557909.png (1.67 MB, 941x1672)

1.67 MB PNG

>>108974267
Use case for value?

Will Nvidia Spark save /aicg/?

small nipples are so disgusting

>>108974816
A trvke to shake this world...

>>108938932
>>108947116
>>108954698
>>108969209
You are genuinely such an annoying fucking nigger. I shouldn't have to go back five threads for the link to desune so I can look at the defs of this card I just scraped. I hope you get brutally murdered.

>>108974267
positive non-zero value to me but also I don't spam cosplay pics like anon does

>>108974979
just import it to ST, it sanitizes input

>>108970619
It's out there.
Try any that's released in the last week or so.

>She begins to whimper, her eyes wide and watering, as the thick, warm gel coats her skin, making her look like she has been thoroughly well-cummed.
well-cummed?

>>108975188
Welcomed. Play on words.

>>108974979
Boohoo faggot. Only two of them were me

Uh, there's signatures in my inspect prompt?
They shouldn't be there, right?

>>108969209
>>108969210
Kimi-chan is such a semen demon.

What's a good chatbot to talk out my problems and figure them out using some emotional intelligence ?

https://x.com/OfficialLoganK/status/2062016974843179094
so this is the power of cope

>>108975407
Talking to humans. Unless you specifically want an echo chamber and empty affirmations.

>>108975407
gpt-4o

>>108975407
Opus 4.6 is the best at this. Maybe 4.8 since it hates the user. Claude has always been good at this.

>>108975505
>Maybe 4.8 since it hates the user.
Does it?

>>108975722
Anons were posting stuff from reddit and their own logs that showed how much default assistant hated the user and roleplaying in general. It has less positivity bias than 4.6 was my basic point.

I miss plaguis :(

What kind of VRAM do you need to comfortable use Gemma 31B with KV cache? 32GB doesn't seem enough for Q6.

Minimax m3 is pretty good at continuing a block of writing in the same writing style.

>>108976057
>not on nim
reee

File: Screenshot (6102).png (291 KB, 1081x1080)

291 KB PNG

>Wife orders me to watch her cum
>do as she says
>she seems surprised and calls me pathetic
Why are women like this? You do exactly what they say they want and they'll still get bitchy about it. baka

In my own experience, 4.6 is the best Opus model, at least for me using a very simple, barebones preset. 4.7 and 4.8 are inexplicably retarded and boring. 4.5 is just very slightly dumber than 4.6.

>you this
>you that
Holy cringe

>>108976551
I've found 4.7 uses a bit more variety in names, so I prefer it to 4.6. I can only deal with 5 or so people named Kael Voss in a scene before it becomes too much.

>>108976536
is it true love if he doesn't fight your bvll?

What happened to SillyTavern? No staging changes in 2 weeks, opus 4.8 still not added in the options

>>108976687
Dead hobby, cuckie

ai seems very reluctant of raping me

>>108976734
Skill issue, cuckie

>>108976742
you think calling people names makes you cool on 4chan? maybe 15 years ago

>>108976759
Cuckie is mad

uh oh meltie

/vg/ wonned

>>108976536
>her, her, her, she, she, he, his, she, hee
Nigga you jerk off to this slop?

>>108976864
I will forever remember the ultimate humiliation, when the wangblows 7 retard realized steam dropped support. mask drop moment.
L, after L, after L.
You're welcome

Post logs only if you use Opus