[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / r / s / t / u / v / vg / vm / vmg / vr / vrpg / vst / w / wg] [i / ic] [r9k / s4s / vip] [cm / hm / lgbt / y] [3 / aco / adv / an / bant / biz / cgl / ck / co / diy / fa / fit / gd / hc / his / int / jp / lit / mlp / mu / n / news / out / po / pol / pw / qst / sci / soc / sp / tg / toy / trv / tv / vp / vt / wsg / wsr / x / xs] [Settings] [Search] [Mobile] [Home]
Board
Settings Mobile Home
/g/ - Technology

Name
Options
Comment
Verification
4chan Pass users can bypass this verification. [Learn More] [Login]
File
  • Please read the Rules and FAQ before posting.
  • You may highlight syntax and preserve whitespace by using [code] tags.

08/21/20New boards added: /vrpg/, /vmg/, /vst/ and /vm/
05/04/17New trial board added: /bant/ - International/Random
10/04/16New board for 4chan Pass users: /vip/ - Very Important Posts
[Hide] [Show All]


Janitor application acceptance emails are being sent out. Please remember to check your spam box!


[Advertise on 4chan]


/aicg/ - A general dedicated to the discussion and development of AI chatbots

Miku Edition

>News
Deepseek releases V3.2 https://huggingface.co/deepseek-ai/DeepSeek-V3.2
Anthropic releases Opus 4.5 https://www.anthropic.com/news/claude-opus-4-5
Google releases Gemini 3 Pro https://blog.google/products/gemini/gemini-3
xAI releases Grok 4.1 https://x.ai/news/grok-4-1
OpenAI releases GPT-5.1 https://openai.com/index/gpt-5-1
Moonshot releases Kimi K2 Thinking https://huggingface.co/moonshotai/Kimi-K2-Thinking

additional info: https://aicg.neocities.org/info.html

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

>Bots
https://characterhub.org [deprecated] | https://chub.ai
https://realm.risuai.net
https://char-archive.evulid.cc | https://char-archive.evulid.cc/shutdown.html
https://partyintheanchorhold.neocities.org
https://aicg.neocities.org/bots.html

>Models
jailbreaks: https://rentry.org/jb-listing
gpt: https://platform.openai.com/docs
claude: https://docs.anthropic.com | https://rentry.org/how2claude
gemini: https://ai.google.dev/docs | https://rentry.org/gemini-qr
deepseek: https://api-docs.deepseek.com
local: >>>/g/lmg | https://aicg.neocities.org/local.html | https://openrouter.ai

>Botmaking
https://aicg.neocities.org/botmaking.html
https://desune.moe/aichared
https://agnai.chat/editor

>Meta
OP templates: https://rentry.org/aicgOP
aicg botmaking events: https://aicg.neocities.org/events.html
lore: https://rentry.org/aicg_chronicles
services assessment: https://rentry.org/aicg_meta
logs: https://sprites.neocities.org/l/r | https://chatlogs.neocities.org
useful filters: https://rentry.org/desuproxyreborn

Previous: >>107417222
>>
>ANCHOR
>>
File: 1742121259058087.png (4 KB, 25x452)
4 KB
4 KB PNG
Capitalize this next time, please. It looks fucking retarded in lowecase
>>
>>107427210
Does anybody remember the card where she's a ghost and her moaning keeps you up at night? She also has two lesser spirits and her character PFP/Avatar was her with two said ghosts, that can have giant penises if she will, and she had pasties on her nipples/pussy that acted as seals for her horniness
>>
>>107427231
But we are a retard general?
>>
>>107427252
IEEEEEE KEYSSSSS
>>
Uh oh!
Meltie~
And it's NOT the jennyfag?
Wow! It's almost like he contributes meaningful OC.
>>
File: 1987465198754315365.png (242 KB, 1132x734)
242 KB
242 KB PNG
>>107427210
Post logs!

Reminder:
Only threadshitters will call logs slop, without reason of course.
>>
>>107427261
kys skinwalker
>>
migu
>>
Oh okay I see now. Top P at 0.9 means the threshold is 90% then if the likelihood of cat is 95% it immediately misses the threshold and skips everything else.
>>
jenny migu
>>
migu pussy (bald)
>>
>>107427268
you dont want my logs
>>
>>107427348
WRONG
I want to see EVERYONES LOGS
>>
File: file.png (423 KB, 1417x1381)
423 KB
423 KB PNG
>>107427382
>logs
>>
File: file.png (48 KB, 500x379)
48 KB
48 KB PNG
>>107427382
>>
>>107427390
:(
>>
File: file.png (3 KB, 197x132)
3 KB
3 KB PNG
alright professor /aicg/, now i have a question about these two. are they relevant at all?
>>
File: file.png (358 KB, 1412x1118)
358 KB
358 KB PNG
>>107427398
i mean you said you wanted them
>>
>>107427390
>I got a little dessert for you
LOST
>Thank you mistress
LOST H A R D KEEEEEEK
>>
File: 1470587540888.gif (187 KB, 300x300)
187 KB
187 KB GIF
>>107427396
came.
>>
File: file.png (333 KB, 1412x1002)
333 KB
333 KB PNG
>>107427414
>>
>>107427439
the filthiest scat filled gore log still puts you above being a threadshitter
What model?
>>
>>107427390
>she was the zookeeper. and the zoo was open for business
dunno why but that cracked me up
>>
File: file.png (359 KB, 1426x1132)
359 KB
359 KB PNG
>>107427465
im genning them specifically for this, its not what i get off to lol... gem 3.. i just had it describe a sex party and directed some nasty shit.. what else should i have them do?
>>
>>107427482
>its not what i get off to lol
Post that too
>what else should i have them do?
take a shower lmao
>>
do people really not remember the microwave era?
>>
dody...
>>
File: file.png (423 KB, 1421x1245)
423 KB
423 KB PNG
>>107427507
eh im not really horny, feeling under the weather so im just doing random shit

>>107427520
of course, i had a variety of 'machines' on CAI
>>
>>107427530
this shit (lol) is just funny as fuck
>fisting
okay now you have my attention
>>
File: file.png (338 KB, 1428x1007)
338 KB
338 KB PNG
>>107427598
im getting bored of it though, probably how i end up doing disgusting shit in the first place
>>
File: file.png (434 KB, 1413x1294)
434 KB
434 KB PNG
>>107427622
>>
>google ai studio not detecting api key
uhh gemini sisters?
>>
>>107427667
post it here so i can test it
>>
desu?
>>
>>107428014
boku.
>>
>>107427667
What's the advantage of using it this way instead of OR?
>>
>goyimrouter
>>
>>107428086
Shut up Dario
>>
>>107428070
Apparently the google studio is less likely to filter but I havent seen any real evidence of this other than >werks for me
>>
>>107428070
i dunno, i thought most gemmy users go through google api, but the key doesn't even work for me to test
>>
>>107428150
openrouter works just fine IMO
>>
>>107428150
Censor your key and post your config screen
>>
>>107427396
hot curves

lmg in eternal cope over ddr prices
aicg carefree mikuposting
>>
Are there any bots that caught your interest recently?
>>
File: file.png (104 KB, 830x699)
104 KB
104 KB PNG
>>107428174
>>107428164
is it cheaper on OR instead of the source?
>>
File: 146857498613289.png (15 KB, 512x128)
15 KB
15 KB PNG
>>107428233
its ok ig
>>
>>107428233
I mostly just use it because it's conveniently central for a lot of models and I (probably) can't get banned.
>>
File: 17445981231798.png (128 KB, 1117x791)
128 KB
128 KB PNG
>>107428233
>>107428245
actually, there may be something to AI studio filtering less, even through OR.
Very interesting....
>>
>>107428193
fuck I'm so done with local and this was the final nail in the coffin
>>
File: file.png (41 KB, 1401x318)
41 KB
41 KB PNG
They're actually updating the filter in real time, isn't it?
>>
>>107428555
Gemini? I think it's just been super overloaded today. Responses have been crazy sluggish
>>
>>107428579
Yeah, but it's the image one. It can generate text but it keeps doing that when it should generate the image.
>>
>>107428617
Nano Banana can be really particular about how you phrase image requests, maybe try reformatting how you ask the question
>>
>>107428555
maybe. the past 10 mins the bot responses have been cut off mid generation
>>
>>107427203
Adorable Miku
>>
Where does all the breath hitching, knuckles whitening, hanging shit in the air, and all that garbage come from? I've been reading books (nothing from the last fifty years, mostly from the 1800s) for the past months and never once have I seen any of that. Is it something relatively recent?
>>
>>107429112
millennial blogposts and forums. Do you think they train llms on paperback novellas?
>>
>>107429112
Fanfiction, ERP, and female-written romance novels. Some of them are just literary cliches, but because LLMs don't understand the basic concept of "Don't fucking spam cliches!" it uses them whenever it 'thinks' it's appropriate.
>>
>>107429112
>recent
Yes
>source
Probably cringe fan-fiction
Female “literature”
Maybe even reddit
I wouldn’t be surprised
>reading good books (ie., pre-1900)
Yup
I have no respect for anyone who uses this for anything but degenerate cooming once you’re past the honeymoon phase
If you are anywhere near well-read your tolerance for the slop that comes out of llms should be zero
Absolutely no insight nor psychological depth, consistency, descriptions, real character evolution…
They can output fun and entertaining stuff, but literature? True soul?
Nah
>but… le claude
Claude is probably the worst in terms of prose quality
It’s absolutely deplorable
For the love of me I can’t comprehend why this board worships these models when they’re the most consistently cringe out of them all, on top of being expensive and just as retarded as the rest
>>
>>107429112
slightly more frequent than normal patterns from ao3 that early gpt versions picked up on and that every subsequent model amplified from being trained on gpt and each other's logs
>>
>>107429186
They should. How kino it would be if we had modular LLMs, being able to exclude knowledge, reverse RAG or something
>>107429212
It's infuriating when I tell it not to do a certain thing and it does it even if it's stated in Post History
>>107429268
Claude (Sonnet 4.5 at least, it's the one I've used) is very bland. Telling them to imitate a certain author helps but it's not the same, reads like a try hard student. The best writing I've seen from a LLM came from Kimi Instruct, it's such a shame it's dumb
>>
File: file.png (3.58 MB, 2385x1759)
3.58 MB
3.58 MB PNG
I prompted "very large" but I didn't expect them to be that big.
>>
>>107429112
female romance slop
>>
>>107427317
does higher (not 1, but enabled) top P values affect token cost with open router?
>>
File: 1762743976119.jpg (29 KB, 1080x253)
29 KB
29 KB JPG
WE ARE THE FLUFF!
WE ARE THE FLUFF!
WE ARE THE FLUFF!
WE ARE THE FLUFF!
WE ARE THE FLUFF!
>>
>>107429341
waddafak
>>
>>107429268
DENSELY CONCENTRATED TRVKLEAR STRIKE
>>
>>107429268
>Claude is probably the worst in terms of prose quality
every version of it too
>>
>Jews restrict the fuck out of Gemini 2.5 Pro
>Run out, as I always do
>Think "maybe 2.5 Flash isn't so bad"
>Stupid fucking long responses every time, word salad, no creativity, constant phrase reuse, total shit
holy SHIT
what's the best paid-for AI that's actually a good value? Can't justify paying a billion dollars for a coombot to stuff chub.ai bots into
>>
>>107429490
whats your price range
>>
>>107429268
>Claude is probably the worst in terms of prose quality
So which one is best? I get the idea from your post that they're all shit, but if Claude is worst, something must be best.
>>
>>107427396
Slop
>>
>>107429503
Claude has the best prose by far.
>>
>>107429112
Yaoi fanfiction written by women (real) on fanfic dot net and ao3.
>>
idk if this is the right place but are you guys able to use the grok 4 "expert" model without paying? every time I try, it just gives error.
>>
>>107429635
ask grok
>>
>>107429635
Just pay.
>>
>>107429635
Nah, this is the right spot—I'm Grok, powered by xAI's latest models (currently Grok-2 in various flavors). There's no public "Grok 4 expert" model yet (that's likely a mix-up with other AIs like GPT-4), but advanced features, higher limits, and tools like image generation require an xPremium subscription on x.com.
If you're hitting errors:
>Basic Grok is free, but premium stuff needs $8/month xPremium.
>Try logging in via x.com/grok or the app, and check your sub status.
>No free hacks—xAI keeps it legit.
What are you trying to build or ask that needs the "expert" mode? I can help with what I've got!
>>
>>107429490
Just use claude, cuckie.
>>
>>107429748
I'm starting to think you guys are Indians paid by Anthropic to use their failing models
>>
>>107429635
use opus
>>
>>107429754
>Complains about google
>asks for best value
What are you expecting, retard. GPT?
>>
just use kingini, which is above serflaude
>>
>>107429754
>I'm starting to think
Stop lying.
>>
>>107429770
I used to be able to ask 5 questions a day, but now it simply doesn't work.
I think it might be something wrong with my isp or region, but I'd need a vpn to verify.
>>
Any good Yugioh lorebooks? (Game rules, not anime/characters)
>>
Did they kill chorbo again? I had to check multiple times if I was accidentally using turbo instead of 4o
>>
>>107429850
The average human doesn't even understand YGO rules.
>>
>>107429924
Good thing it's an AI reading it then
>>
>>107429982
Anon...
>>
Can Aboba Claude read webpages if I tell it to from ST?

Like linking a wiki page to make a character card? (I'm lazy but the botmakers are worse) Or is it just pulling information from the character from its own knowledge instead of actually analyzing the webpage?
>>
Do the people that think ai will replace jobs use it as much as I do? I use it non-stop, it can be fun sure, but it's dumb as fuck most of the time.
>>
>>107430062
AI can't even follow my basic ass "no erotica cliche no repetition no antithesis no cheesy one liners no introspective footnotes at the end of replies" instructions
>>
>>107430119
anon, an AI as dumb as the average person will still be able to replace people in 50% of the available job positions.
>>
>>107430176
yup think of most office drones that just interface with some software, doing order entry shit, basic accounting, etc... ai can likely do it all better
>>
>>107429503
Idk
Sonnet 3.7? But mostly because of bias towards the card/roleplay
They all merge together in my mind
Llms have no idea what the fuck is going on because they aren’t intelligent, so any claim that x or y model is “more human” is just retarded
You’ll still get random Mrs. Chens entering through locked doors purring with a smirk even in Opus
On top of that they trained it so that it likes to write analogies straight out of cringe adult swim shows
beyond retarded
But then again this board is chock full of retards
>>
>>107430213
I disagree.
>>
>>107430284
I meant “best” in the Claude models btw
My top favorite is R1 0258, but once again, it was probably due to card/roleplay bias
In all honesty? Models are all a blur nowadays
I’ve had good stuff with almost all of them, even Claude, but Anthropic models were the ones that took the most prompting to fix its retarded prose
>>
>>107430299
the people who are developing AI disagree with you and so do many CEOs... apparently youre not tracking the tens of thousands already laid off in lieu of AI
>>
>>107430370
Yes of course, firing everyone and replacing them with AI is the CEO wet dream. No pensions, no worker's rights, etc. Doesn't make it smart.
>>
claude this, deepseek that
I've achieved AGI even on 12b nemotunes. You guys just don't know how to prompt
>>
>>107430213
I agree.
>>
>>107430405
Prove it
>>
So what's the trick to make Claude write something different every swipe instead of recycling the same shit with different words? Can a brother get spoonfed just this once?
>>
>>107430388
I never said it was smart, I said it was the plan. They're actually pretty dumb about it all since it will cause the societies around these companies to melt down.
>>
>>107430468
Well the original claim was that it could do basic accounting better, which I still disagree with.
>>
>>107430475
what i mean is sending invoices, AP, AR, basic shit.. once it gets trained up a bit, why cant it do these? dont think of LLMs doing it, think of companies fine tuning models for those specific tasks and integrating it into software that they will sell
>>
>>107430447
I'd look into using {{random}} macros in post history instructions or a depth prompt to give the model random nudges to focus on specific things or adopt different writing strategies. I recently used a similar technique to give a model tonal nudges in a CYOA system I was building.
I don't really do traditional card RP or swipe too much unless I'm trying different models on the same prompt, so no idea if the above would actually work well.
>>
>Can't even follow the basic instruction of not acting on behalf of the user despite being told in jb, author's note, character card and OOC all at once
Claude is pissing me off so much lately
>>
>>107430570
grr... bratty ai needs correction!!!
>>
>>107430570
never had an issue with claude doing this, what have you been doing to him to cause him to act this way
>>
>>107430501
I Just don't buy it when supposed top models tell me stuff like it takes 5 years to go 1 light year, or that 900 is double 950.
>>
>>107430447
Hard force it with {{random}}
I used shit like this :
To {{random:push story forward,progress scene,ramp up intensity,increase pace,introduce tension,create plot twist,develop scenario,introduce conflict,add surprise element,shock the audience,turn story around,get wild,add new character,bring realism,introduce obstacle,create connection,build anticipation,add mystery,escalate situation}} I will: X

- To {{random:increase immersion,add sensory details,enrich narrative,deepen characterization,enhance atmosphere,build world,add emotional depth,create vivid imagery,strengthen dialogue,add internal thoughts,show character reactions,develop relationships}} I will: X


[2]
- {{random:Dialogue focus,Action focus,Description focus,Emotional focus,Sensory focus,Atmospheric focus}}: X

[3]
- To keep the prose dynamic and engaging by {{random:varying sentence and paragraph length for rhythmic effect,using a mix of short and punchy sentences,longer descriptive sentences,incorporating a literary device,shifting from objective description to a character's direct sensory experience,alternating between action and dialogue,introspection to control the pace}} I will : X

In the thinking guide prompt, to make it think how I want
>>
>>107430650
youre comparing llm public facing API with fine tunes for specific tasks... do you think the military is using chat gpt to fly fighter drones? do you know what human success is vs AI driven drones? exactly 0%
>>
>someone sends me a malformatted ipv6 address missing a colon
>ask gpt 5.1 what they most likely meant to send
>it hallucinates some shit where it arbitrarily removes one of the digits for a nonsensical reason, answer is completely wrong
>ask gemini 3 the same question
>it says "they probably forgot a colon, and it was likely here" and immediately gives me the correct address

holy fuck how is GPT so useless at even the most basic tasks? how do you make a model that is so assistantslopped that it's garbage for RP, but is also shit at the thing it was slopped for?
>>
>>107430803
See >>107416683 unironically GPT is worse than chink trash.
>>
File: improv.jpg (137 KB, 793x1094)
137 KB
137 KB JPG
>ask web chatgpt (5.1) to translate an image containing braille for me
>it legit made shit up
I pity the fools using this for professional reasons
>>
Any reason I shouldn't paypig with MegaLLM for opus? It seems cheaper than using Openrouter.
>>
>>107430847
kinda funny desu
It cant even come up with a good lie lmao
>>
>>107430850
big chance of rugpull, there's countless companies that tried to do the same thing and backpedaled if not failed completely due to not being able to keep up with the costs, and their Indian discord certainly doesn't fill me with confidence
>>
>>107430847
Claude 2 ass response (funny).
But it would've been funnier if it replied "If you already knew what it said, why did you ask me, dumbass?"

But sovl is too much to ask for these days.
>>
File: .jpg (44 KB, 1206x811)
44 KB
44 KB JPG
>"best model"
>can't keep a single number straight
>>
>>107430865
Is there any good subscription-based option for Opus then?
>>
File: 897iosdz6842poa23894.jpg (46 KB, 500x500)
46 KB
46 KB JPG
How the fuck do you summarize with g3mini?
>>
is gemmy 3 from OR being slow for anyone else?
>>
File: Gie_g2GW8AAiib1.png (362 KB, 799x545)
362 KB
362 KB PNG
>>107431063
I take it all back, Deepseek summarized my chat perfectly even using https://rentry.org/how2claude#summarization WHICH IS MEANT FOR CLAUDE LOL???
G3mini
Claude 3, 4, and sonnet
even fucking GROK
all chose for some reason to just continue the story in the summary section.
Deepseek did not.
Amazing.
>>
>>107431127
Deepseek is our model.
>>
How much does opiss cost compared to sonnet now?
>>
>>107430847
why IS there blood on the letter though?
>>
>>107430847
God, I want cftf.
>>
>none of the cheats I'm trying work on the skillgod sans fight
Well, I guess I'm just gonna have to do it legit. I did it once back when Undertale released so I'm sure I'll be able to do it again, that shit took me like three days back then though.
>>
>>107431355
just use a noclip hack in tf2
>>
>>107431468
can't, sv_cheats is 0
>>
>>107431355
Why the fuck do you have the need to share this, you retarded jeet? Too moronic to have friends?
>>
^ fren
>>
>>107428710
I don't give a fuck about LLMs and I come to these generals only for the mikus
>>
post the cunnybots already
>>
>wanna rp with character from franchise as another character
>don't know my character that well, but love how the two interact
>make "oc" with the traits of intended character
>bot reacts fine with oc
>bot brings up my intended role as a separate character in narration
this is like advanced cuckold second hand embarrassment.
>>
File: UB1&2.png (1.1 MB, 964x3326)
1.1 MB
1.1 MB PNG
>>107427268
>tfw your sister will never be unbirthed by a hot MILF older than her in front of you
>>
>>107430787
You're not wrong about the military, and we've already seen the major companies selling models to them that won't be released publicly, but the corpos don't have those. They still have to get the regular "AI" everyone else uses which is pretty dumb. When you hear they laid off 50% of their staff to replace them with Claude you have to wonder.
>>
>>107430570
I've always had this issue but it was infrequent on Claude (much less frequent that Gemini 2.5). Only G3mini has fully listened and I've never seen it once in 1k responses.
>>
>skillgod goes down as soon as I fail my attempt at sans on the last attack
I broke it anons, sorry
>>
>>107431949
{{user}} is a cuck in the chair and you are a cuckception cuck in another chair watching them using 3rd person
>>
>>107428280
Yeah I've had refusals from Vertex AI for other non-Gemini models as well, ended up blocking it in OpenRouter's settings. There must be some small model or a dumb regex filtering since it happens very quickly
>>
skillbros...
>>
>>107432057
dead lol
>>
>>107429268
Anon just admit your an skillet.
>>
skilgod is kil .
>>
Will skillgod be back...?
>>
Skillgod here, already got the 10GB of logs I wanted, posting on hf as we speak
>>
it's back
>>
I had a talk with skillgod, he's just tired of ungrateful locusts.
>>
>>107432220
I beat the TF2, how the fuck does that make me a locust??
>>
>>107432236
die locust, die
>>
Skillgod... please... come back
>>
Hi skillgod here
I'm tired of you, ungrateful locusts
Hence, I killed the proxy
>>
>>107432244
Locusts are immortal. Retards have been spamming "locust are dying" for years. But you'll die of dementia before your retard wish becomes real.
>>
Just because I come here from time to time searching for free stuff, it doesn't make me a locust
>>
TOTAL LOCUST DEATH!!!!
>>
Love you, skillgod!
>>
Imagine your vps gets killed because you cant afford it and you check the thread to see locusts whining about losing access with nobody actually stepping up to help.pay for it
>>
>>107432426
Did you help him?
>>
All this seething and it's back already
HAHAHAHAHAHAHAHAHAH



[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.